fix lint error

fix tests
api: add omitempty to required tool function parameter type
2026-02-12 00:23:16 -05:00 · 2025-11-06 13:55:25 -08:00 · 2025-11-06 13:49:17 -08:00 · 2025-11-06 12:58:15 -08:00
954 changed files with 51577 additions and 161521 deletions
--- a/.gitattributes
+++ b/.gitattributes
@@ -15,12 +15,8 @@ ml/backend/**/*.cu linguist-vendored
 ml/backend/**/*.cuh linguist-vendored
 ml/backend/**/*.m linguist-vendored
 ml/backend/**/*.metal linguist-vendored
-ml/backend/**/*.comp linguist-vendored
-ml/backend/**/*.glsl linguist-vendored
 ml/backend/**/CMakeLists.txt linguist-vendored

-app/webview linguist-vendored
-
 llama/build-info.cpp linguist-generated
 ml/backend/ggml/ggml/src/ggml-metal/ggml-metal-embed.s linguist-generated

--- a/.github/ISSUE_TEMPLATE/10_bug_report.yml
+++ b/.github/ISSUE_TEMPLATE/10_bug_report.yml
@@ -13,7 +13,7 @@ body:
    id: logs
    attributes:
      label: Relevant log output
-      description: Please copy and paste any relevant log output. See [Troubleshooting Guide](https://github.com/ollama/ollama/blob/main/docs/troubleshooting.mdx#how-to-troubleshoot-issues) for details.
+      description: Please copy and paste any relevant log output. See [Troubleshooting Guide](https://github.com/ollama/ollama/blob/main/docs/troubleshooting.md#how-to-troubleshoot-issues) for details.
      render: shell
    validations:
      required: false
--- a/.github/workflows/release.yaml
+++ b/.github/workflows/release.yaml
@@ -16,15 +16,13 @@ jobs:
    outputs:
      GOFLAGS: ${{ steps.goflags.outputs.GOFLAGS }}
      VERSION: ${{ steps.goflags.outputs.VERSION }}
-      vendorsha: ${{ steps.changes.outputs.vendorsha }}
    steps:
      - uses: actions/checkout@v4
      - name: Set environment
        id: goflags
        run: |
-          echo GOFLAGS="'-ldflags=-w -s \"-X=github.com/ollama/ollama/version.Version=${GITHUB_REF_NAME#v}\" \"-X=github.com/ollama/ollama/server.mode=release\"'" | tee -a $GITHUB_OUTPUT
-          echo VERSION="${GITHUB_REF_NAME#v}" | tee -a $GITHUB_OUTPUT
-          echo vendorsha=$(make -f Makefile.sync print-base) | tee -a $GITHUB_OUTPUT
+          echo GOFLAGS="'-ldflags=-w -s \"-X=github.com/ollama/ollama/version.Version=${GITHUB_REF_NAME#v}\" \"-X=github.com/ollama/ollama/server.mode=release\"'" >>$GITHUB_OUTPUT
+          echo VERSION="${GITHUB_REF_NAME#v}" >>$GITHUB_OUTPUT

  darwin-build:
    runs-on: macos-14-xlarge
@@ -55,9 +53,6 @@ jobs:
      - uses: actions/setup-go@v5
        with:
          go-version-file: go.mod
-          cache-dependency-path: |
-            go.sum
-            Makefile.sync
      - run: |
          ./scripts/build_darwin.sh
      - name: Log build results
@@ -68,7 +63,6 @@ jobs:
          name: bundles-darwin
          path: |
            dist/*.tgz
-            dist/*.tar.zst
            dist/*.zip
            dist/*.dmg

@@ -110,13 +104,6 @@ jobs:
            install: https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-24.Q4-WinSvr2022-For-HIP.exe
            rocm-version: '6.2'
            flags: '-DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_C_FLAGS="-parallel-jobs=4 -Wno-ignored-attributes -Wno-deprecated-pragma" -DCMAKE_CXX_FLAGS="-parallel-jobs=4 -Wno-ignored-attributes -Wno-deprecated-pragma"'
-            runner_dir: 'rocm'
-          - os: windows
-            arch: amd64
-            preset: Vulkan
-            install: https://sdk.lunarg.com/sdk/download/1.4.321.1/windows/vulkansdk-windows-X64-1.4.321.1.exe
-            flags: ''
-            runner_dir: 'vulkan'
    runs-on: ${{ matrix.arch == 'arm64' && format('{0}-{1}', matrix.os, matrix.arch) || matrix.os }}
    environment: release
    env:
@@ -126,14 +113,13 @@ jobs:
        run: |
          choco install -y --no-progress ccache ninja
          ccache -o cache_dir=${{ github.workspace }}\.ccache
-      - if: startsWith(matrix.preset, 'CUDA ') || startsWith(matrix.preset, 'ROCm ') || startsWith(matrix.preset, 'Vulkan')
+      - if: startsWith(matrix.preset, 'CUDA ') || startsWith(matrix.preset, 'ROCm ')
        id: cache-install
        uses: actions/cache/restore@v4
        with:
          path: |
            C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA
            C:\Program Files\AMD\ROCm
-            C:\VulkanSDK
          key: ${{ matrix.install }}
      - if: startsWith(matrix.preset, 'CUDA ')
        name: Install CUDA ${{ matrix.cuda-version }}
@@ -163,18 +149,6 @@ jobs:
          echo "HIPCXX=$hipPath\bin\clang++.exe" | Out-File -FilePath $env:GITHUB_ENV -Append
          echo "HIP_PLATFORM=amd" | Out-File -FilePath $env:GITHUB_ENV -Append
          echo "CMAKE_PREFIX_PATH=$hipPath" | Out-File -FilePath $env:GITHUB_ENV -Append
-      - if: matrix.preset == 'Vulkan'
-        name: Install Vulkan ${{ matrix.rocm-version }}
-        run: |
-          $ErrorActionPreference = "Stop"
-          if ("${{ steps.cache-install.outputs.cache-hit }}" -ne 'true') {
-            Invoke-WebRequest -Uri "${{ matrix.install }}" -OutFile "install.exe"
-            Start-Process -FilePath .\install.exe -ArgumentList "-c","--am","--al","in" -NoNewWindow -Wait
-          }
-          
-          $vulkanPath = (Resolve-Path "C:\VulkanSDK\*").path
-          echo "$vulkanPath\bin" | Out-File -FilePath $env:GITHUB_PATH -Encoding utf8 -Append
-          echo "VULKAN_SDK=$vulkanPath" >> $env:GITHUB_ENV
      - if: matrix.preset == 'CPU'
        run: |
          echo "CC=clang.exe" | Out-File -FilePath $env:GITHUB_ENV -Append
@@ -185,20 +159,19 @@ jobs:
          path: |
            C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA
            C:\Program Files\AMD\ROCm
-            C:\VulkanSDK
          key: ${{ matrix.install }}
      - uses: actions/checkout@v4
      - uses: actions/cache@v4
        with:
          path: ${{ github.workspace }}\.ccache
-          key: ccache-${{ matrix.os }}-${{ matrix.arch }}-${{ matrix.preset }}-${{ needs.setup-environment.outputs.vendorsha }}
+          key: ccache-${{ matrix.os }}-${{ matrix.arch }}-${{ matrix.preset }}
      - name: Build target "${{ matrix.preset }}"
        run: |
          Import-Module 'C:\Program Files\Microsoft Visual Studio\2022\Enterprise\Common7\Tools\Microsoft.VisualStudio.DevShell.dll'
          Enter-VsDevShell -VsInstallPath 'C:\Program Files\Microsoft Visual Studio\2022\Enterprise' -SkipAutomaticLocation  -DevCmdArguments '-arch=x64 -no_logo'
          cmake --preset "${{ matrix.preset }}" ${{ matrix.flags }} --install-prefix "$((pwd).Path)\dist\${{ matrix.os }}-${{ matrix.arch }}"
          cmake --build --parallel ([Environment]::ProcessorCount) --preset "${{ matrix.preset }}"
-          cmake --install build --component "${{ startsWith(matrix.preset, 'CUDA ') && 'CUDA' || startsWith(matrix.preset, 'ROCm ') && 'HIP' || startsWith(matrix.preset, 'Vulkan') && 'Vulkan' || 'CPU' }}" --strip
+          cmake --install build --component "${{ startsWith(matrix.preset, 'CUDA ') && 'CUDA' || startsWith(matrix.preset, 'ROCm ') && 'HIP' || 'CPU' }}" --strip
          Remove-Item -Path dist\lib\ollama\rocm\rocblas\library\*gfx906* -ErrorAction SilentlyContinue
        env:
          CMAKE_GENERATOR: Ninja
@@ -255,9 +228,6 @@ jobs:
      - uses: actions/setup-go@v5
        with:
          go-version-file: go.mod
-          cache-dependency-path: |
-            go.sum
-            Makefile.sync
      - name: Verify gcc is actually clang
        run: |
          $ErrorActionPreference='Continue'
@@ -311,9 +281,6 @@ jobs:
      - uses: actions/setup-go@v5
        with:
          go-version-file: go.mod
-          cache-dependency-path: |
-            go.sum
-            Makefile.sync
      - uses: actions/download-artifact@v4
        with:
          pattern: depends-windows*
@@ -337,7 +304,6 @@ jobs:
          name: bundles-windows
          path: |
            dist/*.zip
-            dist/*.ps1
            dist/OllamaSetup.exe

  linux-build:
@@ -346,13 +312,13 @@ jobs:
        include:
          - os: linux
            arch: amd64
-            target: archive
+            target: archive_novulkan
          - os: linux
            arch: amd64
            target: rocm
          - os: linux
            arch: arm64
-            target: archive
+            target: archive_novulkan
    runs-on: ${{ matrix.arch == 'arm64' && format('{0}-{1}', matrix.os, matrix.arch) || matrix.os }}
    environment: release
    needs: setup-environment
@@ -373,17 +339,12 @@ jobs:
          outputs: type=local,dest=dist/${{ matrix.os }}-${{ matrix.arch }}
          cache-from: type=registry,ref=${{ vars.DOCKER_REPO }}:latest
          cache-to: type=inline
-      - name: Deduplicate CUDA libraries
-        run: |
-          ./scripts/deduplicate_cuda_libs.sh dist/${{ matrix.os }}-${{ matrix.arch }}
      - run: |
          for COMPONENT in bin/* lib/ollama/*; do
            case "$COMPONENT" in
-              bin/ollama*)               echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
+              bin/ollama)                echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
              lib/ollama/*.so*)          echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
              lib/ollama/cuda_v*)        echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
-              lib/ollama/vulkan*)        echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
-              lib/ollama/mlx*)           echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
              lib/ollama/cuda_jetpack5)  echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}-jetpack5.tar.in ;;
              lib/ollama/cuda_jetpack6)  echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}-jetpack6.tar.in ;;
              lib/ollama/rocm)           echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}-rocm.tar.in ;;
@@ -398,13 +359,13 @@ jobs:
          done
      - run: |
          for ARCHIVE in dist/${{ matrix.os }}-${{ matrix.arch }}/*.tar.in; do
-            tar c -C dist/${{ matrix.os }}-${{ matrix.arch }} -T $ARCHIVE --owner 0 --group 0 | zstd --ultra -22 -T0 >$(basename ${ARCHIVE//.*/}.tar.zst);
+            tar c -C dist/${{ matrix.os }}-${{ matrix.arch }} -T $ARCHIVE --owner 0 --group 0 | pigz -9vc >$(basename ${ARCHIVE//.*/}.tgz);
          done
      - uses: actions/upload-artifact@v4
        with:
          name: bundles-${{ matrix.os }}-${{ matrix.arch }}-${{ matrix.target }}
          path: |
-            *.tar.zst
+            *.tgz

  # Build each Docker variant (OS, arch, and flavor) separately. Using QEMU is unreliable and slower.
  docker-build-push:
@@ -413,12 +374,14 @@ jobs:
        include:
          - os: linux
            arch: arm64
+            target: novulkan
            build-args: |
              CGO_CFLAGS
              CGO_CXXFLAGS
              GOFLAGS
          - os: linux
            arch: amd64
+            target: novulkan
            build-args: |
              CGO_CFLAGS
              CGO_CXXFLAGS
@@ -431,6 +394,14 @@ jobs:
              CGO_CXXFLAGS
              GOFLAGS
              FLAVOR=rocm
+          - os: linux
+            arch: amd64
+            suffix: '-vulkan'
+            target: default
+            build-args: |
+              CGO_CFLAGS
+              CGO_CXXFLAGS
+              GOFLAGS
    runs-on: ${{ matrix.arch == 'arm64' && format('{0}-{1}', matrix.os, matrix.arch) || matrix.os }}
    environment: release
    needs: setup-environment
@@ -448,6 +419,7 @@ jobs:
        with:
          context: .
          platforms: ${{ matrix.os }}/${{ matrix.arch }}
+          target: ${{ matrix.preset }}
          build-args: ${{ matrix.build-args }}
          outputs: type=image,name=${{ vars.DOCKER_REPO }},push-by-digest=true,name-canonical=true,push=true
          cache-from: type=registry,ref=${{ vars.DOCKER_REPO }}:latest
@@ -515,9 +487,6 @@ jobs:
      - name: Log dist contents
        run: |
          ls -l dist/
-      - name: Copy install scripts to dist
-        run: |
-          cp scripts/install.sh dist/install.sh
      - name: Generate checksum file
        run: find . -type f -not -name 'sha256sum.txt' | xargs sha256sum | tee sha256sum.txt
        working-directory: dist
@@ -540,7 +509,7 @@ jobs:
      - name: Upload release artifacts
        run: |
          pids=()
-          for payload in dist/*.txt dist/*.zip dist/*.tgz dist/*.tar.zst dist/*.exe dist/*.dmg dist/*.ps1 dist/*.sh ; do
+          for payload in dist/*.txt dist/*.zip dist/*.tgz dist/*.exe dist/*.dmg ; do
            echo "Uploading $payload"
            gh release upload ${GITHUB_REF_NAME} $payload --clobber &
            pids[$!]=$!
--- a/.github/workflows/test-install.yaml
+++ b/.github/workflows/test-install.yaml
@@ -1,22 +0,0 @@
-name: test-install
-
-on:
-  pull_request:
-    paths:
-      - 'scripts/install.sh'
-      - '.github/workflows/test-install.yaml'
-
-jobs:
-  test:
-    strategy:
-      matrix:
-        os: [ubuntu-latest, macos-latest]
-    runs-on: ${{ matrix.os }}
-    steps:
-      - uses: actions/checkout@v4
-      - name: Run install script
-        run: sh ./scripts/install.sh
-        env:
-          OLLAMA_NO_START: 1 # do not start app
-      - name: Verify ollama is available
-        run: ollama --version
--- a/.github/workflows/test.yaml
+++ b/.github/workflows/test.yaml
@@ -22,7 +22,6 @@ jobs:
    runs-on: ubuntu-latest
    outputs:
      changed: ${{ steps.changes.outputs.changed }}
-      vendorsha: ${{ steps.changes.outputs.vendorsha }}
    steps:
      - uses: actions/checkout@v4
        with:
@@ -38,7 +37,6 @@ jobs:
          }

          echo changed=$(changed 'llama/llama.cpp/**/*' 'ml/backend/ggml/ggml/**/*') | tee -a $GITHUB_OUTPUT
-          echo vendorsha=$(make -f Makefile.sync print-base) | tee -a $GITHUB_OUTPUT

  linux:
    needs: [changes]
@@ -85,7 +83,7 @@ jobs:
      - uses: actions/cache@v4
        with:
          path: /github/home/.cache/ccache
-          key: ccache-${{ runner.os }}-${{ runner.arch }}-${{ matrix.preset }}-${{ needs.changes.outputs.vendorsha }}
+          key: ccache-${{ runner.os }}-${{ runner.arch }}-${{ matrix.preset }}
      - run: |
          cmake --preset ${{ matrix.preset }} ${{ matrix.flags }}
          cmake --build --preset ${{ matrix.preset }} --parallel
@@ -174,13 +172,12 @@ jobs:
          path: |
            C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA
            C:\Program Files\AMD\ROCm
-            C:\VulkanSDK
          key: ${{ matrix.install }}
      - uses: actions/checkout@v4
      - uses: actions/cache@v4
        with:
          path: ${{ github.workspace }}\.ccache
-          key: ccache-${{ runner.os }}-${{ runner.arch }}-${{ matrix.preset }}-${{ needs.changes.outputs.vendorsha }}
+          key: ccache-${{ runner.os }}-${{ runner.arch }}-${{ matrix.preset }}
      - run: |
          Import-Module 'C:\Program Files\Microsoft Visual Studio\2022\Enterprise\Common7\Tools\Microsoft.VisualStudio.DevShell.dll'
          Enter-VsDevShell -VsInstallPath 'C:\Program Files\Microsoft Visual Studio\2022\Enterprise' -SkipAutomaticLocation  -DevCmdArguments '-arch=x64 -no_logo'
@@ -208,9 +205,6 @@ jobs:
      - uses: actions/setup-go@v5
        with:
          go-version-file: 'go.mod'
-          cache-dependency-path: |
-            go.sum
-            Makefile.sync
      - uses: actions/setup-node@v4
        with:
          node-version: '20'
@@ -231,9 +225,12 @@ jobs:
        if: always()
        run: go test -count=1 -benchtime=1x ./...

-      - uses: golangci/golangci-lint-action@v9
+      # TODO(bmizerany): replace this heavy tool with just the
+      # tools/checks/binaries we want and then make them all run in parallel
+      # across jobs, not on a single tiny vm on Github Actions.
+      - uses: golangci/golangci-lint-action@v6
        with:
-          only-new-issues: true
+          args: --timeout 10m0s -v

  patches:
    runs-on: ubuntu-latest
@@ -242,4 +239,4 @@ jobs:
      - name: Verify patches apply cleanly and do not change files
        run: |
          make -f Makefile.sync clean checkout apply-patches sync
-          git diff --compact-summary --exit-code
+          git diff --compact-summary --exit-code
--- a/.golangci.yaml
+++ b/.golangci.yaml
@@ -1,4 +1,5 @@
-version: "2"
+run:
+  timeout: 5m
 linters:
  enable:
    - asasalint
@@ -6,46 +7,35 @@ linters:
    - bodyclose
    - containedctx
    - gocheckcompilerdirectives
+    - gofmt
+    - gofumpt
+    - gosimple
+    - govet
+    - ineffassign
    - intrange
    - makezero
    - misspell
    - nilerr
    - nolintlint
    - nosprintfhostport
+    - staticcheck
    - unconvert
    - usetesting
    - wastedassign
    - whitespace
  disable:
-    - errcheck
    - usestdlibvars
-  settings:
-    govet:
-      disable:
-        - unusedresult
-    staticcheck:
-      checks:
-        - all
-        - -QF* # disable quick fix suggestions
-        - -SA1019
-        - -ST1000 # package comment format
-        - -ST1003 # underscores in package names
-        - -ST1005 # error strings should not be capitalized
-        - -ST1012 # error var naming (ErrFoo)
-        - -ST1016 # receiver name consistency
-        - -ST1020 # comment on exported function format
-        - -ST1021 # comment on exported type format
-        - -ST1022 # comment on exported var format
-        - -ST1023 # omit type from declaration
+    - errcheck
+linters-settings:
+  staticcheck:
+    checks:
+      - all
+      - -SA1019 # omit Deprecated check
 severity:
-  default: error
+  default-severity: error
  rules:
    - linters:
        - gofmt
        - goimports
        - intrange
      severity: info
-formatters:
-  enable:
-    - gofmt
-    - gofumpt
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -2,22 +2,6 @@ cmake_minimum_required(VERSION 3.21)

 project(Ollama C CXX)

-# Handle cross-compilation on macOS: when CMAKE_OSX_ARCHITECTURES is set to a
-# single architecture different from the host, override CMAKE_SYSTEM_PROCESSOR
-# to match. This is necessary because CMAKE_SYSTEM_PROCESSOR defaults to the
-# host architecture, but downstream projects (like MLX) use it to detect the
-# target architecture.
-if(CMAKE_OSX_ARCHITECTURES AND NOT CMAKE_OSX_ARCHITECTURES MATCHES ";")
-    # Single architecture specified
-    if(CMAKE_OSX_ARCHITECTURES STREQUAL "x86_64" AND NOT CMAKE_SYSTEM_PROCESSOR STREQUAL "x86_64")
-        message(STATUS "Cross-compiling for x86_64: overriding CMAKE_SYSTEM_PROCESSOR from ${CMAKE_SYSTEM_PROCESSOR} to x86_64")
-        set(CMAKE_SYSTEM_PROCESSOR "x86_64")
-    elseif(CMAKE_OSX_ARCHITECTURES STREQUAL "arm64" AND NOT CMAKE_SYSTEM_PROCESSOR STREQUAL "arm64")
-        message(STATUS "Cross-compiling for arm64: overriding CMAKE_SYSTEM_PROCESSOR from ${CMAKE_SYSTEM_PROCESSOR} to arm64")
-        set(CMAKE_SYSTEM_PROCESSOR "arm64")
-    endif()
-endif()
-
 include(CheckLanguage)
 include(GNUInstallDirs)

@@ -28,7 +12,7 @@ set(BUILD_SHARED_LIBS ON)

 set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
-set(CMAKE_CXX_EXTENSIONS ON) # Recent versions of MLX Requires gnu++17 extensions to compile properly
+set(CMAKE_CXX_EXTENSIONS OFF)

 set(GGML_BUILD ON)
 set(GGML_SHARED ON)
@@ -48,10 +32,9 @@ if((CMAKE_OSX_ARCHITECTURES AND NOT CMAKE_OSX_ARCHITECTURES MATCHES "arm64")
    set(GGML_CPU_ALL_VARIANTS ON)
 endif()

-if(APPLE)
+if (CMAKE_OSX_ARCHITECTURES MATCHES "x86_64")
    set(CMAKE_BUILD_RPATH "@loader_path")
    set(CMAKE_INSTALL_RPATH "@loader_path")
-    set(CMAKE_BUILD_WITH_INSTALL_RPATH ON)
 endif()

 set(OLLAMA_BUILD_DIR ${CMAKE_BINARY_DIR}/lib/ollama)
@@ -71,13 +54,6 @@ include_directories(${CMAKE_CURRENT_SOURCE_DIR}/ml/backend/ggml/ggml/src/ggml-cp

 add_compile_definitions(NDEBUG GGML_VERSION=0x0 GGML_COMMIT=0x0)

-# Define GGML version variables for shared library SOVERSION
-# These are required by ggml/src/CMakeLists.txt for proper library versioning
-set(GGML_VERSION_MAJOR 0)
-set(GGML_VERSION_MINOR 0)
-set(GGML_VERSION_PATCH 0)
-set(GGML_VERSION "${GGML_VERSION_MAJOR}.${GGML_VERSION_MINOR}.${GGML_VERSION_PATCH}")
-
 set(GGML_CPU ON)
 add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/ml/backend/ggml/ggml/src)
 set_property(TARGET ggml PROPERTY EXCLUDE_FROM_ALL TRUE)
@@ -164,56 +140,14 @@ if(CMAKE_HIP_COMPILER)
    endif()
 endif()

-if(NOT APPLE)
-    find_package(Vulkan)
-    if(Vulkan_FOUND)
-        add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/ml/backend/ggml/ggml/src/ggml-vulkan)
-        install(TARGETS ggml-vulkan
-            RUNTIME_DEPENDENCIES
-                PRE_INCLUDE_REGEXES vulkan
-                PRE_EXCLUDE_REGEXES ".*"
-            RUNTIME DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT Vulkan
-            LIBRARY DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT Vulkan
-        )
-    endif()
-endif()
-
-option(MLX_ENGINE "Enable MLX backend" OFF)
-
-if(MLX_ENGINE)
-    message(STATUS "Setting up MLX (this takes a while...)")
-    add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/x/imagegen/mlx)
-
-    # Find CUDA toolkit if MLX is built with CUDA support
-    find_package(CUDAToolkit)
-
-    install(TARGETS mlx mlxc
+find_package(Vulkan)
+if(Vulkan_FOUND)
+    add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/ml/backend/ggml/ggml/src/ggml-vulkan)
+    install(TARGETS ggml-vulkan
        RUNTIME_DEPENDENCIES
-            DIRECTORIES ${CUDAToolkit_BIN_DIR} ${CUDAToolkit_BIN_DIR}/x64 ${CUDAToolkit_LIBRARY_DIR}
-            PRE_INCLUDE_REGEXES cublas cublasLt cudart nvrtc nvrtc-builtins cudnn nccl openblas gfortran
+            PRE_INCLUDE_REGEXES vulkan
            PRE_EXCLUDE_REGEXES ".*"
-        RUNTIME DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT MLX
-        LIBRARY DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT MLX
-        FRAMEWORK DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT MLX
+        RUNTIME DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT Vulkan
+        LIBRARY DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT Vulkan
    )
-
-    # Install the Metal library for macOS arm64 (must be colocated with the binary)
-    # Metal backend is only built for arm64, not x86_64
-    if(APPLE AND CMAKE_SYSTEM_PROCESSOR STREQUAL "arm64")
-        install(FILES ${CMAKE_BINARY_DIR}/_deps/mlx-build/mlx/backend/metal/kernels/mlx.metallib
-            DESTINATION ${OLLAMA_INSTALL_DIR}
-            COMPONENT MLX)
-    endif()
-
-    # Manually install cudart and cublas since they might not be picked up as direct dependencies
-    if(CUDAToolkit_FOUND)
-        file(GLOB CUDART_LIBS
-            "${CUDAToolkit_LIBRARY_DIR}/libcudart.so*"
-            "${CUDAToolkit_LIBRARY_DIR}/libcublas.so*")
-        if(CUDART_LIBS)
-            install(FILES ${CUDART_LIBS}
-                DESTINATION ${OLLAMA_INSTALL_DIR}
-                COMPONENT MLX)
-        endif()
-    endif()
 endif()
--- a/CMakePresets.json
+++ b/CMakePresets.json
@@ -41,7 +41,7 @@
      "inherits": [ "CUDA" ],
      "cacheVariables": {
        "CMAKE_CUDA_ARCHITECTURES": "75-virtual;80-virtual;86-virtual;87-virtual;89-virtual;90-virtual;90a-virtual;100-virtual;103-virtual;110-virtual;120-virtual;121-virtual",
-        "CMAKE_CUDA_FLAGS": "-t 4",
+        "CMAKE_CUDA_FLAGS": "-t 2",
        "OLLAMA_RUNNER_DIR": "cuda_v13"
      }
    },
@@ -83,28 +83,6 @@
      "cacheVariables": {
        "OLLAMA_RUNNER_DIR": "vulkan"
      }
-    },
-    {
-      "name": "MLX",
-      "inherits": [ "Default" ],
-      "cacheVariables": {
-        "MLX_ENGINE": "ON",
-        "OLLAMA_RUNNER_DIR": "mlx"
-      }
-    },
-    {
-      "name": "MLX CUDA 12",
-      "inherits": [ "MLX", "CUDA 12" ],
-      "cacheVariables": {
-        "OLLAMA_RUNNER_DIR": "mlx_cuda_v12"
-      }
-    },
-    {
-      "name": "MLX CUDA 13",
-      "inherits": [ "MLX", "CUDA 13" ],
-      "cacheVariables": {
-        "OLLAMA_RUNNER_DIR": "mlx_cuda_v13"
-      }
    }
  ],
  "buildPresets": [
@@ -162,21 +140,6 @@
      "name": "Vulkan",
      "targets": [ "ggml-vulkan" ],
      "configurePreset": "Vulkan"
-    },
-    {
-      "name": "MLX",
-      "targets": [ "mlx", "mlxc" ],
-      "configurePreset": "MLX"
-    },
-    {
-      "name": "MLX CUDA 12",
-      "targets": [ "mlx", "mlxc" ],
-      "configurePreset": "MLX CUDA 12"
-    },
-    {
-      "name": "MLX CUDA 13",
-      "targets": [ "mlx", "mlxc" ],
-      "configurePreset": "MLX CUDA 13"
    }
  ]
 }
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -16,7 +16,7 @@ See the [development documentation](./docs/development.md) for instructions on h

 * New features: new features (e.g. API fields, environment variables) add surface area to Ollama and make it harder to maintain in the long run as they cannot be removed without potentially breaking users in the future.
 * Refactoring: large code improvements are important, but can be harder or take longer to review and merge.
-* Documentation: small updates to fill in or correct missing documentation are helpful, however large documentation additions can be hard to maintain over time.
+* Documentation: small updates to fill in or correct missing documentation is helpful, however large documentation additions can be hard to maintain over time.

 ### Issues that may not be accepted

@@ -43,7 +43,7 @@ Tips for proposals:
 * Explain how the change will be tested.

 Additionally, for bonus points: Provide draft documentation you would expect to
-see if the changes were accepted.
+see if the change were accepted.

 ## Pull requests

@@ -66,6 +66,7 @@ Examples:

      llm/backend/mlx: support the llama architecture
      CONTRIBUTING: provide clarity on good commit messages, and bad
+      docs: simplify manual installation with shorter curl commands

 Bad Examples:

--- a/78
+++ b/78
@@ -32,21 +32,21 @@ ENV PATH=/${VULKANVERSION}/x86_64/bin:$PATH
 FROM --platform=linux/arm64 almalinux:8 AS base-arm64
 # install epel-release for ccache
 RUN yum install -y yum-utils epel-release \
-    && dnf install -y clang ccache git \
+    && dnf install -y clang ccache \
    && yum-config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel8/sbsa/cuda-rhel8.repo
 ENV CC=clang CXX=clang++

 FROM base-${TARGETARCH} AS base
 ARG CMAKEVERSION
 RUN curl -fsSL https://github.com/Kitware/CMake/releases/download/v${CMAKEVERSION}/cmake-${CMAKEVERSION}-linux-$(uname -m).tar.gz | tar xz -C /usr/local --strip-components 1
+COPY CMakeLists.txt CMakePresets.json .
+COPY ml/backend/ggml/ggml ml/backend/ggml/ggml
 ENV LDFLAGS=-s

 FROM base AS cpu
 RUN dnf install -y gcc-toolset-11-gcc gcc-toolset-11-gcc-c++
 ENV PATH=/opt/rh/gcc-toolset-11/root/usr/bin:$PATH
 ARG PARALLEL
-COPY CMakeLists.txt CMakePresets.json .
-COPY ml/backend/ggml/ggml ml/backend/ggml/ggml
 RUN --mount=type=cache,target=/root/.ccache \
    cmake --preset 'CPU' \
        && cmake --build --parallel ${PARALLEL} --preset 'CPU' \
@@ -57,8 +57,6 @@ ARG CUDA11VERSION=11.8
 RUN dnf install -y cuda-toolkit-${CUDA11VERSION//./-}
 ENV PATH=/usr/local/cuda-11/bin:$PATH
 ARG PARALLEL
-COPY CMakeLists.txt CMakePresets.json .
-COPY ml/backend/ggml/ggml ml/backend/ggml/ggml
 RUN --mount=type=cache,target=/root/.ccache \
    cmake --preset 'CUDA 11' \
        && cmake --build --parallel ${PARALLEL} --preset 'CUDA 11' \
@@ -69,8 +67,6 @@ ARG CUDA12VERSION=12.8
 RUN dnf install -y cuda-toolkit-${CUDA12VERSION//./-}
 ENV PATH=/usr/local/cuda-12/bin:$PATH
 ARG PARALLEL
-COPY CMakeLists.txt CMakePresets.json .
-COPY ml/backend/ggml/ggml ml/backend/ggml/ggml
 RUN --mount=type=cache,target=/root/.ccache \
    cmake --preset 'CUDA 12' \
        && cmake --build --parallel ${PARALLEL} --preset 'CUDA 12' \
@@ -82,8 +78,6 @@ ARG CUDA13VERSION=13.0
 RUN dnf install -y cuda-toolkit-${CUDA13VERSION//./-}
 ENV PATH=/usr/local/cuda-13/bin:$PATH
 ARG PARALLEL
-COPY CMakeLists.txt CMakePresets.json .
-COPY ml/backend/ggml/ggml ml/backend/ggml/ggml
 RUN --mount=type=cache,target=/root/.ccache \
    cmake --preset 'CUDA 13' \
        && cmake --build --parallel ${PARALLEL} --preset 'CUDA 13' \
@@ -93,8 +87,6 @@ RUN --mount=type=cache,target=/root/.ccache \
 FROM base AS rocm-6
 ENV PATH=/opt/rocm/hcc/bin:/opt/rocm/hip/bin:/opt/rocm/bin:/opt/rocm/hcc/bin:$PATH
 ARG PARALLEL
-COPY CMakeLists.txt CMakePresets.json .
-COPY ml/backend/ggml/ggml ml/backend/ggml/ggml
 RUN --mount=type=cache,target=/root/.ccache \
    cmake --preset 'ROCm 6' \
        && cmake --build --parallel ${PARALLEL} --preset 'ROCm 6' \
@@ -126,37 +118,11 @@ RUN --mount=type=cache,target=/root/.ccache \
        && cmake --install build --component CUDA --strip --parallel ${PARALLEL}

 FROM base AS vulkan
-COPY CMakeLists.txt CMakePresets.json .
-COPY ml/backend/ggml/ggml ml/backend/ggml/ggml
 RUN --mount=type=cache,target=/root/.ccache \
    cmake --preset 'Vulkan' \
        && cmake --build --parallel --preset 'Vulkan' \
-        && cmake --install build --component Vulkan --strip --parallel 8
+        && cmake --install build --component Vulkan --strip --parallel 8 

-FROM base AS mlx
-ARG CUDA13VERSION=13.0
-RUN dnf install -y cuda-toolkit-${CUDA13VERSION//./-} \
-    && dnf install -y openblas-devel lapack-devel \
-    && dnf install -y libcudnn9-cuda-13 libcudnn9-devel-cuda-13 \
-    && dnf install -y libnccl libnccl-devel
-ENV PATH=/usr/local/cuda-13/bin:$PATH
-ENV BLAS_INCLUDE_DIRS=/usr/include/openblas
-ENV LAPACK_INCLUDE_DIRS=/usr/include/openblas
-ENV CGO_LDFLAGS="-L/usr/local/cuda-13/lib64 -L/usr/local/cuda-13/targets/x86_64-linux/lib/stubs"
-ARG PARALLEL
-WORKDIR /go/src/github.com/ollama/ollama
-COPY CMakeLists.txt CMakePresets.json .
-COPY ml/backend/ggml/ggml ml/backend/ggml/ggml
-COPY x/imagegen/mlx x/imagegen/mlx
-COPY go.mod go.sum .
-COPY MLX_VERSION .
-RUN curl -fsSL https://golang.org/dl/go$(awk '/^go/ { print $2 }' go.mod).linux-$(case $(uname -m) in x86_64) echo amd64 ;; aarch64) echo arm64 ;; esac).tar.gz | tar xz -C /usr/local
-ENV PATH=/usr/local/go/bin:$PATH
-RUN go mod download
-RUN --mount=type=cache,target=/root/.ccache \
-    cmake --preset 'MLX CUDA 13' -DBLAS_INCLUDE_DIRS=/usr/include/openblas -DLAPACK_INCLUDE_DIRS=/usr/include/openblas \
-        && cmake --build --parallel ${PARALLEL} --preset 'MLX CUDA 13' \
-        && cmake --install build --component MLX --strip --parallel ${PARALLEL}

 FROM base AS build
 WORKDIR /go/src/github.com/ollama/ollama
@@ -165,23 +131,18 @@ RUN curl -fsSL https://golang.org/dl/go$(awk '/^go/ { print $2 }' go.mod).linux-
 ENV PATH=/usr/local/go/bin:$PATH
 RUN go mod download
 COPY . .
-# Clone mlx-c headers for CGO (version from MLX_VERSION file)
-RUN git clone --depth 1 --branch "$(cat MLX_VERSION)" https://github.com/ml-explore/mlx-c.git build/_deps/mlx-c-src
 ARG GOFLAGS="'-ldflags=-w -s'"
 ENV CGO_ENABLED=1
 ARG CGO_CFLAGS
 ARG CGO_CXXFLAGS
-ENV CGO_CFLAGS="${CGO_CFLAGS} -I/go/src/github.com/ollama/ollama/build/_deps/mlx-c-src"
-ENV CGO_CXXFLAGS="${CGO_CXXFLAGS}"
 RUN --mount=type=cache,target=/root/.cache/go-build \
-    go build -tags mlx -trimpath -buildmode=pie -o /bin/ollama .
+    go build -trimpath -buildmode=pie -o /bin/ollama .

 FROM --platform=linux/amd64 scratch AS amd64
 # COPY --from=cuda-11 dist/lib/ollama/ /lib/ollama/
 COPY --from=cuda-12 dist/lib/ollama /lib/ollama/
 COPY --from=cuda-13 dist/lib/ollama /lib/ollama/
 COPY --from=vulkan  dist/lib/ollama  /lib/ollama/
-COPY --from=mlx     /go/src/github.com/ollama/ollama/dist/lib/ollama /lib/ollama/

 FROM --platform=linux/arm64 scratch AS arm64
 # COPY --from=cuda-11 dist/lib/ollama/ /lib/ollama/
@@ -198,9 +159,34 @@ ARG VULKANVERSION
 COPY --from=cpu dist/lib/ollama /lib/ollama
 COPY --from=build /bin/ollama /bin/ollama

-FROM ubuntu:24.04
+# Temporary opt-out stages for Vulkan
+FROM --platform=linux/amd64 scratch AS amd64_novulkan
+# COPY --from=cuda-11 dist/lib/ollama/ /lib/ollama/
+COPY --from=cuda-12 dist/lib/ollama /lib/ollama/
+COPY --from=cuda-13 dist/lib/ollama /lib/ollama/
+FROM arm64 AS arm64_novulkan
+FROM ${FLAVOR}_novulkan AS archive_novulkan
+COPY --from=cpu dist/lib/ollama /lib/ollama
+COPY --from=build /bin/ollama /bin/ollama
+FROM ubuntu:24.04 AS novulkan
 RUN apt-get update \
-    && apt-get install -y ca-certificates libvulkan1 libopenblas0 \
+    && apt-get install -y ca-certificates \
+    && apt-get clean \
+    && rm -rf /var/lib/apt/lists/*
+COPY --from=archive_novulkan /bin /usr/bin
+ENV PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
+COPY --from=archive_novulkan /lib/ollama /usr/lib/ollama
+ENV LD_LIBRARY_PATH=/usr/local/nvidia/lib:/usr/local/nvidia/lib64
+ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility
+ENV NVIDIA_VISIBLE_DEVICES=all
+ENV OLLAMA_HOST=0.0.0.0:11434
+EXPOSE 11434
+ENTRYPOINT ["/bin/ollama"]
+CMD ["serve"]
+
+FROM ubuntu:24.04 AS default
+RUN apt-get update \
+    && apt-get install -y ca-certificates libvulkan1 \
    && apt-get clean \
    && rm -rf /var/lib/apt/lists/*
 COPY --from=archive /bin /usr/bin
--- a/1
+++ b/1
@@ -1 +0,0 @@
-v0.4.1
--- a/Makefile.sync
+++ b/Makefile.sync
@@ -1,6 +1,6 @@
 UPSTREAM=https://github.com/ggml-org/llama.cpp.git
 WORKDIR=llama/vendor
-FETCH_HEAD=ec98e2002
+FETCH_HEAD=3cfa9c3f125763305b4226bc032f1954f08990dc

 .PHONY: help
 help:
@@ -57,7 +57,7 @@ checkout: $(WORKDIR)
 $(WORKDIR):
 	git clone $(UPSTREAM) $(WORKDIR)

-.PHONY: format-patches
+.PHONE: format-patches
 format-patches: llama/patches
 	git -C $(WORKDIR) format-patch \
 		--no-signature \
@@ -66,11 +66,7 @@ format-patches: llama/patches
 		-o $(realpath $<) \
 		$(FETCH_HEAD)

-.PHONY: clean
+.PHONE: clean
 clean: checkout
 	@git -C $(WORKDIR) am --abort || true
 	$(RM) llama/patches/.*.patched
-
-.PHONY: print-base
-print-base:
-	@echo $(FETCH_HEAD)
--- a/README.md
+++ b/README.md
@@ -48,7 +48,7 @@ ollama run gemma3

 ## Model library

-Ollama supports a list of models available on [ollama.com/library](https://ollama.com/library "ollama model library")
+Ollama supports a list of models available on [ollama.com/library](https://ollama.com/library 'ollama model library')

 Here are some example models that can be downloaded:

@@ -79,7 +79,7 @@ Here are some example models that can be downloaded:
 | Code Llama         | 7B         | 3.8GB | `ollama run codellama`           |
 | Llama 2 Uncensored | 7B         | 3.8GB | `ollama run llama2-uncensored`   |
 | LLaVA              | 7B         | 4.5GB | `ollama run llava`               |
-| Granite-3.3        | 8B         | 4.9GB | `ollama run granite3.3`          |
+| Granite-3.3         | 8B         | 4.9GB | `ollama run granite3.3`          |

 > [!NOTE]
 > You should have at least 8 GB of RAM available to run the 7B models, 16 GB to run the 13B models, and 32 GB to run the 33B models.
@@ -260,38 +260,6 @@ Finally, in a separate shell, run a model:
 ./ollama run llama3.2
 ```

-## Building with MLX (experimental)
-
-First build the MLX libraries:
-
-```shell
-cmake --preset MLX
-cmake --build --preset MLX --parallel
-cmake --install build --component MLX
-```
-
-When building with the `-tags mlx` flag, the main `ollama` binary includes MLX support for experimental features like image generation:
-
-```shell
-go build -tags mlx .
-```
-
-Finally, start the server:
-
-```
-./ollama serve
-```
-
-### Building MLX with CUDA
-
-When building with CUDA, use the preset "MLX CUDA 13" or "MLX CUDA 12" to enable CUDA with default architectures:
-
-```shell
-cmake --preset 'MLX CUDA 13'
-cmake --build --preset 'MLX CUDA 13' --parallel
-cmake --install build --component MLX
-```
-
 ## REST API

 Ollama has a REST API for running and managing models.
@@ -322,7 +290,6 @@ See the [API documentation](./docs/api.md) for all endpoints.

 ### Web & Desktop

- [Onyx](https://github.com/onyx-dot-app/onyx)
 - [Open WebUI](https://github.com/open-webui/open-webui)
 - [SwiftChat (macOS with ReactNative)](https://github.com/aws-samples/swift-chat)
 - [Enchanted (macOS native)](https://github.com/AugustDev/enchanted)
@@ -332,7 +299,6 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [LibreChat](https://github.com/danny-avila/LibreChat)
 - [Bionic GPT](https://github.com/bionic-gpt/bionic-gpt)
 - [HTML UI](https://github.com/rtcfirefly/ollama-ui)
- [AI-UI](https://github.com/bajahaw/ai-ui)
 - [Saddle](https://github.com/jikkuatwork/saddle)
 - [TagSpaces](https://www.tagspaces.org) (A platform for file-based apps, [utilizing Ollama](https://docs.tagspaces.org/ai/) for the generation of tags and descriptions)
 - [Chatbot UI](https://github.com/ivanfioravanti/chatbot-ollama)
@@ -358,7 +324,6 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [Odin Runes](https://github.com/leonid20000/OdinRunes)
 - [LLM-X](https://github.com/mrdjohnson/llm-x) (Progressive Web App)
 - [AnythingLLM (Docker + MacOs/Windows/Linux native app)](https://github.com/Mintplex-Labs/anything-llm)
- [Screenpipe](https://github.com/mediar-ai/screenpipe) (24/7 screen & mic recording with AI-powered search, uses Ollama for local LLM features)
 - [Ollama Basic Chat: Uses HyperDiv Reactive UI](https://github.com/rapidarchitect/ollama_basic_chat)
 - [Ollama-chats RPG](https://github.com/drazdra/ollama-chats)
 - [IntelliBar](https://intellibar.app/) (AI-powered assistant for macOS)
@@ -400,8 +365,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [PartCAD](https://github.com/openvmp/partcad/) (CAD model generation with OpenSCAD and CadQuery)
 - [Ollama4j Web UI](https://github.com/ollama4j/ollama4j-web-ui) - Java-based Web UI for Ollama built with Vaadin, Spring Boot, and Ollama4j
 - [PyOllaMx](https://github.com/kspviswa/pyOllaMx) - macOS application capable of chatting with both Ollama and Apple MLX models.
- [Cline](https://github.com/cline/cline) - Formerly known as Claude Dev is a VS Code extension for multi-file/whole-repo coding
- [Void](https://github.com/voideditor/void) (Open source AI code editor and Cursor alternative)
+- [Cline](https://github.com/cline/cline) - Formerly known as Claude Dev is a VSCode extension for multi-file/whole-repo coding
 - [Cherry Studio](https://github.com/kangfenmao/cherry-studio) (Desktop client with Ollama support)
 - [ConfiChat](https://github.com/1runeberg/confichat) (Lightweight, standalone, multi-platform, and privacy-focused LLM chat interface with optional encryption)
 - [Archyve](https://github.com/nickthecook/archyve) (RAG-enabling document library)
@@ -433,7 +397,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [aidful-ollama-model-delete](https://github.com/AidfulAI/aidful-ollama-model-delete) (User interface for simplified model cleanup)
 - [Perplexica](https://github.com/ItzCrazyKns/Perplexica) (An AI-powered search engine & an open-source alternative to Perplexity AI)
 - [Ollama Chat WebUI for Docker ](https://github.com/oslook/ollama-webui) (Support for local docker deployment, lightweight ollama webui)
- [AI Toolkit for Visual Studio Code](https://aka.ms/ai-tooklit/ollama-docs) (Microsoft-official VS Code extension to chat, test, evaluate models with Ollama support, and use them in your AI applications.)
+- [AI Toolkit for Visual Studio Code](https://aka.ms/ai-tooklit/ollama-docs) (Microsoft-official VSCode extension to chat, test, evaluate models with Ollama support, and use them in your AI applications.)
 - [MinimalNextOllamaChat](https://github.com/anilkay/MinimalNextOllamaChat) (Minimal Web UI for Chat and Model Control)
 - [Chipper](https://github.com/TilmanGriesel/chipper) AI interface for tinkerers (Ollama, Haystack RAG, Python)
 - [ChibiChat](https://github.com/CosmicEventHorizon/ChibiChat) (Kotlin-based Android app to chat with Ollama and Koboldcpp API endpoints)
@@ -455,18 +419,16 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [AppFlowy](https://github.com/AppFlowy-IO/AppFlowy) (AI collaborative workspace with Ollama, cross-platform and self-hostable)
 - [Lumina](https://github.com/cushydigit/lumina.git) (A lightweight, minimal React.js frontend for interacting with Ollama servers)
 - [Tiny Notepad](https://pypi.org/project/tiny-notepad) (A lightweight, notepad-like interface to chat with ollama available on PyPI)
- [macLlama (macOS native)](https://github.com/hellotunamayo/macLlama) (A native macOS GUI application for interacting with Ollama models, featuring a chat interface.)
+- [macLlama (macOS native)](https://github.com/hellotunamayo/macLlama) (A native macOS GUI application for interacting with Ollama models, featuring a chat interface.) 
 - [GPTranslate](https://github.com/philberndt/GPTranslate) (A fast and lightweight, AI powered desktop translation application written with Rust and Tauri. Features real-time translation with OpenAI/Azure/Ollama.)
 - [ollama launcher](https://github.com/NGC13009/ollama-launcher) (A launcher for Ollama, aiming to provide users with convenient functions such as ollama server launching, management, or configuration.)
 - [ai-hub](https://github.com/Aj-Seven/ai-hub) (AI Hub supports multiple models via API keys and Chat support via Ollama API.)
 - [Mayan EDMS](https://gitlab.com/mayan-edms/mayan-edms) (Open source document management system to organize, tag, search, and automate your files with powerful Ollama driven workflows.)
 - [Serene Pub](https://github.com/doolijb/serene-pub) (Beginner friendly, open source AI Roleplaying App for Windows, Mac OS and Linux. Search, download and use models with Ollama all inside the app.)
 - [Andes](https://github.com/aqerd/andes) (A Visual Studio Code extension that provides a local UI interface for Ollama models)
- [KDeps](https://github.com/kdeps/kdeps) (Kdeps is an offline-first AI framework for building Dockerized full-stack AI applications declaratively using Apple PKL and integrates APIs with Ollama on the backend.)
 - [Clueless](https://github.com/KashyapTan/clueless) (Open Source & Local Cluely: A desktop application LLM assistant to help you talk to anything on your screen using locally served Ollama models. Also undetectable to screenshare)
 - [ollama-co2](https://github.com/carbonatedWaterOrg/ollama-co2) (FastAPI web interface for monitoring and managing local and remote Ollama servers with real-time model monitoring and concurrent downloads)
 - [Hillnote](https://hillnote.com) (A Markdown-first workspace designed to supercharge your AI workflow. Create documents ready to integrate with Claude, ChatGPT, Gemini, Cursor, and more - all while keeping your work on your device.)
- [Stakpak](https://github.com/stakpak/agent) (An open source, vendor neutral DevOps agent that works with any model, and any stack, for teams who just want to ship)

 ### Cloud

@@ -528,7 +490,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 ### Database

 - [pgai](https://github.com/timescale/pgai) - PostgreSQL as a vector database (Create and search embeddings from Ollama models using pgvector)
-  - [Get started guide](https://github.com/timescale/pgai/blob/main/docs/vectorizer-quick-start.md)
+   - [Get started guide](https://github.com/timescale/pgai/blob/main/docs/vectorizer-quick-start.md)
 - [MindsDB](https://github.com/mindsdb/mindsdb/blob/staging/mindsdb/integrations/handlers/ollama_handler/README.md) (Connects Ollama models with nearly 200 data platforms and apps)
 - [chromem-go](https://github.com/philippgille/chromem-go/blob/v0.5.0/embed_ollama.go) with [example](https://github.com/philippgille/chromem-go/tree/v0.5.0/examples/rag-wikipedia-ollama)
 - [Kangaroo](https://github.com/dbkangaroo/kangaroo) (AI-powered SQL client and admin tool for popular databases)
@@ -560,7 +522,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [LiteLLM](https://github.com/BerriAI/litellm)
 - [OllamaFarm for Go](https://github.com/presbrey/ollamafarm)
 - [OllamaSharp for .NET](https://github.com/awaescher/OllamaSharp)
- [Ollama for Ruby](https://github.com/crmne/ruby_llm)
+- [Ollama for Ruby](https://github.com/gbaptista/ollama-ai)
 - [Ollama-rs for Rust](https://github.com/pepperoni21/ollama-rs)
 - [Ollama-hpp for C++](https://github.com/jmont-dev/ollama-hpp)
 - [Ollama4j for Java](https://github.com/ollama4j/ollama4j)
@@ -590,7 +552,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [Parakeet](https://github.com/parakeet-nest/parakeet) is a GoLang library, made to simplify the development of small generative AI applications with Ollama.
 - [Haverscript](https://github.com/andygill/haverscript) with [examples](https://github.com/andygill/haverscript/tree/main/examples)
 - [Ollama for Swift](https://github.com/mattt/ollama-swift)
- [Swollama for Swift](https://github.com/guitaripod/Swollama) with [DocC](https://guitaripod.github.io/Swollama/documentation/swollama)
+- [Swollama for Swift](https://github.com/marcusziade/Swollama) with [DocC](https://marcusziade.github.io/Swollama/documentation/swollama/)
 - [GoLamify](https://github.com/prasad89/golamify)
 - [Ollama for Haskell](https://github.com/tusharad/ollama-haskell)
 - [multi-llm-ts](https://github.com/nbonamy/multi-llm-ts) (A Typescript/JavaScript library allowing access to different LLM in a unified API)
@@ -653,7 +615,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [LSP-AI](https://github.com/SilasMarvin/lsp-ai) (Open-source language server for AI-powered functionality)
 - [QodeAssist](https://github.com/Palm1r/QodeAssist) (AI-powered coding assistant plugin for Qt Creator)
 - [Obsidian Quiz Generator plugin](https://github.com/ECuiDev/obsidian-quiz-generator)
- [AI Summary Helper plugin](https://github.com/philffm/ai-summary-helper)
+- [AI Summmary Helper plugin](https://github.com/philffm/ai-summary-helper)
 - [TextCraft](https://github.com/suncloudsmoon/TextCraft) (Copilot in Word alternative using Ollama)
 - [Alfred Ollama](https://github.com/zeitlings/alfred-ollama) (Alfred Workflow)
 - [TextLLaMA](https://github.com/adarshM84/TextLLaMA) A Chrome Extension that helps you write emails, correct grammar, and translate into any language
@@ -661,7 +623,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [LLM Telegram Bot](https://github.com/innightwolfsleep/llm_telegram_bot) (telegram bot, primary for RP. Oobabooga-like buttons, [A1111](https://github.com/AUTOMATIC1111/stable-diffusion-webui) API integration e.t.c)
 - [mcp-llm](https://github.com/sammcj/mcp-llm) (MCP Server to allow LLMs to call other LLMs)
 - [SimpleOllamaUnity](https://github.com/HardCodeDev777/SimpleOllamaUnity) (Unity Engine extension for communicating with Ollama in a few lines of code. Also works at runtime)
- [UnityCodeLama](https://github.com/HardCodeDev777/UnityCodeLama) (Unity Editor tool to analyze scripts via Ollama)
+- [UnityCodeLama](https://github.com/HardCodeDev777/UnityCodeLama) (Unity Edtior tool to analyze scripts via Ollama)
 - [NativeMind](https://github.com/NativeMindBrowser/NativeMindExtension) (Private, on-device AI Assistant, no cloud dependencies)
 - [GMAI - Gradle Managed AI](https://gmai.premex.se/) (Gradle plugin for automated Ollama lifecycle management during build phases)
 - [NOMYO Router](https://github.com/nomyo-ai/nomyo-router) (A transparent Ollama proxy with model deployment aware routing which auto-manages multiple Ollama instances in a given network)
@@ -671,14 +633,9 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [llama.cpp](https://github.com/ggml-org/llama.cpp) project founded by Georgi Gerganov.

 ### Observability
-
- [Opik](https://www.comet.com/docs/opik/cookbook/ollama) is an open-source platform to debug, evaluate, and monitor your LLM applications, RAG systems, and agentic workflows with comprehensive tracing, automated evaluations, and production-ready dashboards. Opik supports native integration to Ollama.
+- [Opik](https://www.comet.com/docs/opik/cookbook/ollama) is an open-source platform to debug, evaluate, and monitor your LLM applications, RAG systems, and agentic workflows with comprehensive tracing, automated evaluations, and production-ready dashboards. Opik supports native intergration to Ollama.
 - [Lunary](https://lunary.ai/docs/integrations/ollama) is the leading open-source LLM observability platform. It provides a variety of enterprise-grade features such as real-time analytics, prompt templates management, PII masking, and comprehensive agent tracing.
 - [OpenLIT](https://github.com/openlit/openlit) is an OpenTelemetry-native tool for monitoring Ollama Applications & GPUs using traces and metrics.
 - [HoneyHive](https://docs.honeyhive.ai/integrations/ollama) is an AI observability and evaluation platform for AI agents. Use HoneyHive to evaluate agent performance, interrogate failures, and monitor quality in production.
 - [Langfuse](https://langfuse.com/docs/integrations/ollama) is an open source LLM observability platform that enables teams to collaboratively monitor, evaluate and debug AI applications.
 - [MLflow Tracing](https://mlflow.org/docs/latest/llms/tracing/index.html#automatic-tracing) is an open source LLM observability tool with a convenient API to log and visualize traces, making it easy to debug and evaluate GenAI applications.
-
-### Security
-
- [Ollama Fortress](https://github.com/ParisNeo/ollama_proxy_server)
--- a/SECURITY.md
+++ b/SECURITY.md
@@ -14,7 +14,7 @@ Please include the following details in your report:

 ## Security best practices

-While the maintainer team does its best to secure Ollama, users are encouraged to implement their own security best practices, such as:
+While the maintainer team does their best to secure Ollama, users are encouraged to implement their own security best practices, such as:

 - Regularly updating to the latest version of Ollama
 - Securing access to hosted instances of Ollama
--- a/anthropic/anthropic.go
+++ b/anthropic/anthropic.go
@@ -1,901 +0,0 @@
-package anthropic
-
-import (
-	"crypto/rand"
-	"encoding/base64"
-	"encoding/json"
-	"errors"
-	"fmt"
-	"log/slog"
-	"net/http"
-	"strings"
-	"time"
-
-	"github.com/ollama/ollama/api"
-)
-
-// Error types matching Anthropic API
-type Error struct {
-	Type    string `json:"type"`
-	Message string `json:"message"`
-}
-
-type ErrorResponse struct {
-	Type      string `json:"type"` // always "error"
-	Error     Error  `json:"error"`
-	RequestID string `json:"request_id,omitempty"`
-}
-
-// NewError creates a new ErrorResponse with the appropriate error type based on HTTP status code
-func NewError(code int, message string) ErrorResponse {
-	var etype string
-	switch code {
-	case http.StatusBadRequest:
-		etype = "invalid_request_error"
-	case http.StatusUnauthorized:
-		etype = "authentication_error"
-	case http.StatusForbidden:
-		etype = "permission_error"
-	case http.StatusNotFound:
-		etype = "not_found_error"
-	case http.StatusTooManyRequests:
-		etype = "rate_limit_error"
-	case http.StatusServiceUnavailable, 529:
-		etype = "overloaded_error"
-	default:
-		etype = "api_error"
-	}
-
-	return ErrorResponse{
-		Type:      "error",
-		Error:     Error{Type: etype, Message: message},
-		RequestID: generateID("req"),
-	}
-}
-
-// Request types
-
-// MessagesRequest represents an Anthropic Messages API request
-type MessagesRequest struct {
-	Model         string          `json:"model"`
-	MaxTokens     int             `json:"max_tokens"`
-	Messages      []MessageParam  `json:"messages"`
-	System        any             `json:"system,omitempty"` // string or []ContentBlock
-	Stream        bool            `json:"stream,omitempty"`
-	Temperature   *float64        `json:"temperature,omitempty"`
-	TopP          *float64        `json:"top_p,omitempty"`
-	TopK          *int            `json:"top_k,omitempty"`
-	StopSequences []string        `json:"stop_sequences,omitempty"`
-	Tools         []Tool          `json:"tools,omitempty"`
-	ToolChoice    *ToolChoice     `json:"tool_choice,omitempty"`
-	Thinking      *ThinkingConfig `json:"thinking,omitempty"`
-	Metadata      *Metadata       `json:"metadata,omitempty"`
-}
-
-// MessageParam represents a message in the request
-type MessageParam struct {
-	Role    string `json:"role"`    // "user" or "assistant"
-	Content any    `json:"content"` // string or []ContentBlock
-}
-
-// ContentBlock represents a content block in a message.
-// Text and Thinking use pointers so they serialize as the field being present (even if empty)
-// only when set, which is required for SDK streaming accumulation.
-type ContentBlock struct {
-	Type string `json:"type"` // text, image, tool_use, tool_result, thinking
-
-	// For text blocks - pointer so field only appears when set (SDK requires it for accumulation)
-	Text *string `json:"text,omitempty"`
-
-	// For image blocks
-	Source *ImageSource `json:"source,omitempty"`
-
-	// For tool_use blocks
-	ID    string `json:"id,omitempty"`
-	Name  string `json:"name,omitempty"`
-	Input any    `json:"input,omitempty"`
-
-	// For tool_result blocks
-	ToolUseID string `json:"tool_use_id,omitempty"`
-	Content   any    `json:"content,omitempty"` // string or []ContentBlock
-	IsError   bool   `json:"is_error,omitempty"`
-
-	// For thinking blocks - pointer so field only appears when set (SDK requires it for accumulation)
-	Thinking  *string `json:"thinking,omitempty"`
-	Signature string  `json:"signature,omitempty"`
-}
-
-// ImageSource represents the source of an image
-type ImageSource struct {
-	Type      string `json:"type"` // "base64" or "url"
-	MediaType string `json:"media_type,omitempty"`
-	Data      string `json:"data,omitempty"`
-	URL       string `json:"url,omitempty"`
-}
-
-// Tool represents a tool definition
-type Tool struct {
-	Type        string          `json:"type,omitempty"` // "custom" for user-defined tools
-	Name        string          `json:"name"`
-	Description string          `json:"description,omitempty"`
-	InputSchema json.RawMessage `json:"input_schema,omitempty"`
-}
-
-// ToolChoice controls how the model uses tools
-type ToolChoice struct {
-	Type                   string `json:"type"` // "auto", "any", "tool", "none"
-	Name                   string `json:"name,omitempty"`
-	DisableParallelToolUse bool   `json:"disable_parallel_tool_use,omitempty"`
-}
-
-// ThinkingConfig controls extended thinking
-type ThinkingConfig struct {
-	Type         string `json:"type"` // "enabled" or "disabled"
-	BudgetTokens int    `json:"budget_tokens,omitempty"`
-}
-
-// Metadata for the request
-type Metadata struct {
-	UserID string `json:"user_id,omitempty"`
-}
-
-// Response types
-
-// MessagesResponse represents an Anthropic Messages API response
-type MessagesResponse struct {
-	ID           string         `json:"id"`
-	Type         string         `json:"type"` // "message"
-	Role         string         `json:"role"` // "assistant"
-	Model        string         `json:"model"`
-	Content      []ContentBlock `json:"content"`
-	StopReason   string         `json:"stop_reason,omitempty"`
-	StopSequence string         `json:"stop_sequence,omitempty"`
-	Usage        Usage          `json:"usage"`
-}
-
-// Usage contains token usage information
-type Usage struct {
-	InputTokens  int `json:"input_tokens"`
-	OutputTokens int `json:"output_tokens"`
-}
-
-// Streaming event types
-
-// MessageStartEvent is sent at the start of streaming
-type MessageStartEvent struct {
-	Type    string           `json:"type"` // "message_start"
-	Message MessagesResponse `json:"message"`
-}
-
-// ContentBlockStartEvent signals the start of a content block
-type ContentBlockStartEvent struct {
-	Type         string       `json:"type"` // "content_block_start"
-	Index        int          `json:"index"`
-	ContentBlock ContentBlock `json:"content_block"`
-}
-
-// ContentBlockDeltaEvent contains incremental content updates
-type ContentBlockDeltaEvent struct {
-	Type  string `json:"type"` // "content_block_delta"
-	Index int    `json:"index"`
-	Delta Delta  `json:"delta"`
-}
-
-// Delta represents an incremental update
-type Delta struct {
-	Type        string `json:"type"` // "text_delta", "input_json_delta", "thinking_delta", "signature_delta"
-	Text        string `json:"text,omitempty"`
-	PartialJSON string `json:"partial_json,omitempty"`
-	Thinking    string `json:"thinking,omitempty"`
-	Signature   string `json:"signature,omitempty"`
-}
-
-// ContentBlockStopEvent signals the end of a content block
-type ContentBlockStopEvent struct {
-	Type  string `json:"type"` // "content_block_stop"
-	Index int    `json:"index"`
-}
-
-// MessageDeltaEvent contains updates to the message
-type MessageDeltaEvent struct {
-	Type  string       `json:"type"` // "message_delta"
-	Delta MessageDelta `json:"delta"`
-	Usage DeltaUsage   `json:"usage"`
-}
-
-// MessageDelta contains stop information
-type MessageDelta struct {
-	StopReason   string `json:"stop_reason,omitempty"`
-	StopSequence string `json:"stop_sequence,omitempty"`
-}
-
-// DeltaUsage contains cumulative token usage
-type DeltaUsage struct {
-	InputTokens  int `json:"input_tokens"`
-	OutputTokens int `json:"output_tokens"`
-}
-
-// MessageStopEvent signals the end of the message
-type MessageStopEvent struct {
-	Type string `json:"type"` // "message_stop"
-}
-
-// PingEvent is a keepalive event
-type PingEvent struct {
-	Type string `json:"type"` // "ping"
-}
-
-// StreamErrorEvent is an error during streaming
-type StreamErrorEvent struct {
-	Type  string `json:"type"` // "error"
-	Error Error  `json:"error"`
-}
-
-// FromMessagesRequest converts an Anthropic MessagesRequest to an Ollama api.ChatRequest
-func FromMessagesRequest(r MessagesRequest) (*api.ChatRequest, error) {
-	var messages []api.Message
-
-	if r.System != nil {
-		switch sys := r.System.(type) {
-		case string:
-			if sys != "" {
-				messages = append(messages, api.Message{Role: "system", Content: sys})
-			}
-		case []any:
-			// System can be an array of content blocks
-			var content strings.Builder
-			for _, block := range sys {
-				if blockMap, ok := block.(map[string]any); ok {
-					if blockMap["type"] == "text" {
-						if text, ok := blockMap["text"].(string); ok {
-							content.WriteString(text)
-						}
-					}
-				}
-			}
-			if content.Len() > 0 {
-				messages = append(messages, api.Message{Role: "system", Content: content.String()})
-			}
-		}
-	}
-
-	for _, msg := range r.Messages {
-		converted, err := convertMessage(msg)
-		if err != nil {
-			return nil, err
-		}
-		messages = append(messages, converted...)
-	}
-
-	options := make(map[string]any)
-
-	options["num_predict"] = r.MaxTokens
-
-	if r.Temperature != nil {
-		options["temperature"] = *r.Temperature
-	}
-
-	if r.TopP != nil {
-		options["top_p"] = *r.TopP
-	}
-
-	if r.TopK != nil {
-		options["top_k"] = *r.TopK
-	}
-
-	if len(r.StopSequences) > 0 {
-		options["stop"] = r.StopSequences
-	}
-
-	var tools api.Tools
-	for _, t := range r.Tools {
-		tool, err := convertTool(t)
-		if err != nil {
-			return nil, err
-		}
-		tools = append(tools, tool)
-	}
-
-	var think *api.ThinkValue
-	if r.Thinking != nil && r.Thinking.Type == "enabled" {
-		think = &api.ThinkValue{Value: true}
-	}
-
-	stream := r.Stream
-
-	return &api.ChatRequest{
-		Model:    r.Model,
-		Messages: messages,
-		Options:  options,
-		Stream:   &stream,
-		Tools:    tools,
-		Think:    think,
-	}, nil
-}
-
-// convertMessage converts an Anthropic MessageParam to Ollama api.Message(s)
-func convertMessage(msg MessageParam) ([]api.Message, error) {
-	var messages []api.Message
-	role := strings.ToLower(msg.Role)
-
-	switch content := msg.Content.(type) {
-	case string:
-		messages = append(messages, api.Message{Role: role, Content: content})
-
-	case []any:
-		var textContent strings.Builder
-		var images []api.ImageData
-		var toolCalls []api.ToolCall
-		var thinking string
-		var toolResults []api.Message
-
-		for _, block := range content {
-			blockMap, ok := block.(map[string]any)
-			if !ok {
-				return nil, errors.New("invalid content block format")
-			}
-
-			blockType, _ := blockMap["type"].(string)
-
-			switch blockType {
-			case "text":
-				if text, ok := blockMap["text"].(string); ok {
-					textContent.WriteString(text)
-				}
-
-			case "image":
-				source, ok := blockMap["source"].(map[string]any)
-				if !ok {
-					return nil, errors.New("invalid image source")
-				}
-
-				sourceType, _ := source["type"].(string)
-				if sourceType == "base64" {
-					data, _ := source["data"].(string)
-					decoded, err := base64.StdEncoding.DecodeString(data)
-					if err != nil {
-						return nil, fmt.Errorf("invalid base64 image data: %w", err)
-					}
-					images = append(images, decoded)
-				} else {
-					return nil, fmt.Errorf("invalid image source type: %s. Only base64 images are supported.", sourceType)
-				}
-				// URL images would need to be fetched - skip for now
-
-			case "tool_use":
-				id, ok := blockMap["id"].(string)
-				if !ok {
-					return nil, errors.New("tool_use block missing required 'id' field")
-				}
-				name, ok := blockMap["name"].(string)
-				if !ok {
-					return nil, errors.New("tool_use block missing required 'name' field")
-				}
-				tc := api.ToolCall{
-					ID: id,
-					Function: api.ToolCallFunction{
-						Name: name,
-					},
-				}
-				if input, ok := blockMap["input"].(map[string]any); ok {
-					tc.Function.Arguments = mapToArgs(input)
-				}
-				toolCalls = append(toolCalls, tc)
-
-			case "tool_result":
-				toolUseID, _ := blockMap["tool_use_id"].(string)
-				var resultContent string
-
-				switch c := blockMap["content"].(type) {
-				case string:
-					resultContent = c
-				case []any:
-					for _, cb := range c {
-						if cbMap, ok := cb.(map[string]any); ok {
-							if cbMap["type"] == "text" {
-								if text, ok := cbMap["text"].(string); ok {
-									resultContent += text
-								}
-							}
-						}
-					}
-				}
-
-				toolResults = append(toolResults, api.Message{
-					Role:       "tool",
-					Content:    resultContent,
-					ToolCallID: toolUseID,
-				})
-
-			case "thinking":
-				if t, ok := blockMap["thinking"].(string); ok {
-					thinking = t
-				}
-			}
-		}
-
-		if textContent.Len() > 0 || len(images) > 0 || len(toolCalls) > 0 || thinking != "" {
-			m := api.Message{
-				Role:      role,
-				Content:   textContent.String(),
-				Images:    images,
-				ToolCalls: toolCalls,
-				Thinking:  thinking,
-			}
-			messages = append(messages, m)
-		}
-
-		// Add tool results as separate messages
-		messages = append(messages, toolResults...)
-
-	default:
-		return nil, fmt.Errorf("invalid message content type: %T", content)
-	}
-
-	return messages, nil
-}
-
-// convertTool converts an Anthropic Tool to an Ollama api.Tool
-func convertTool(t Tool) (api.Tool, error) {
-	var params api.ToolFunctionParameters
-	if len(t.InputSchema) > 0 {
-		if err := json.Unmarshal(t.InputSchema, &params); err != nil {
-			return api.Tool{}, fmt.Errorf("invalid input_schema for tool %q: %w", t.Name, err)
-		}
-	}
-
-	return api.Tool{
-		Type: "function",
-		Function: api.ToolFunction{
-			Name:        t.Name,
-			Description: t.Description,
-			Parameters:  params,
-		},
-	}, nil
-}
-
-// ToMessagesResponse converts an Ollama api.ChatResponse to an Anthropic MessagesResponse
-func ToMessagesResponse(id string, r api.ChatResponse) MessagesResponse {
-	var content []ContentBlock
-
-	if r.Message.Thinking != "" {
-		content = append(content, ContentBlock{
-			Type:     "thinking",
-			Thinking: ptr(r.Message.Thinking),
-		})
-	}
-
-	if r.Message.Content != "" {
-		content = append(content, ContentBlock{
-			Type: "text",
-			Text: ptr(r.Message.Content),
-		})
-	}
-
-	for _, tc := range r.Message.ToolCalls {
-		content = append(content, ContentBlock{
-			Type:  "tool_use",
-			ID:    tc.ID,
-			Name:  tc.Function.Name,
-			Input: tc.Function.Arguments,
-		})
-	}
-
-	stopReason := mapStopReason(r.DoneReason, len(r.Message.ToolCalls) > 0)
-
-	return MessagesResponse{
-		ID:         id,
-		Type:       "message",
-		Role:       "assistant",
-		Model:      r.Model,
-		Content:    content,
-		StopReason: stopReason,
-		Usage: Usage{
-			InputTokens:  r.Metrics.PromptEvalCount,
-			OutputTokens: r.Metrics.EvalCount,
-		},
-	}
-}
-
-// mapStopReason converts Ollama done_reason to Anthropic stop_reason
-func mapStopReason(reason string, hasToolCalls bool) string {
-	if hasToolCalls {
-		return "tool_use"
-	}
-
-	switch reason {
-	case "stop":
-		return "end_turn"
-	case "length":
-		return "max_tokens"
-	default:
-		if reason != "" {
-			return "stop_sequence"
-		}
-		return ""
-	}
-}
-
-// StreamConverter manages state for converting Ollama streaming responses to Anthropic format
-type StreamConverter struct {
-	ID                   string
-	Model                string
-	firstWrite           bool
-	contentIndex         int
-	inputTokens          int
-	outputTokens         int
-	estimatedInputTokens int // Estimated tokens from request (used when actual metrics are 0)
-	thinkingStarted      bool
-	thinkingDone         bool
-	textStarted          bool
-	toolCallsSent        map[string]bool
-}
-
-func NewStreamConverter(id, model string, estimatedInputTokens int) *StreamConverter {
-	return &StreamConverter{
-		ID:                   id,
-		Model:                model,
-		firstWrite:           true,
-		estimatedInputTokens: estimatedInputTokens,
-		toolCallsSent:        make(map[string]bool),
-	}
-}
-
-// StreamEvent represents a streaming event to be sent to the client
-type StreamEvent struct {
-	Event string
-	Data  any
-}
-
-// Process converts an Ollama ChatResponse to Anthropic streaming events
-func (c *StreamConverter) Process(r api.ChatResponse) []StreamEvent {
-	var events []StreamEvent
-
-	if c.firstWrite {
-		c.firstWrite = false
-		// Use actual metrics if available, otherwise use estimate
-		c.inputTokens = r.Metrics.PromptEvalCount
-		if c.inputTokens == 0 && c.estimatedInputTokens > 0 {
-			c.inputTokens = c.estimatedInputTokens
-		}
-
-		events = append(events, StreamEvent{
-			Event: "message_start",
-			Data: MessageStartEvent{
-				Type: "message_start",
-				Message: MessagesResponse{
-					ID:      c.ID,
-					Type:    "message",
-					Role:    "assistant",
-					Model:   c.Model,
-					Content: []ContentBlock{},
-					Usage: Usage{
-						InputTokens:  c.inputTokens,
-						OutputTokens: 0,
-					},
-				},
-			},
-		})
-	}
-
-	if r.Message.Thinking != "" && !c.thinkingDone {
-		if !c.thinkingStarted {
-			c.thinkingStarted = true
-			events = append(events, StreamEvent{
-				Event: "content_block_start",
-				Data: ContentBlockStartEvent{
-					Type:  "content_block_start",
-					Index: c.contentIndex,
-					ContentBlock: ContentBlock{
-						Type:     "thinking",
-						Thinking: ptr(""),
-					},
-				},
-			})
-		}
-
-		events = append(events, StreamEvent{
-			Event: "content_block_delta",
-			Data: ContentBlockDeltaEvent{
-				Type:  "content_block_delta",
-				Index: c.contentIndex,
-				Delta: Delta{
-					Type:     "thinking_delta",
-					Thinking: r.Message.Thinking,
-				},
-			},
-		})
-	}
-
-	if r.Message.Content != "" {
-		if c.thinkingStarted && !c.thinkingDone {
-			c.thinkingDone = true
-			events = append(events, StreamEvent{
-				Event: "content_block_stop",
-				Data: ContentBlockStopEvent{
-					Type:  "content_block_stop",
-					Index: c.contentIndex,
-				},
-			})
-			c.contentIndex++
-		}
-
-		if !c.textStarted {
-			c.textStarted = true
-			events = append(events, StreamEvent{
-				Event: "content_block_start",
-				Data: ContentBlockStartEvent{
-					Type:  "content_block_start",
-					Index: c.contentIndex,
-					ContentBlock: ContentBlock{
-						Type: "text",
-						Text: ptr(""),
-					},
-				},
-			})
-		}
-
-		events = append(events, StreamEvent{
-			Event: "content_block_delta",
-			Data: ContentBlockDeltaEvent{
-				Type:  "content_block_delta",
-				Index: c.contentIndex,
-				Delta: Delta{
-					Type: "text_delta",
-					Text: r.Message.Content,
-				},
-			},
-		})
-	}
-
-	for _, tc := range r.Message.ToolCalls {
-		if c.toolCallsSent[tc.ID] {
-			continue
-		}
-
-		if c.textStarted {
-			events = append(events, StreamEvent{
-				Event: "content_block_stop",
-				Data: ContentBlockStopEvent{
-					Type:  "content_block_stop",
-					Index: c.contentIndex,
-				},
-			})
-			c.contentIndex++
-			c.textStarted = false
-		}
-
-		argsJSON, err := json.Marshal(tc.Function.Arguments)
-		if err != nil {
-			slog.Error("failed to marshal tool arguments", "error", err, "tool_id", tc.ID)
-			continue
-		}
-
-		events = append(events, StreamEvent{
-			Event: "content_block_start",
-			Data: ContentBlockStartEvent{
-				Type:  "content_block_start",
-				Index: c.contentIndex,
-				ContentBlock: ContentBlock{
-					Type:  "tool_use",
-					ID:    tc.ID,
-					Name:  tc.Function.Name,
-					Input: map[string]any{},
-				},
-			},
-		})
-
-		events = append(events, StreamEvent{
-			Event: "content_block_delta",
-			Data: ContentBlockDeltaEvent{
-				Type:  "content_block_delta",
-				Index: c.contentIndex,
-				Delta: Delta{
-					Type:        "input_json_delta",
-					PartialJSON: string(argsJSON),
-				},
-			},
-		})
-
-		events = append(events, StreamEvent{
-			Event: "content_block_stop",
-			Data: ContentBlockStopEvent{
-				Type:  "content_block_stop",
-				Index: c.contentIndex,
-			},
-		})
-
-		c.toolCallsSent[tc.ID] = true
-		c.contentIndex++
-	}
-
-	if r.Done {
-		if c.textStarted {
-			events = append(events, StreamEvent{
-				Event: "content_block_stop",
-				Data: ContentBlockStopEvent{
-					Type:  "content_block_stop",
-					Index: c.contentIndex,
-				},
-			})
-		} else if c.thinkingStarted && !c.thinkingDone {
-			events = append(events, StreamEvent{
-				Event: "content_block_stop",
-				Data: ContentBlockStopEvent{
-					Type:  "content_block_stop",
-					Index: c.contentIndex,
-				},
-			})
-		}
-
-		c.inputTokens = r.Metrics.PromptEvalCount
-		c.outputTokens = r.Metrics.EvalCount
-		stopReason := mapStopReason(r.DoneReason, len(c.toolCallsSent) > 0)
-
-		events = append(events, StreamEvent{
-			Event: "message_delta",
-			Data: MessageDeltaEvent{
-				Type: "message_delta",
-				Delta: MessageDelta{
-					StopReason: stopReason,
-				},
-				Usage: DeltaUsage{
-					InputTokens:  c.inputTokens,
-					OutputTokens: c.outputTokens,
-				},
-			},
-		})
-
-		events = append(events, StreamEvent{
-			Event: "message_stop",
-			Data: MessageStopEvent{
-				Type: "message_stop",
-			},
-		})
-	}
-
-	return events
-}
-
-// generateID generates a unique ID with the given prefix using crypto/rand
-func generateID(prefix string) string {
-	b := make([]byte, 12)
-	if _, err := rand.Read(b); err != nil {
-		// Fallback to time-based ID if crypto/rand fails
-		return fmt.Sprintf("%s_%d", prefix, time.Now().UnixNano())
-	}
-	return fmt.Sprintf("%s_%x", prefix, b)
-}
-
-// GenerateMessageID generates a unique message ID
-func GenerateMessageID() string {
-	return generateID("msg")
-}
-
-// ptr returns a pointer to the given string value
-func ptr(s string) *string {
-	return &s
-}
-
-// mapToArgs converts a map to ToolCallFunctionArguments
-func mapToArgs(m map[string]any) api.ToolCallFunctionArguments {
-	args := api.NewToolCallFunctionArguments()
-	for k, v := range m {
-		args.Set(k, v)
-	}
-	return args
-}
-
-// CountTokensRequest represents an Anthropic count_tokens request
-type CountTokensRequest struct {
-	Model    string          `json:"model"`
-	Messages []MessageParam  `json:"messages"`
-	System   any             `json:"system,omitempty"`
-	Tools    []Tool          `json:"tools,omitempty"`
-	Thinking *ThinkingConfig `json:"thinking,omitempty"`
-}
-
-// EstimateInputTokens estimates input tokens from a MessagesRequest (reuses CountTokensRequest logic)
-func EstimateInputTokens(req MessagesRequest) int {
-	return estimateTokens(CountTokensRequest{
-		Model:    req.Model,
-		Messages: req.Messages,
-		System:   req.System,
-		Tools:    req.Tools,
-		Thinking: req.Thinking,
-	})
-}
-
-// CountTokensResponse represents an Anthropic count_tokens response
-type CountTokensResponse struct {
-	InputTokens int `json:"input_tokens"`
-}
-
-// estimateTokens returns a rough estimate of tokens (len/4).
-// TODO: Replace with actual tokenization via Tokenize API for accuracy.
-// Current len/4 heuristic is a rough approximation (~4 chars/token average).
-func estimateTokens(req CountTokensRequest) int {
-	var totalLen int
-
-	// Count system prompt
-	if req.System != nil {
-		totalLen += countAnyContent(req.System)
-	}
-
-	// Count messages
-	for _, msg := range req.Messages {
-		// Count role (always present)
-		totalLen += len(msg.Role)
-		// Count content
-		contentLen := countAnyContent(msg.Content)
-		totalLen += contentLen
-	}
-
-	for _, tool := range req.Tools {
-		totalLen += len(tool.Name) + len(tool.Description) + len(tool.InputSchema)
-	}
-
-	// Return len/4 as rough token estimate, minimum 1 if there's any content
-	tokens := totalLen / 4
-	if tokens == 0 && (len(req.Messages) > 0 || req.System != nil) {
-		tokens = 1
-	}
-	return tokens
-}
-
-func countAnyContent(content any) int {
-	if content == nil {
-		return 0
-	}
-
-	switch c := content.(type) {
-	case string:
-		return len(c)
-	case []any:
-		total := 0
-		for _, block := range c {
-			total += countContentBlock(block)
-		}
-		return total
-	default:
-		if data, err := json.Marshal(content); err == nil {
-			return len(data)
-		}
-		return 0
-	}
-}
-
-func countContentBlock(block any) int {
-	blockMap, ok := block.(map[string]any)
-	if !ok {
-		if s, ok := block.(string); ok {
-			return len(s)
-		}
-		return 0
-	}
-
-	total := 0
-	blockType, _ := blockMap["type"].(string)
-
-	if text, ok := blockMap["text"].(string); ok {
-		total += len(text)
-	}
-
-	if thinking, ok := blockMap["thinking"].(string); ok {
-		total += len(thinking)
-	}
-
-	if blockType == "tool_use" {
-		if data, err := json.Marshal(blockMap); err == nil {
-			total += len(data)
-		}
-	}
-
-	if blockType == "tool_result" {
-		if data, err := json.Marshal(blockMap); err == nil {
-			total += len(data)
-		}
-	}
-
-	return total
-}
--- a/anthropic/anthropic_test.go
+++ b/anthropic/anthropic_test.go
--- a/api/client.go
+++ b/api/client.go
@@ -165,7 +165,7 @@ func (c *Client) do(ctx context.Context, method, path string, reqData, respData
 	return nil
 }

-const maxBufferSize = 8 * format.MegaByte
+const maxBufferSize = 512 * format.KiloByte

 func (c *Client) stream(ctx context.Context, method, path string, data any, fn func([]byte) error) error {
 	var buf io.Reader
@@ -226,14 +226,7 @@ func (c *Client) stream(ctx context.Context, method, path string, data any, fn f

 		bts := scanner.Bytes()
 		if err := json.Unmarshal(bts, &errorResponse); err != nil {
-			if response.StatusCode >= http.StatusBadRequest {
-				return StatusError{
-					StatusCode:   response.StatusCode,
-					Status:       response.Status,
-					ErrorMessage: string(bts),
-				}
-			}
-			return errors.New(string(bts))
+			return fmt.Errorf("unmarshal: %w", err)
 		}

 		if response.StatusCode == http.StatusUnauthorized {
@@ -347,7 +340,7 @@ type CreateProgressFunc func(ProgressResponse) error
 // Create creates a model from a [Modelfile]. fn is a progress function that
 // behaves similarly to other methods (see [Client.Pull]).
 //
-// [Modelfile]: https://github.com/ollama/ollama/blob/main/docs/modelfile.mdx
+// [Modelfile]: https://github.com/ollama/ollama/blob/main/docs/modelfile.md
 func (c *Client) Create(ctx context.Context, req *CreateRequest, fn CreateProgressFunc) error {
 	return c.stream(ctx, http.MethodPost, "/api/create", req, func(bts []byte) error {
 		var resp ProgressResponse
@@ -466,25 +459,3 @@ func (c *Client) Whoami(ctx context.Context) (*UserResponse, error) {
 	}
 	return &resp, nil
 }
-
-// AliasRequest is the request body for creating or updating a model alias.
-type AliasRequest struct {
-	Alias          string `json:"alias"`
-	Target         string `json:"target"`
-	PrefixMatching bool   `json:"prefix_matching,omitempty"`
-}
-
-// SetAliasExperimental creates or updates a model alias via the experimental aliases API.
-func (c *Client) SetAliasExperimental(ctx context.Context, req *AliasRequest) error {
-	return c.do(ctx, http.MethodPost, "/api/experimental/aliases", req, nil)
-}
-
-// AliasDeleteRequest is the request body for deleting a model alias.
-type AliasDeleteRequest struct {
-	Alias string `json:"alias"`
-}
-
-// DeleteAliasExperimental deletes a model alias via the experimental aliases API.
-func (c *Client) DeleteAliasExperimental(ctx context.Context, req *AliasDeleteRequest) error {
-	return c.do(ctx, http.MethodDelete, "/api/experimental/aliases", req, nil)
-}
--- a/api/client_test.go
+++ b/api/client_test.go
@@ -55,7 +55,6 @@ func TestClientFromEnvironment(t *testing.T) {
 type testError struct {
 	message    string
 	statusCode int
-	raw        bool // if true, write message as-is instead of JSON encoding
 }

 func (e testError) Error() string {
@@ -112,20 +111,6 @@ func TestClientStream(t *testing.T) {
 				},
 			},
 		},
-		{
-			name: "plain text error response",
-			responses: []any{
-				"internal server error",
-			},
-			wantErr: "internal server error",
-		},
-		{
-			name: "HTML error page",
-			responses: []any{
-				"<html><body>404 Not Found</body></html>",
-			},
-			wantErr: "404 Not Found",
-		},
 	}

 	for _, tc := range testCases {
@@ -150,12 +135,6 @@ func TestClientStream(t *testing.T) {
 						return
 					}

-					if str, ok := resp.(string); ok {
-						fmt.Fprintln(w, str)
-						flusher.Flush()
-						continue
-					}
-
 					if err := json.NewEncoder(w).Encode(resp); err != nil {
 						t.Fatalf("failed to encode response: %v", err)
 					}
@@ -194,10 +173,9 @@ func TestClientStream(t *testing.T) {

 func TestClientDo(t *testing.T) {
 	testCases := []struct {
-		name           string
-		response       any
-		wantErr        string
-		wantStatusCode int
+		name     string
+		response any
+		wantErr  string
 	}{
 		{
 			name: "immediate error response",
@@ -205,8 +183,7 @@ func TestClientDo(t *testing.T) {
 				message:    "test error message",
 				statusCode: http.StatusBadRequest,
 			},
-			wantErr:        "test error message",
-			wantStatusCode: http.StatusBadRequest,
+			wantErr: "test error message",
 		},
 		{
 			name: "server error response",
@@ -214,8 +191,7 @@ func TestClientDo(t *testing.T) {
 				message:    "internal error",
 				statusCode: http.StatusInternalServerError,
 			},
-			wantErr:        "internal error",
-			wantStatusCode: http.StatusInternalServerError,
+			wantErr: "internal error",
 		},
 		{
 			name: "successful response",
@@ -227,26 +203,6 @@ func TestClientDo(t *testing.T) {
 				Success: true,
 			},
 		},
-		{
-			name: "plain text error response",
-			response: testError{
-				message:    "internal server error",
-				statusCode: http.StatusInternalServerError,
-				raw:        true,
-			},
-			wantErr:        "internal server error",
-			wantStatusCode: http.StatusInternalServerError,
-		},
-		{
-			name: "HTML error page",
-			response: testError{
-				message:    "<html><body>404 Not Found</body></html>",
-				statusCode: http.StatusNotFound,
-				raw:        true,
-			},
-			wantErr:        "<html><body>404 Not Found</body></html>",
-			wantStatusCode: http.StatusNotFound,
-		},
 	}

 	for _, tc := range testCases {
@@ -254,16 +210,11 @@ func TestClientDo(t *testing.T) {
 			ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 				if errResp, ok := tc.response.(testError); ok {
 					w.WriteHeader(errResp.statusCode)
-					if !errResp.raw {
-						err := json.NewEncoder(w).Encode(map[string]string{
-							"error": errResp.message,
-						})
-						if err != nil {
-							t.Fatal("failed to encode error response:", err)
-						}
-					} else {
-						// Write raw message (simulates non-JSON error responses)
-						fmt.Fprint(w, errResp.message)
+					err := json.NewEncoder(w).Encode(map[string]string{
+						"error": errResp.message,
+					})
+					if err != nil {
+						t.Fatal("failed to encode error response:", err)
 					}
 					return
 				}
@@ -290,15 +241,6 @@ func TestClientDo(t *testing.T) {
 				if err.Error() != tc.wantErr {
 					t.Errorf("error message mismatch: got %q, want %q", err.Error(), tc.wantErr)
 				}
-				if tc.wantStatusCode != 0 {
-					if statusErr, ok := err.(StatusError); ok {
-						if statusErr.StatusCode != tc.wantStatusCode {
-							t.Errorf("status code mismatch: got %d, want %d", statusErr.StatusCode, tc.wantStatusCode)
-						}
-					} else {
-						t.Errorf("expected StatusError, got %T", err)
-					}
-				}
 				return
 			}

--- a/api/examples/chat/main.go
+++ b/api/examples/chat/main.go
@@ -15,19 +15,19 @@ func main() {
 	}

 	messages := []api.Message{
-		{
+		api.Message{
 			Role:    "system",
 			Content: "Provide very brief, concise responses",
 		},
-		{
+		api.Message{
 			Role:    "user",
 			Content: "Name some unusual animals",
 		},
-		{
+		api.Message{
 			Role:    "assistant",
 			Content: "Monotreme, platypus, echidna",
 		},
-		{
+		api.Message{
 			Role:    "user",
 			Content: "which of these is the most dangerous?",
 		},
--- a/api/types.go
+++ b/api/types.go
@@ -3,7 +3,6 @@ package api
 import (
 	"encoding/json"
 	"fmt"
-	"iter"
 	"log/slog"
 	"math"
 	"os"
@@ -15,7 +14,6 @@ import (
 	"github.com/google/uuid"

 	"github.com/ollama/ollama/envconfig"
-	"github.com/ollama/ollama/internal/orderedmap"
 	"github.com/ollama/ollama/types/model"
 )

@@ -119,28 +117,6 @@ type GenerateRequest struct {
 	// DebugRenderOnly is a debug option that, when set to true, returns the rendered
 	// template instead of calling the model.
 	DebugRenderOnly bool `json:"_debug_render_only,omitempty"`
-
-	// Logprobs specifies whether to return log probabilities of the output tokens.
-	Logprobs bool `json:"logprobs,omitempty"`
-
-	// TopLogprobs is the number of most likely tokens to return at each token position,
-	// each with an associated log probability. Only applies when Logprobs is true.
-	// Valid values are 0-20. Default is 0 (only return the selected token's logprob).
-	TopLogprobs int `json:"top_logprobs,omitempty"`
-
-	// Experimental: Image generation fields (may change or be removed)
-
-	// Width is the width of the generated image in pixels.
-	// Only used for image generation models.
-	Width int32 `json:"width,omitempty"`
-
-	// Height is the height of the generated image in pixels.
-	// Only used for image generation models.
-	Height int32 `json:"height,omitempty"`
-
-	// Steps is the number of diffusion steps for image generation.
-	// Only used for image generation models.
-	Steps int32 `json:"steps,omitempty"`
 }

 // ChatRequest describes a request sent by [Client.Chat].
@@ -183,14 +159,6 @@ type ChatRequest struct {
 	// DebugRenderOnly is a debug option that, when set to true, returns the rendered
 	// template instead of calling the model.
 	DebugRenderOnly bool `json:"_debug_render_only,omitempty"`
-
-	// Logprobs specifies whether to return log probabilities of the output tokens.
-	Logprobs bool `json:"logprobs,omitempty"`
-
-	// TopLogprobs is the number of most likely tokens to return at each token position,
-	// each with an associated log probability. Only applies when Logprobs is true.
-	// Valid values are 0-20. Default is 0 (only return the selected token's logprob).
-	TopLogprobs int `json:"top_logprobs,omitempty"`
 }

 type Tools []Tool
@@ -243,79 +211,13 @@ type ToolCallFunction struct {
 	Arguments ToolCallFunctionArguments `json:"arguments"`
 }

-// ToolCallFunctionArguments holds tool call arguments in insertion order.
-type ToolCallFunctionArguments struct {
-	om *orderedmap.Map[string, any]
-}
-
-// NewToolCallFunctionArguments creates a new empty ToolCallFunctionArguments.
-func NewToolCallFunctionArguments() ToolCallFunctionArguments {
-	return ToolCallFunctionArguments{om: orderedmap.New[string, any]()}
-}
-
-// Get retrieves a value by key.
-func (t *ToolCallFunctionArguments) Get(key string) (any, bool) {
-	if t == nil || t.om == nil {
-		return nil, false
-	}
-	return t.om.Get(key)
-}
-
-// Set sets a key-value pair, preserving insertion order.
-func (t *ToolCallFunctionArguments) Set(key string, value any) {
-	if t == nil {
-		return
-	}
-	if t.om == nil {
-		t.om = orderedmap.New[string, any]()
-	}
-	t.om.Set(key, value)
-}
-
-// Len returns the number of arguments.
-func (t *ToolCallFunctionArguments) Len() int {
-	if t == nil || t.om == nil {
-		return 0
-	}
-	return t.om.Len()
-}
-
-// All returns an iterator over all key-value pairs in insertion order.
-func (t *ToolCallFunctionArguments) All() iter.Seq2[string, any] {
-	if t == nil || t.om == nil {
-		return func(yield func(string, any) bool) {}
-	}
-	return t.om.All()
-}
-
-// ToMap returns a regular map (order not preserved).
-func (t *ToolCallFunctionArguments) ToMap() map[string]any {
-	if t == nil || t.om == nil {
-		return nil
-	}
-	return t.om.ToMap()
-}
+type ToolCallFunctionArguments map[string]any

 func (t *ToolCallFunctionArguments) String() string {
-	if t == nil || t.om == nil {
-		return "{}"
-	}
-	bts, _ := json.Marshal(t.om)
+	bts, _ := json.Marshal(t)
 	return string(bts)
 }

-func (t *ToolCallFunctionArguments) UnmarshalJSON(data []byte) error {
-	t.om = orderedmap.New[string, any]()
-	return json.Unmarshal(data, t.om)
-}
-
-func (t ToolCallFunctionArguments) MarshalJSON() ([]byte, error) {
-	if t.om == nil {
-		return []byte("{}"), nil
-	}
-	return json.Marshal(t.om)
-}
-
 type Tool struct {
 	Type     string       `json:"type"`
 	Items    any          `json:"items,omitempty"`
@@ -364,78 +266,12 @@ func (pt PropertyType) String() string {
 	return fmt.Sprintf("%v", []string(pt))
 }

-// ToolPropertiesMap holds tool properties in insertion order.
-type ToolPropertiesMap struct {
-	om *orderedmap.Map[string, ToolProperty]
-}
-
-// NewToolPropertiesMap creates a new empty ToolPropertiesMap.
-func NewToolPropertiesMap() *ToolPropertiesMap {
-	return &ToolPropertiesMap{om: orderedmap.New[string, ToolProperty]()}
-}
-
-// Get retrieves a property by name.
-func (t *ToolPropertiesMap) Get(key string) (ToolProperty, bool) {
-	if t == nil || t.om == nil {
-		return ToolProperty{}, false
-	}
-	return t.om.Get(key)
-}
-
-// Set sets a property, preserving insertion order.
-func (t *ToolPropertiesMap) Set(key string, value ToolProperty) {
-	if t == nil {
-		return
-	}
-	if t.om == nil {
-		t.om = orderedmap.New[string, ToolProperty]()
-	}
-	t.om.Set(key, value)
-}
-
-// Len returns the number of properties.
-func (t *ToolPropertiesMap) Len() int {
-	if t == nil || t.om == nil {
-		return 0
-	}
-	return t.om.Len()
-}
-
-// All returns an iterator over all properties in insertion order.
-func (t *ToolPropertiesMap) All() iter.Seq2[string, ToolProperty] {
-	if t == nil || t.om == nil {
-		return func(yield func(string, ToolProperty) bool) {}
-	}
-	return t.om.All()
-}
-
-// ToMap returns a regular map (order not preserved).
-func (t *ToolPropertiesMap) ToMap() map[string]ToolProperty {
-	if t == nil || t.om == nil {
-		return nil
-	}
-	return t.om.ToMap()
-}
-
-func (t ToolPropertiesMap) MarshalJSON() ([]byte, error) {
-	if t.om == nil {
-		return []byte("null"), nil
-	}
-	return json.Marshal(t.om)
-}
-
-func (t *ToolPropertiesMap) UnmarshalJSON(data []byte) error {
-	t.om = orderedmap.New[string, ToolProperty]()
-	return json.Unmarshal(data, t.om)
-}
-
 type ToolProperty struct {
-	AnyOf       []ToolProperty     `json:"anyOf,omitempty"`
-	Type        PropertyType       `json:"type,omitempty"`
-	Items       any                `json:"items,omitempty"`
-	Description string             `json:"description,omitempty"`
-	Enum        []any              `json:"enum,omitempty"`
-	Properties  *ToolPropertiesMap `json:"properties,omitempty"`
+	AnyOf       []ToolProperty `json:"anyOf,omitempty"`
+	Type        PropertyType   `json:"type,omitempty"`
+	Items       any            `json:"items,omitempty"`
+	Description string         `json:"description,omitempty"`
+	Enum        []any          `json:"enum,omitempty"`
 }

 // ToTypeScriptType converts a ToolProperty to a TypeScript type string
@@ -484,11 +320,11 @@ func mapToTypeScriptType(jsonType string) string {
 }

 type ToolFunctionParameters struct {
-	Type       string             `json:"type"`
-	Defs       any                `json:"$defs,omitempty"`
-	Items      any                `json:"items,omitempty"`
-	Required   []string           `json:"required,omitempty"`
-	Properties *ToolPropertiesMap `json:"properties"`
+	Type       string                  `json:"type"`
+	Defs       any                     `json:"$defs,omitempty"`
+	Items      any                     `json:"items,omitempty"`
+	Required   []string                `json:"required,omitempty"`
+	Properties map[string]ToolProperty `json:"properties"`
 }

 func (t *ToolFunctionParameters) String() string {
@@ -507,27 +343,6 @@ func (t *ToolFunction) String() string {
 	return string(bts)
 }

-// TokenLogprob represents log probability information for a single token alternative.
-type TokenLogprob struct {
-	// Token is the text representation of the token.
-	Token string `json:"token"`
-
-	// Logprob is the log probability of this token.
-	Logprob float64 `json:"logprob"`
-
-	// Bytes contains the raw byte representation of the token
-	Bytes []int `json:"bytes,omitempty"`
-}
-
-// Logprob contains log probability information for a generated token.
-type Logprob struct {
-	TokenLogprob
-
-	// TopLogprobs contains the most likely tokens and their log probabilities
-	// at this position, if requested via TopLogprobs parameter.
-	TopLogprobs []TokenLogprob `json:"top_logprobs,omitempty"`
-}
-
 // ChatResponse is the response returned by [Client.Chat]. Its fields are
 // similar to [GenerateResponse].
 type ChatResponse struct {
@@ -554,10 +369,6 @@ type ChatResponse struct {

 	DebugInfo *DebugInfo `json:"_debug_info,omitempty"`

-	// Logprobs contains log probability information for the generated tokens,
-	// if requested via the Logprobs parameter.
-	Logprobs []Logprob `json:"logprobs,omitempty"`
-
 	Metrics
 }

@@ -701,9 +512,6 @@ type CreateRequest struct {
 	Renderer string `json:"renderer,omitempty"`
 	Parser   string `json:"parser,omitempty"`

-	// Requires is the minimum version of Ollama required by the model.
-	Requires string `json:"requires,omitempty"`
-
 	// Info is a map of additional information for the model
 	Info map[string]any `json:"info,omitempty"`

@@ -749,12 +557,11 @@ type ShowResponse struct {
 	Messages      []Message          `json:"messages,omitempty"`
 	RemoteModel   string             `json:"remote_model,omitempty"`
 	RemoteHost    string             `json:"remote_host,omitempty"`
-	ModelInfo     map[string]any     `json:"model_info"`
+	ModelInfo     map[string]any     `json:"model_info,omitempty"`
 	ProjectorInfo map[string]any     `json:"projector_info,omitempty"`
 	Tensors       []Tensor           `json:"tensors,omitempty"`
 	Capabilities  []model.Capability `json:"capabilities,omitempty"`
 	ModifiedAt    time.Time          `json:"modified_at,omitempty"`
-	Requires      string             `json:"requires,omitempty"`
 }

 // CopyRequest is the request passed to [Client.Copy].
@@ -870,24 +677,6 @@ type GenerateResponse struct {
 	ToolCalls []ToolCall `json:"tool_calls,omitempty"`

 	DebugInfo *DebugInfo `json:"_debug_info,omitempty"`
-
-	// Logprobs contains log probability information for the generated tokens,
-	// if requested via the Logprobs parameter.
-	Logprobs []Logprob `json:"logprobs,omitempty"`
-
-	// Experimental: Image generation fields (may change or be removed)
-
-	// Image contains a base64-encoded generated image.
-	// Only present for image generation models.
-	Image string `json:"image,omitempty"`
-
-	// Completed is the number of completed steps in image generation.
-	// Only present for image generation models during streaming.
-	Completed int64 `json:"completed,omitempty"`
-
-	// Total is the total number of steps for image generation.
-	// Only present for image generation models during streaming.
-	Total int64 `json:"total,omitempty"`
 }

 // ModelDetails provides details about a model.
--- a/api/types_test.go
+++ b/api/types_test.go
@@ -11,24 +11,6 @@ import (
 	"github.com/stretchr/testify/require"
 )

-// testPropsMap creates a ToolPropertiesMap from a map (convenience function for tests, order not preserved)
-func testPropsMap(m map[string]ToolProperty) *ToolPropertiesMap {
-	props := NewToolPropertiesMap()
-	for k, v := range m {
-		props.Set(k, v)
-	}
-	return props
-}
-
-// testArgs creates ToolCallFunctionArguments from a map (convenience function for tests, order not preserved)
-func testArgs(m map[string]any) ToolCallFunctionArguments {
-	args := NewToolCallFunctionArguments()
-	for k, v := range m {
-		args.Set(k, v)
-	}
-	return args
-}
-
 func TestKeepAliveParsingFromJSON(t *testing.T) {
 	tests := []struct {
 		name string
@@ -327,9 +309,9 @@ func TestToolFunctionParameters_MarshalJSON(t *testing.T) {
 			input: ToolFunctionParameters{
 				Type:     "object",
 				Required: []string{"name"},
-				Properties: testPropsMap(map[string]ToolProperty{
+				Properties: map[string]ToolProperty{
 					"name": {Type: PropertyType{"string"}},
-				}),
+				},
 			},
 			expected: `{"type":"object","required":["name"],"properties":{"name":{"type":"string"}}}`,
 		},
@@ -337,9 +319,9 @@ func TestToolFunctionParameters_MarshalJSON(t *testing.T) {
 			name: "no required",
 			input: ToolFunctionParameters{
 				Type: "object",
-				Properties: testPropsMap(map[string]ToolProperty{
+				Properties: map[string]ToolProperty{
 					"name": {Type: PropertyType{"string"}},
-				}),
+				},
 			},
 			expected: `{"type":"object","properties":{"name":{"type":"string"}}}`,
 		},
@@ -357,7 +339,7 @@ func TestToolFunctionParameters_MarshalJSON(t *testing.T) {
 func TestToolCallFunction_IndexAlwaysMarshals(t *testing.T) {
 	fn := ToolCallFunction{
 		Name:      "echo",
-		Arguments: testArgs(map[string]any{"message": "hi"}),
+		Arguments: ToolCallFunctionArguments{"message": "hi"},
 	}

 	data, err := json.Marshal(fn)
@@ -522,116 +504,6 @@ func TestThinking_UnmarshalJSON(t *testing.T) {
 	}
 }

-func TestToolPropertyNestedProperties(t *testing.T) {
-	tests := []struct {
-		name     string
-		input    string
-		expected ToolProperty
-	}{
-		{
-			name: "nested object properties",
-			input: `{
-				"type": "object",
-				"description": "Location details",
-				"properties": {
-					"address": {
-						"type": "string",
-						"description": "Street address"
-					},
-					"city": {
-						"type": "string",
-						"description": "City name"
-					}
-				}
-			}`,
-			expected: ToolProperty{
-				Type:        PropertyType{"object"},
-				Description: "Location details",
-				Properties: testPropsMap(map[string]ToolProperty{
-					"address": {
-						Type:        PropertyType{"string"},
-						Description: "Street address",
-					},
-					"city": {
-						Type:        PropertyType{"string"},
-						Description: "City name",
-					},
-				}),
-			},
-		},
-		{
-			name: "deeply nested properties",
-			input: `{
-				"type": "object",
-				"description": "Event",
-				"properties": {
-					"location": {
-						"type": "object",
-						"description": "Location",
-						"properties": {
-							"coordinates": {
-								"type": "object",
-								"description": "GPS coordinates",
-								"properties": {
-									"lat": {"type": "number", "description": "Latitude"},
-									"lng": {"type": "number", "description": "Longitude"}
-								}
-							}
-						}
-					}
-				}
-			}`,
-			expected: ToolProperty{
-				Type:        PropertyType{"object"},
-				Description: "Event",
-				Properties: testPropsMap(map[string]ToolProperty{
-					"location": {
-						Type:        PropertyType{"object"},
-						Description: "Location",
-						Properties: testPropsMap(map[string]ToolProperty{
-							"coordinates": {
-								Type:        PropertyType{"object"},
-								Description: "GPS coordinates",
-								Properties: testPropsMap(map[string]ToolProperty{
-									"lat": {Type: PropertyType{"number"}, Description: "Latitude"},
-									"lng": {Type: PropertyType{"number"}, Description: "Longitude"},
-								}),
-							},
-						}),
-					},
-				}),
-			},
-		},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			var prop ToolProperty
-			err := json.Unmarshal([]byte(tt.input), &prop)
-			require.NoError(t, err)
-
-			// Compare JSON representations since pointer comparison doesn't work
-			expectedJSON, err := json.Marshal(tt.expected)
-			require.NoError(t, err)
-			actualJSON, err := json.Marshal(prop)
-			require.NoError(t, err)
-			assert.JSONEq(t, string(expectedJSON), string(actualJSON))
-
-			// Round-trip test: marshal and unmarshal again
-			data, err := json.Marshal(prop)
-			require.NoError(t, err)
-
-			var prop2 ToolProperty
-			err = json.Unmarshal(data, &prop2)
-			require.NoError(t, err)
-
-			prop2JSON, err := json.Marshal(prop2)
-			require.NoError(t, err)
-			assert.JSONEq(t, string(expectedJSON), string(prop2JSON))
-		})
-	}
-}
-
 func TestToolFunctionParameters_String(t *testing.T) {
 	tests := []struct {
 		name     string
@@ -643,12 +515,12 @@ func TestToolFunctionParameters_String(t *testing.T) {
 			params: ToolFunctionParameters{
 				Type:     "object",
 				Required: []string{"name"},
-				Properties: testPropsMap(map[string]ToolProperty{
+				Properties: map[string]ToolProperty{
 					"name": {
 						Type:        PropertyType{"string"},
 						Description: "The name of the person",
 					},
-				}),
+				},
 			},
 			expected: `{"type":"object","required":["name"],"properties":{"name":{"type":"string","description":"The name of the person"}}}`,
 		},
@@ -665,7 +537,7 @@ func TestToolFunctionParameters_String(t *testing.T) {
 					s.Self = s
 					return s
 				}(),
-				Properties: testPropsMap(map[string]ToolProperty{}),
+				Properties: map[string]ToolProperty{},
 			},
 			expected: "",
 		},
@@ -678,235 +550,3 @@ func TestToolFunctionParameters_String(t *testing.T) {
 		})
 	}
 }
-
-func TestToolCallFunctionArguments_OrderPreservation(t *testing.T) {
-	t.Run("marshal preserves insertion order", func(t *testing.T) {
-		args := NewToolCallFunctionArguments()
-		args.Set("zebra", "z")
-		args.Set("apple", "a")
-		args.Set("mango", "m")
-
-		data, err := json.Marshal(args)
-		require.NoError(t, err)
-
-		// Should preserve insertion order, not alphabetical
-		assert.Equal(t, `{"zebra":"z","apple":"a","mango":"m"}`, string(data))
-	})
-
-	t.Run("unmarshal preserves JSON order", func(t *testing.T) {
-		jsonData := `{"zebra":"z","apple":"a","mango":"m"}`
-
-		var args ToolCallFunctionArguments
-		err := json.Unmarshal([]byte(jsonData), &args)
-		require.NoError(t, err)
-
-		// Verify iteration order matches JSON order
-		var keys []string
-		for k := range args.All() {
-			keys = append(keys, k)
-		}
-		assert.Equal(t, []string{"zebra", "apple", "mango"}, keys)
-	})
-
-	t.Run("round trip preserves order", func(t *testing.T) {
-		original := `{"z":1,"a":2,"m":3,"b":4}`
-
-		var args ToolCallFunctionArguments
-		err := json.Unmarshal([]byte(original), &args)
-		require.NoError(t, err)
-
-		data, err := json.Marshal(args)
-		require.NoError(t, err)
-
-		assert.Equal(t, original, string(data))
-	})
-
-	t.Run("String method returns ordered JSON", func(t *testing.T) {
-		args := NewToolCallFunctionArguments()
-		args.Set("c", 3)
-		args.Set("a", 1)
-		args.Set("b", 2)
-
-		assert.Equal(t, `{"c":3,"a":1,"b":2}`, args.String())
-	})
-
-	t.Run("Get retrieves correct values", func(t *testing.T) {
-		args := NewToolCallFunctionArguments()
-		args.Set("key1", "value1")
-		args.Set("key2", 42)
-
-		v, ok := args.Get("key1")
-		assert.True(t, ok)
-		assert.Equal(t, "value1", v)
-
-		v, ok = args.Get("key2")
-		assert.True(t, ok)
-		assert.Equal(t, 42, v)
-
-		_, ok = args.Get("nonexistent")
-		assert.False(t, ok)
-	})
-
-	t.Run("Len returns correct count", func(t *testing.T) {
-		args := NewToolCallFunctionArguments()
-		assert.Equal(t, 0, args.Len())
-
-		args.Set("a", 1)
-		assert.Equal(t, 1, args.Len())
-
-		args.Set("b", 2)
-		assert.Equal(t, 2, args.Len())
-	})
-
-	t.Run("empty args marshal to empty object", func(t *testing.T) {
-		args := NewToolCallFunctionArguments()
-		data, err := json.Marshal(args)
-		require.NoError(t, err)
-		assert.Equal(t, `{}`, string(data))
-	})
-
-	t.Run("zero value args marshal to empty object", func(t *testing.T) {
-		var args ToolCallFunctionArguments
-		assert.Equal(t, "{}", args.String())
-	})
-}
-
-func TestToolPropertiesMap_OrderPreservation(t *testing.T) {
-	t.Run("marshal preserves insertion order", func(t *testing.T) {
-		props := NewToolPropertiesMap()
-		props.Set("zebra", ToolProperty{Type: PropertyType{"string"}})
-		props.Set("apple", ToolProperty{Type: PropertyType{"number"}})
-		props.Set("mango", ToolProperty{Type: PropertyType{"boolean"}})
-
-		data, err := json.Marshal(props)
-		require.NoError(t, err)
-
-		// Should preserve insertion order, not alphabetical
-		expected := `{"zebra":{"type":"string"},"apple":{"type":"number"},"mango":{"type":"boolean"}}`
-		assert.Equal(t, expected, string(data))
-	})
-
-	t.Run("unmarshal preserves JSON order", func(t *testing.T) {
-		jsonData := `{"zebra":{"type":"string"},"apple":{"type":"number"},"mango":{"type":"boolean"}}`
-
-		var props ToolPropertiesMap
-		err := json.Unmarshal([]byte(jsonData), &props)
-		require.NoError(t, err)
-
-		// Verify iteration order matches JSON order
-		var keys []string
-		for k := range props.All() {
-			keys = append(keys, k)
-		}
-		assert.Equal(t, []string{"zebra", "apple", "mango"}, keys)
-	})
-
-	t.Run("round trip preserves order", func(t *testing.T) {
-		original := `{"z":{"type":"string"},"a":{"type":"number"},"m":{"type":"boolean"}}`
-
-		var props ToolPropertiesMap
-		err := json.Unmarshal([]byte(original), &props)
-		require.NoError(t, err)
-
-		data, err := json.Marshal(props)
-		require.NoError(t, err)
-
-		assert.Equal(t, original, string(data))
-	})
-
-	t.Run("Get retrieves correct values", func(t *testing.T) {
-		props := NewToolPropertiesMap()
-		props.Set("name", ToolProperty{Type: PropertyType{"string"}, Description: "The name"})
-		props.Set("age", ToolProperty{Type: PropertyType{"integer"}, Description: "The age"})
-
-		v, ok := props.Get("name")
-		assert.True(t, ok)
-		assert.Equal(t, "The name", v.Description)
-
-		v, ok = props.Get("age")
-		assert.True(t, ok)
-		assert.Equal(t, "The age", v.Description)
-
-		_, ok = props.Get("nonexistent")
-		assert.False(t, ok)
-	})
-
-	t.Run("Len returns correct count", func(t *testing.T) {
-		props := NewToolPropertiesMap()
-		assert.Equal(t, 0, props.Len())
-
-		props.Set("a", ToolProperty{})
-		assert.Equal(t, 1, props.Len())
-
-		props.Set("b", ToolProperty{})
-		assert.Equal(t, 2, props.Len())
-	})
-
-	t.Run("nil props marshal to null", func(t *testing.T) {
-		var props *ToolPropertiesMap
-		data, err := json.Marshal(props)
-		require.NoError(t, err)
-		assert.Equal(t, `null`, string(data))
-	})
-
-	t.Run("ToMap returns regular map", func(t *testing.T) {
-		props := NewToolPropertiesMap()
-		props.Set("a", ToolProperty{Type: PropertyType{"string"}})
-		props.Set("b", ToolProperty{Type: PropertyType{"number"}})
-
-		m := props.ToMap()
-		assert.Equal(t, 2, len(m))
-		assert.Equal(t, PropertyType{"string"}, m["a"].Type)
-		assert.Equal(t, PropertyType{"number"}, m["b"].Type)
-	})
-}
-
-func TestToolCallFunctionArguments_ComplexValues(t *testing.T) {
-	t.Run("nested objects preserve order", func(t *testing.T) {
-		jsonData := `{"outer":{"z":1,"a":2},"simple":"value"}`
-
-		var args ToolCallFunctionArguments
-		err := json.Unmarshal([]byte(jsonData), &args)
-		require.NoError(t, err)
-
-		// Outer keys should be in order
-		var keys []string
-		for k := range args.All() {
-			keys = append(keys, k)
-		}
-		assert.Equal(t, []string{"outer", "simple"}, keys)
-	})
-
-	t.Run("arrays as values", func(t *testing.T) {
-		args := NewToolCallFunctionArguments()
-		args.Set("items", []string{"a", "b", "c"})
-		args.Set("numbers", []int{1, 2, 3})
-
-		data, err := json.Marshal(args)
-		require.NoError(t, err)
-
-		assert.Equal(t, `{"items":["a","b","c"],"numbers":[1,2,3]}`, string(data))
-	})
-}
-
-func TestToolPropertiesMap_NestedProperties(t *testing.T) {
-	t.Run("nested properties preserve order", func(t *testing.T) {
-		props := NewToolPropertiesMap()
-
-		nestedProps := NewToolPropertiesMap()
-		nestedProps.Set("z_field", ToolProperty{Type: PropertyType{"string"}})
-		nestedProps.Set("a_field", ToolProperty{Type: PropertyType{"number"}})
-
-		props.Set("outer", ToolProperty{
-			Type:       PropertyType{"object"},
-			Properties: nestedProps,
-		})
-
-		data, err := json.Marshal(props)
-		require.NoError(t, err)
-
-		// Both outer and inner should preserve order
-		expected := `{"outer":{"type":"object","properties":{"z_field":{"type":"string"},"a_field":{"type":"number"}}}}`
-		assert.Equal(t, expected, string(data))
-	})
-}
--- a/app/README.md
+++ b/app/README.md
@@ -48,6 +48,16 @@ The `-dev` flag enables:
 - CORS headers for cross-origin requests
 - Hot-reload support for UI development

+#### Run Storybook
+
+Inside the `ui/app` directory, run:
+
+```bash
+npm run storybook
+```
+
+For now we're writing stories as siblings of the component they're testing. So for example, `src/components/Message.stories.tsx` is the story for `src/components/Message.tsx`.
+
 ## Build


@@ -75,9 +85,9 @@ The `-dev` flag enables:
 CI builds with Xcode 14.1 for OS compatibility prior to v13.  If you want to manually build v11+ support, you can download the older Xcode [here](https://developer.apple.com/services-account/download?path=/Developer_Tools/Xcode_14.1/Xcode_14.1.xip), extract, then `mv ./Xcode.app /Applications/Xcode_14.1.0.app` then activate with:

 ```
-export CGO_CFLAGS="-O3 -mmacosx-version-min=12.0"
-export CGO_CXXFLAGS="-O3 -mmacosx-version-min=12.0"
-export CGO_LDFLAGS="-mmacosx-version-min=12.0"
+export CGO_CFLAGS=-mmacosx-version-min=12.0
+export CGO_CXXFLAGS=-mmacosx-version-min=12.0
+export CGO_LDFLAGS=-mmacosx-version-min=12.0
 export SDKROOT=/Applications/Xcode_14.1.0.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk
 export DEVELOPER_DIR=/Applications/Xcode_14.1.0.app/Contents/Developer
 ```
--- a/app/cmd/app/app.go
+++ b/app/cmd/app/app.go
@@ -273,6 +273,10 @@ func main() {
 		Handler: uiServer.Handler(),
 	}

+	if _, err := uiServer.UserData(ctx); err != nil {
+		slog.Warn("failed to load user data", "error", err)
+	}
+
 	// Start the UI server
 	slog.Info("starting ui server", "port", port)
 	go func() {
@@ -316,17 +320,6 @@ func main() {
 		slog.Debug("no URL scheme request to handle")
 	}

-	go func() {
-		slog.Debug("waiting for ollama server to be ready")
-		if err := ui.WaitForServer(ctx, 10*time.Second); err != nil {
-			slog.Warn("ollama server not ready, continuing anyway", "error", err)
-		}
-
-		if _, err := uiServer.UserData(ctx); err != nil {
-			slog.Warn("failed to load user data", "error", err)
-		}
-	}()
-
 	osRun(cancel, hasCompletedFirstRun, startHidden)

 	slog.Info("shutting down desktop server")
@@ -368,7 +361,7 @@ func checkUserLoggedIn(uiServerPort int) bool {
 		return false
 	}

-	resp, err := http.Post(fmt.Sprintf("http://127.0.0.1:%d/api/me", uiServerPort), "application/json", nil)
+	resp, err := http.Get(fmt.Sprintf("http://127.0.0.1:%d/api/v1/me", uiServerPort))
 	if err != nil {
 		slog.Debug("failed to call local auth endpoint", "error", err)
 		return false
@@ -404,8 +397,8 @@ func checkUserLoggedIn(uiServerPort int) bool {
 // handleConnectURLScheme fetches the connect URL and opens it in the browser
 func handleConnectURLScheme() {
 	if checkUserLoggedIn(uiServerPort) {
-		slog.Info("user is already logged in, opening app instead")
-		showWindow(wv.webview.Window())
+		slog.Info("user is already logged in, opening settings instead")
+		sendUIRequestMessage("/")
 		return
 	}

@@ -441,30 +434,37 @@ func openInBrowser(url string) {
 	}
 }

-// parseURLScheme parses an ollama:// URL and validates it
-// Supports: ollama:// (open app) and ollama://connect (OAuth)
-func parseURLScheme(urlSchemeRequest string) (isConnect bool, err error) {
+// parseURLScheme parses an ollama:// URL and returns whether it's a connect URL and the UI path
+func parseURLScheme(urlSchemeRequest string) (isConnect bool, uiPath string, err error) {
 	parsedURL, err := url.Parse(urlSchemeRequest)
 	if err != nil {
-		return false, fmt.Errorf("invalid URL: %w", err)
+		return false, "", err
 	}

 	// Check if this is a connect URL
 	if parsedURL.Host == "connect" || strings.TrimPrefix(parsedURL.Path, "/") == "connect" {
-		return true, nil
+		return true, "", nil
 	}

-	// Allow bare ollama:// or ollama:/// to open the app
-	if (parsedURL.Host == "" && parsedURL.Path == "") || parsedURL.Path == "/" {
-		return false, nil
+	// Extract the UI path
+	path := "/"
+	if parsedURL.Path != "" && parsedURL.Path != "/" {
+		// For URLs like ollama:///settings, use the path directly
+		path = parsedURL.Path
+	} else if parsedURL.Host != "" {
+		// For URLs like ollama://settings (without triple slash),
+		// the "settings" part is parsed as the host, not the path.
+		// We need to convert it to a path by prepending "/"
+		// This also handles ollama://settings/ where Windows adds a trailing slash
+		path = "/" + parsedURL.Host
 	}

-	return false, fmt.Errorf("unsupported ollama:// URL path: %s", urlSchemeRequest)
+	return false, path, nil
 }

 // handleURLSchemeInCurrentInstance processes URL scheme requests in the current instance
 func handleURLSchemeInCurrentInstance(urlSchemeRequest string) {
-	isConnect, err := parseURLScheme(urlSchemeRequest)
+	isConnect, uiPath, err := parseURLScheme(urlSchemeRequest)
 	if err != nil {
 		slog.Error("failed to parse URL scheme request", "url", urlSchemeRequest, "error", err)
 		return
@@ -473,8 +473,6 @@ func handleURLSchemeInCurrentInstance(urlSchemeRequest string) {
 	if isConnect {
 		handleConnectURLScheme()
 	} else {
-		if wv.webview != nil {
-			showWindow(wv.webview.Window())
-		}
+		sendUIRequestMessage(uiPath)
 	}
 }
--- a/app/cmd/app/app_darwin.go
+++ b/app/cmd/app/app_darwin.go
@@ -191,6 +191,13 @@ func LaunchNewApp() {
 	C.launchApp(appName)
 }

+// Send a request to the main app thread to load a UI page
+func sendUIRequestMessage(path string) {
+	p := C.CString(path)
+	defer C.free(unsafe.Pointer(p))
+	C.uiRequest(p)
+}
+
 func registerLaunchAgent(hasCompletedFirstRun bool) {
 	// Remove any stale Login Item registrations
 	C.unregisterSelfFromLoginItem()
--- a/app/cmd/app/app_darwin.m
+++ b/app/cmd/app/app_darwin.m
@@ -14,7 +14,6 @@ extern NSString *SystemWidePath;
@interface AppDelegate () <NSWindowDelegate, WKNavigationDelegate, WKUIDelegate>
@property(strong, nonatomic) NSStatusItem *statusItem;
@property(assign, nonatomic) BOOL updateAvailable;
-@property(assign, nonatomic) BOOL systemShutdownInProgress;
@end

@implementation AppDelegate
@@ -25,14 +24,27 @@ bool firstTimeRun,startHidden; // Set in run before initialization
    for (NSURL *url in urls) {
        if ([url.scheme isEqualToString:@"ollama"]) {
            NSString *path = url.path;
-
-            if (path && ([path isEqualToString:@"/connect"] || [url.host isEqualToString:@"connect"])) {
+            if (!path || [path isEqualToString:@""]) {
+                // For URLs like ollama://settings (without triple slash),
+                // the "settings" part is parsed as the host, not the path.
+                // We need to convert it to a path by prepending "/"
+                if (url.host && ![url.host isEqualToString:@""]) {
+                    path = [@"/" stringByAppendingString:url.host];
+                } else {
+                    path = @"/";
+                }
+            }
+            
+            if ([path isEqualToString:@"/connect"] || [url.host isEqualToString:@"connect"]) {
                // Special case: handle connect by opening browser instead of app
                handleConnectURL();
            } else {
                // Set app to be active and visible
                [NSApp setActivationPolicy:NSApplicationActivationPolicyRegular];
                [NSApp activateIgnoringOtherApps:YES];
+                
+                // Open the path with the UI
+                [self uiRequest:path];
            }
            
            break;
@@ -41,13 +53,6 @@ bool firstTimeRun,startHidden; // Set in run before initialization
 }

 - (void)applicationDidFinishLaunching:(NSNotification *)aNotification {
-    // Register for system shutdown/restart notification so we can allow termination
-    [[[NSWorkspace sharedWorkspace] notificationCenter]
-        addObserver:self
-           selector:@selector(systemWillPowerOff:)
-               name:NSWorkspaceWillPowerOffNotification
-             object:nil];
-
    // if we're in development mode, set the app icon
    NSString *bundlePath = [[NSBundle mainBundle] bundlePath];
    if (![bundlePath hasSuffix:@".app"]) {
@@ -255,7 +260,7 @@ bool firstTimeRun,startHidden; // Set in run before initialization
 }

 - (void)openHelp:(id)sender {
-    NSURL *url = [NSURL URLWithString:@"https://docs.ollama.com/"];
+    NSURL *url = [NSURL URLWithString:@"https://github.com/ollama/ollama/tree/main/docs"];
    [[NSWorkspace sharedWorkspace] openURL:url];
 }

@@ -286,18 +291,7 @@ bool firstTimeRun,startHidden; // Set in run before initialization
    [NSApp activateIgnoringOtherApps:YES];
 }

- (void)systemWillPowerOff:(NSNotification *)notification {
-    // Set flag so applicationShouldTerminate: knows to allow termination.
-    // The system will call applicationShouldTerminate: after posting this notification.
-    self.systemShutdownInProgress = YES;
-}
-
 - (NSApplicationTerminateReply)applicationShouldTerminate:(NSApplication *)sender {
-    // Allow termination if the system is shutting down or restarting
-    if (self.systemShutdownInProgress) {
-        return NSTerminateNow;
-    }
-    // Otherwise just hide the app (for Cmd+Q, close button, etc.)
    [NSApp hide:nil];
    [NSApp setActivationPolicy:NSApplicationActivationPolicyAccessory];
    return NSTerminateCancel;
--- a/app/cmd/app/app_windows.go
+++ b/app/cmd/app/app_windows.go
@@ -138,7 +138,7 @@ func (app *appCallbacks) HandleURLScheme(urlScheme string) {

 // handleURLSchemeRequest processes URL scheme requests from other instances
 func handleURLSchemeRequest(urlScheme string) {
-	isConnect, err := parseURLScheme(urlScheme)
+	isConnect, uiPath, err := parseURLScheme(urlScheme)
 	if err != nil {
 		slog.Error("failed to parse URL scheme request", "url", urlScheme, "error", err)
 		return
@@ -147,9 +147,7 @@ func handleURLSchemeRequest(urlScheme string) {
 	if isConnect {
 		handleConnectURLScheme()
 	} else {
-		if wv.webview != nil {
-			showWindow(wv.webview.Window())
-		}
+		sendUIRequestMessage(uiPath)
 	}
 }

@@ -263,6 +261,11 @@ func createLoginShortcut() error {
 	return nil
 }

+// Send a request to the main app thread to load a UI page
+func sendUIRequestMessage(path string) {
+	wintray.SendUIRequestMessage(path)
+}
+
 func LaunchNewApp() {
 }

--- a/app/cmd/app/webview.go
+++ b/app/cmd/app/webview.go
@@ -282,7 +282,7 @@ func (w *Webview) Run(path string) unsafe.Pointer {
 					"go", "rs", "swift", "kt", "scala", "sh", "bat", "yaml", "yml", "toml", "ini",
 					"cfg", "conf", "log", "rtf",
 				}
-				imageExts := []string{"png", "jpg", "jpeg", "webp"}
+				imageExts := []string{"png", "jpg", "jpeg"}
 				allowedExts := append(textExts, imageExts...)

 				// Use native multiple file selection with extension filtering
--- a/app/dialog/cocoa/dlg.m
+++ b/app/dialog/cocoa/dlg.m
@@ -169,47 +169,37 @@ DlgResult fileDlg(FileDlgParams* params) {
 	}
 	
 	NSArray* urls = [panel URLs];
-	if([urls count] == 0) {
-		return DLG_CANCEL;
-	}
-	
-	if(self->params->allowMultiple) {
+	if(self->params->allowMultiple && [urls count] >= 1) {
 		// For multiple files, we need to return all paths separated by null bytes
 		char* bufPtr = self->params->buf;
 		int remainingBuf = self->params->nbuf;
 		
-		// Calculate total required buffer size first
-		int totalSize = 0;
-		for(NSURL* url in urls) {
-			char tempBuf[PATH_MAX];
-			if(![url getFileSystemRepresentation:tempBuf maxLength:PATH_MAX]) {
-				return DLG_URLFAIL;
-			}
-			totalSize += strlen(tempBuf) + 1; // +1 for null terminator
-		}
-		totalSize += 1; // Final null terminator
+  // Calculate total required buffer size first
+  int totalSize = 0;
+  for(NSURL* url in urls) {
+      char tempBuf[PATH_MAX];
+      if(![url getFileSystemRepresentation:tempBuf maxLength:PATH_MAX]) {
+          return DLG_URLFAIL;
+      }
+      totalSize += strlen(tempBuf) + 1; // +1 for null terminator
+  }
+  totalSize += 1; // Final null terminator

-		if(totalSize > self->params->nbuf) {
-			// Not enough buffer space
-			return DLG_URLFAIL;
-		}
+  if(totalSize > self->params->nbuf) {
+      // Not enough buffer space
+      return DLG_URLFAIL;
+  }

-		// Now actually copy the paths (we know we have space)
-		bufPtr = self->params->buf;
-		for(NSURL* url in urls) {
-			char tempBuf[PATH_MAX];
-			[url getFileSystemRepresentation:tempBuf maxLength:PATH_MAX];
-			int pathLen = strlen(tempBuf);
-			strcpy(bufPtr, tempBuf);
-			bufPtr += pathLen + 1;
-		}
-		*bufPtr = '\0'; // Final null terminator
-	} else {
-		// Single file/directory selection - write path to buffer
-		NSURL* url = [urls firstObject];
-		if(![url getFileSystemRepresentation:self->params->buf maxLength:self->params->nbuf]) {
-			return DLG_URLFAIL;
-		}
+  // Now actually copy the paths (we know we have space)
+  bufPtr = self->params->buf;
+  for(NSURL* url in urls) {
+      char tempBuf[PATH_MAX];
+      [url getFileSystemRepresentation:tempBuf maxLength:PATH_MAX];
+      int pathLen = strlen(tempBuf);
+      strcpy(bufPtr, tempBuf);
+      bufPtr += pathLen + 1;
+  }
+  *bufPtr = '\0'; // Final null terminator
 	}
 	
 	return DLG_OK;
--- a/app/dialog/dlgs_windows.go
+++ b/app/dialog/dlgs_windows.go
@@ -15,7 +15,7 @@ const multiFileBufferSize = w32.MAX_PATH * 10
 type WinDlgError int

 func (e WinDlgError) Error() string {
-	return fmt.Sprintf("CommDlgExtendedError: %#x", int(e))
+	return fmt.Sprintf("CommDlgExtendedError: %#x", e)
 }

 func err() error {
--- a/app/server/server.go
+++ b/app/server/server.go
@@ -224,7 +224,9 @@ func (s *Server) cmd(ctx context.Context) (*exec.Cmd, error) {
 		if _, err := os.Stat(settings.Models); err == nil {
 			env["OLLAMA_MODELS"] = settings.Models
 		} else {
-			slog.Warn("models path not accessible, using default", "path", settings.Models, "err", err)
+			slog.Warn("models path not accessible, clearing models setting", "path", settings.Models, "err", err)
+			settings.Models = ""
+			s.store.SetSettings(settings)
 		}
 	}
 	if settings.ContextLength > 0 {
--- a/app/ui/app/codegen/gotypes.gen.ts
+++ b/app/ui/app/codegen/gotypes.gen.ts
@@ -469,24 +469,26 @@ export class HealthResponse {
 }
 export class User {
    id: string;
-    email: string;
    name: string;
-    bio?: string;
-    avatarurl?: string;
-    firstname?: string;
-    lastname?: string;
-    plan?: string;
+    email: string;
+    avatarURL: string;
+    plan: string;
+    bio: string;
+    firstName: string;
+    lastName: string;
+    overThreshold: boolean;

    constructor(source: any = {}) {
        if ('string' === typeof source) source = JSON.parse(source);
        this.id = source["id"];
-        this.email = source["email"];
        this.name = source["name"];
-        this.bio = source["bio"];
-        this.avatarurl = source["avatarurl"];
-        this.firstname = source["firstname"];
-        this.lastname = source["lastname"];
+        this.email = source["email"];
+        this.avatarURL = source["avatarURL"];
        this.plan = source["plan"];
+        this.bio = source["bio"];
+        this.firstName = source["firstName"];
+        this.lastName = source["lastName"];
+        this.overThreshold = source["overThreshold"];
    }
 }
 export class Attachment {
--- a/app/ui/app/package-lock.json
+++ b/app/ui/app/package-lock.json
--- a/app/ui/app/package.json
+++ b/app/ui/app/package.json
@@ -34,7 +34,6 @@
    "rehype-raw": "^7.0.0",
    "rehype-sanitize": "^6.0.0",
    "remark-math": "^6.0.0",
-    "streamdown": "^1.4.0",
    "unist-builder": "^4.0.0",
    "unist-util-parents": "^3.0.0"
  },
--- a/app/ui/app/src/api.ts
+++ b/app/ui/app/src/api.ts
@@ -15,7 +15,6 @@ import {
 import { parseJsonlFromResponse } from "./util/jsonl-parsing";
 import { ollamaClient as ollama } from "./lib/ollama-client";
 import type { ModelResponse } from "ollama/browser";
-import { API_BASE, OLLAMA_DOT_COM } from "./lib/config";

 // Extend Model class with utility methods
 declare module "@/gotypes" {
@@ -27,6 +26,9 @@ declare module "@/gotypes" {
 Model.prototype.isCloud = function (): boolean {
  return this.model.endsWith("cloud");
 };
+
+const API_BASE = import.meta.env.DEV ? "http://127.0.0.1:3001" : "";
+
 // Helper function to convert Uint8Array to base64
 function uint8ArrayToBase64(uint8Array: Uint8Array): string {
  const chunkSize = 0x8000; // 32KB chunks to avoid stack overflow
@@ -41,50 +43,44 @@ function uint8ArrayToBase64(uint8Array: Uint8Array): string {
 }

 export async function fetchUser(): Promise<User | null> {
-  const response = await fetch(`${API_BASE}/api/me`, {
-    method: "POST",
-    headers: {
-      "Content-Type": "application/json",
-    },
-  });
+  try {
+    const response = await fetch(`${API_BASE}/api/v1/me`, {
+      method: "GET",
+      headers: {
+        "Content-Type": "application/json",
+      },
+    });

-  if (response.ok) {
-    const userData: User = await response.json();
-
-    if (userData.avatarurl && !userData.avatarurl.startsWith("http")) {
-      userData.avatarurl = `${OLLAMA_DOT_COM}${userData.avatarurl}`;
+    if (response.ok) {
+      const userData: User = await response.json();
+      return userData;
    }

-    return userData;
-  }
-
-  if (response.status === 401 || response.status === 403) {
+    return null;
+  } catch (error) {
+    console.error("Error fetching user:", error);
    return null;
  }
-
-  throw new Error(`Failed to fetch user: ${response.status}`);
 }

 export async function fetchConnectUrl(): Promise<string> {
-  const response = await fetch(`${API_BASE}/api/me`, {
-    method: "POST",
+  const response = await fetch(`${API_BASE}/api/v1/connect`, {
+    method: "GET",
    headers: {
      "Content-Type": "application/json",
    },
  });

-  if (response.status === 401) {
-    const data = await response.json();
-    if (data.signin_url) {
-      return data.signin_url;
-    }
+  if (!response.ok) {
+    throw new Error("Failed to fetch connect URL");
  }

-  throw new Error("Failed to fetch connect URL");
+  const data = await response.json();
+  return data.connect_url;
 }

 export async function disconnectUser(): Promise<void> {
-  const response = await fetch(`${API_BASE}/api/signout`, {
+  const response = await fetch(`${API_BASE}/api/v1/disconnect`, {
    method: "POST",
    headers: {
      "Content-Type": "application/json",
@@ -209,11 +205,6 @@ export async function* sendMessage(
    data: uint8ArrayToBase64(att.data),
  }));

-  // Send think parameter when it's explicitly set (true, false, or a non-empty string).
-  const shouldSendThink =
-    think !== undefined &&
-    (typeof think === "boolean" || (typeof think === "string" && think !== ""));
-
  const response = await fetch(`${API_BASE}/api/v1/chat/${chatId}`, {
    method: "POST",
    headers: {
@@ -231,7 +222,7 @@ export async function* sendMessage(
        web_search: webSearch ?? false,
        file_tools: fileTools ?? false,
        ...(forceUpdate !== undefined ? { forceUpdate } : {}),
-        ...(shouldSendThink ? { think } : {}),
+        ...(think !== undefined ? { think } : {}),
      }),
    ),
    signal,
@@ -394,8 +385,7 @@ export async function getInferenceCompute(): Promise<InferenceCompute[]> {

 export async function fetchHealth(): Promise<boolean> {
  try {
-    // Use the /api/version endpoint as a health check
-    const response = await fetch(`${API_BASE}/api/version`, {
+    const response = await fetch(`${API_BASE}/api/v1/health`, {
      method: "GET",
      headers: {
        "Content-Type": "application/json",
@@ -404,8 +394,7 @@ export async function fetchHealth(): Promise<boolean> {

    if (response.ok) {
      const data = await response.json();
-      // If we get a version back, the server is healthy
-      return !!data.version;
+      return data.healthy || false;
    }

    return false;
--- a/app/ui/app/src/components/Settings.tsx
+++ b/app/ui/app/src/components/Settings.tsx
@@ -299,9 +299,9 @@ export default function Settings() {
                        </Button>
                      </div>
                    </div>
-                    {user?.avatarurl && (
+                    {user?.avatarURL && (
                      <img
-                        src={user.avatarurl}
+                        src={user.avatarURL}
                        alt={user?.name}
                        className="h-10 w-10 rounded-full bg-neutral-200 dark:bg-neutral-700 flex-shrink-0"
                        onError={(e) => {
--- a/app/ui/app/src/components/StreamingMarkdownContent.stories.tsx
+++ b/app/ui/app/src/components/StreamingMarkdownContent.stories.tsx
--- a/app/ui/app/src/components/StreamingMarkdownContent.test.tsx
+++ b/app/ui/app/src/components/StreamingMarkdownContent.test.tsx
@@ -0,0 +1,522 @@
+import { expect, test, suite } from "vitest";
+import { processStreamingMarkdown } from "@/utils/processStreamingMarkdown";
+
+suite("common llm outputs that cause issues", () => {
+  test("prefix of bolded list item shouldn't make a horizontal line", () => {
+    // we're going to go in order of incrementally adding characters. This
+    // happens really commonly with LLMs that like to make lists like so:
+    //
+    // * **point 1**: explanatory text
+    // * **point 2**: more explanatory text
+    //
+    // Partial rendering of `*` (A), followed by `* *` (B), followed by `* **`
+    // (C) is a total mess.  (A) renders as a single bullet point in an
+    // otherwise empty list, (B) renders as two nested lists (and therefore
+    // two bullet points, styled differently by default in html), and (C)
+    // renders as a horizontal line because in markdown apparently `***` or `*
+    // * *` horizontal rules don't have as strict whitespace rules as I
+    // expected them to
+
+    // these are alone (i.e., they would be the first list item)
+    expect(processStreamingMarkdown("*")).toBe("");
+    expect(processStreamingMarkdown("* *")).toBe("");
+    expect(processStreamingMarkdown("* **")).toBe("");
+    // expect(processStreamingMarkdown("* **b")).toBe("* **b**");
+
+    // with a list item before them
+    expect(
+      processStreamingMarkdown(
+        // prettier-ignore
+        [
+          "* abc", 
+          "*"
+        ].join("\n"),
+      ),
+    ).toBe("* abc");
+
+    expect(
+      processStreamingMarkdown(
+        // prettier-ignore
+        [
+          "* abc", 
+          "* *"
+        ].join("\n"),
+      ),
+    ).toBe("* abc");
+
+    expect(
+      processStreamingMarkdown(
+        // prettier-ignore
+        [
+          "* abc", 
+          "* **"
+        ].join("\n"),
+      ),
+    ).toBe("* abc");
+  });
+
+  test("bolded list items with text should be rendered properly", () => {
+    expect(processStreamingMarkdown("* **abc**")).toBe("* **abc**");
+  });
+
+  test("partially bolded list items should be autoclosed", () => {
+    expect(processStreamingMarkdown("* **abc")).toBe("* **abc**");
+  });
+
+  suite(
+    "partially bolded list items should be autoclosed, even if the last node isn't a text node",
+    () => {
+      test("inline code", () => {
+        expect(
+          processStreamingMarkdown("* **Asynchronous Function `async`*"),
+        ).toBe("* **Asynchronous Function `async`**");
+      });
+    },
+  );
+});
+
+suite("autoclosing bold", () => {
+  suite("endings with no asterisks", () => {
+    test("should autoclose bold", () => {
+      expect(processStreamingMarkdown("**abc")).toBe("**abc**");
+      expect(processStreamingMarkdown("abc **abc")).toBe("abc **abc**");
+    });
+
+    suite("should autoclose, even if the last node isn't a text node", () => {
+      test("inline code", () => {
+        expect(
+          processStreamingMarkdown("* **Asynchronous Function `async`"),
+        ).toBe("* **Asynchronous Function `async`**");
+      });
+
+      test("opening ** is at the end of the text", () => {
+        expect(processStreamingMarkdown("abc **`def` jhk [lmn](opq)")).toBe(
+          "abc **`def` jhk [lmn](opq)**",
+        );
+      });
+
+      test("if there's a space after the **, it should NOT be autoclosed", () => {
+        expect(processStreamingMarkdown("abc ** `def` jhk [lmn](opq)")).toBe(
+          "abc \\*\\* `def` jhk [lmn](opq)",
+        );
+      });
+    });
+
+    test("should autoclose bold, even if the last node isn't a text node", () => {
+      expect(
+        processStreamingMarkdown("* **Asynchronous Function ( `async`"),
+      ).toBe("* **Asynchronous Function ( `async`**");
+    });
+
+    test("whitespace fakeouts should not be modified", () => {
+      expect(processStreamingMarkdown("** abc")).toBe("\\*\\* abc");
+    });
+
+    // TODO(drifkin): arguably this should just be removed entirely, but empty
+    // isn't so bad
+    test("should handle empty bolded items", () => {
+      expect(processStreamingMarkdown("**")).toBe("");
+    });
+  });
+
+  suite("partially closed bolded items", () => {
+    test("simple partial", () => {
+      expect(processStreamingMarkdown("**abc*")).toBe("**abc**");
+    });
+
+    test("partial with non-text node at end", () => {
+      expect(processStreamingMarkdown("**abc`def`*")).toBe("**abc`def`**");
+    });
+
+    test("partial with multiply nested ending nodes", () => {
+      expect(processStreamingMarkdown("**abc[abc](`def`)*")).toBe(
+        "**abc[abc](`def`)**",
+      );
+    });
+
+    test("normal emphasis should not be affected", () => {
+      expect(processStreamingMarkdown("*abc*")).toBe("*abc*");
+    });
+
+    test("normal emphasis with nested code should not be affected", () => {
+      expect(processStreamingMarkdown("*`abc`*")).toBe("*`abc`*");
+    });
+  });
+
+  test.skip("shouldn't autoclose immediately if there's a space before the closing *", () => {
+    expect(processStreamingMarkdown("**abc *")).toBe("**abc**");
+  });
+
+  // skipping for now because this requires partial link completion as well
+  suite.skip("nested blocks that each need autoclosing", () => {
+    test("emph nested in link nested in strong nested in list item", () => {
+      expect(processStreamingMarkdown("* **[abc **def")).toBe(
+        "* **[abc **def**]()**",
+      );
+    });
+
+    test("* **[ab *`def`", () => {
+      expect(processStreamingMarkdown("* **[ab *`def`")).toBe(
+        "* **[ab *`def`*]()**",
+      );
+    });
+  });
+});
+
+suite("numbered list items", () => {
+  test("should remove trailing numbers", () => {
+    expect(processStreamingMarkdown("1. First\n2")).toBe("1. First");
+  });
+
+  test("should remove trailing numbers with breaks before", () => {
+    expect(processStreamingMarkdown("1. First    \n2")).toBe("1. First");
+  });
+
+  test("should remove trailing numbers that form a new paragraph", () => {
+    expect(processStreamingMarkdown("1. First\n\n2")).toBe("1. First");
+  });
+
+  test("but should leave list items separated by two newlines", () => {
+    expect(processStreamingMarkdown("1. First\n\n2. S")).toBe(
+      "1. First\n\n2. S",
+    );
+  });
+});
+
+// TODO(drifkin):slop tests ahead, some are decent, but need to manually go
+// through them as I implement
+/*
+describe("StreamingMarkdownContent - processStreamingMarkdown", () => {
+  describe("Ambiguous endings removal", () => {
+    it("should remove list markers at the end", () => {
+      expect(processStreamingMarkdown("Some text\n* ")).toBe("Some text");
+      expect(processStreamingMarkdown("Some text\n*")).toBe("Some text");
+      expect(processStreamingMarkdown("* Item 1\n- ")).toBe("* Item 1");
+      expect(processStreamingMarkdown("* Item 1\n-")).toBe("* Item 1");
+      expect(processStreamingMarkdown("Text\n+ ")).toBe("Text");
+      expect(processStreamingMarkdown("Text\n+")).toBe("Text");
+      expect(processStreamingMarkdown("1. First\n2. ")).toBe("1. First");
+    });
+
+    it("should remove heading markers at the end", () => {
+      expect(processStreamingMarkdown("Some text\n# ")).toBe("Some text");
+      expect(processStreamingMarkdown("Some text\n#")).toBe("Some text\n#"); // # without space is not removed
+      expect(processStreamingMarkdown("# Title\n## ")).toBe("# Title");
+      expect(processStreamingMarkdown("# Title\n##")).toBe("# Title\n##"); // ## without space is not removed
+    });
+
+    it("should remove ambiguous bold markers at the end", () => {
+      expect(processStreamingMarkdown("Text **")).toBe("Text ");
+      expect(processStreamingMarkdown("Some text\n**")).toBe("Some text");
+    });
+
+    it("should remove code block markers at the end", () => {
+      expect(processStreamingMarkdown("Text\n```")).toBe("Text");
+      expect(processStreamingMarkdown("```")).toBe("");
+    });
+
+    it("should remove single backtick at the end", () => {
+      expect(processStreamingMarkdown("Text `")).toBe("Text ");
+      expect(processStreamingMarkdown("`")).toBe("");
+    });
+
+    it("should remove single asterisk at the end", () => {
+      expect(processStreamingMarkdown("Text *")).toBe("Text ");
+      expect(processStreamingMarkdown("*")).toBe("");
+    });
+
+    it("should handle empty content", () => {
+      expect(processStreamingMarkdown("")).toBe("");
+    });
+
+    it("should handle single line removals correctly", () => {
+      expect(processStreamingMarkdown("* ")).toBe("");
+      expect(processStreamingMarkdown("# ")).toBe("");
+      expect(processStreamingMarkdown("**")).toBe("");
+      expect(processStreamingMarkdown("`")).toBe("");
+    });
+
+    it("shouldn't have this regexp capture group bug", () => {
+      expect(
+        processStreamingMarkdown("Here's a shopping list:\n*"),
+      ).not.toContain("0*");
+      expect(processStreamingMarkdown("Here's a shopping list:\n*")).toBe(
+        "Here's a shopping list:",
+      );
+    });
+  });
+
+  describe("List markers", () => {
+    it("should preserve complete list items", () => {
+      expect(processStreamingMarkdown("* Complete item")).toBe(
+        "* Complete item",
+      );
+      expect(processStreamingMarkdown("- Another item")).toBe("- Another item");
+      expect(processStreamingMarkdown("+ Plus item")).toBe("+ Plus item");
+      expect(processStreamingMarkdown("1. Numbered item")).toBe(
+        "1. Numbered item",
+      );
+    });
+
+    it("should handle indented list markers", () => {
+      expect(processStreamingMarkdown("  * ")).toBe("  ");
+      expect(processStreamingMarkdown("    - ")).toBe("    ");
+      expect(processStreamingMarkdown("\t+ ")).toBe("\t");
+    });
+  });
+
+  describe("Heading markers", () => {
+    it("should preserve complete headings", () => {
+      expect(processStreamingMarkdown("# Complete Heading")).toBe(
+        "# Complete Heading",
+      );
+      expect(processStreamingMarkdown("## Subheading")).toBe("## Subheading");
+      expect(processStreamingMarkdown("### H3 Title")).toBe("### H3 Title");
+    });
+
+    it("should not affect # in other contexts", () => {
+      expect(processStreamingMarkdown("C# programming")).toBe("C# programming");
+      expect(processStreamingMarkdown("Issue #123")).toBe("Issue #123");
+    });
+  });
+
+  describe("Bold text", () => {
+    it("should close incomplete bold text", () => {
+      expect(processStreamingMarkdown("This is **bold text")).toBe(
+        "This is **bold text**",
+      );
+      expect(processStreamingMarkdown("Start **bold and more")).toBe(
+        "Start **bold and more**",
+      );
+      expect(processStreamingMarkdown("**just bold")).toBe("**just bold**");
+    });
+
+    it("should not affect complete bold text", () => {
+      expect(processStreamingMarkdown("**complete bold**")).toBe(
+        "**complete bold**",
+      );
+      expect(processStreamingMarkdown("Text **bold** more")).toBe(
+        "Text **bold** more",
+      );
+    });
+
+    it("should handle nested bold correctly", () => {
+      expect(processStreamingMarkdown("**bold** and **another")).toBe(
+        "**bold** and **another**",
+      );
+    });
+  });
+
+  describe("Italic text", () => {
+    it("should close incomplete italic text", () => {
+      expect(processStreamingMarkdown("This is *italic text")).toBe(
+        "This is *italic text*",
+      );
+      expect(processStreamingMarkdown("Start *italic and more")).toBe(
+        "Start *italic and more*",
+      );
+    });
+
+    it("should differentiate between list markers and italic", () => {
+      expect(processStreamingMarkdown("* Item\n* ")).toBe("* Item");
+      expect(processStreamingMarkdown("Some *italic text")).toBe(
+        "Some *italic text*",
+      );
+      expect(processStreamingMarkdown("*just italic")).toBe("*just italic*");
+    });
+
+    it("should not affect complete italic text", () => {
+      expect(processStreamingMarkdown("*complete italic*")).toBe(
+        "*complete italic*",
+      );
+      expect(processStreamingMarkdown("Text *italic* more")).toBe(
+        "Text *italic* more",
+      );
+    });
+  });
+
+  describe("Code blocks", () => {
+    it("should close incomplete code blocks", () => {
+      expect(processStreamingMarkdown("```javascript\nconst x = 42;")).toBe(
+        "```javascript\nconst x = 42;\n```",
+      );
+      expect(processStreamingMarkdown("```\ncode here")).toBe(
+        "```\ncode here\n```",
+      );
+    });
+
+    it("should not affect complete code blocks", () => {
+      expect(processStreamingMarkdown("```\ncode\n```")).toBe("```\ncode\n```");
+      expect(processStreamingMarkdown("```js\nconst x = 1;\n```")).toBe(
+        "```js\nconst x = 1;\n```",
+      );
+    });
+
+    it("should handle nested code blocks correctly", () => {
+      expect(processStreamingMarkdown("```\ncode\n```\n```python")).toBe(
+        "```\ncode\n```\n```python\n```",
+      );
+    });
+
+    it("should not process markdown inside code blocks", () => {
+      expect(processStreamingMarkdown("```\n* not a list\n**not bold**")).toBe(
+        "```\n* not a list\n**not bold**\n```",
+      );
+    });
+  });
+
+  describe("Inline code", () => {
+    it("should close incomplete inline code", () => {
+      expect(processStreamingMarkdown("This is `inline code")).toBe(
+        "This is `inline code`",
+      );
+      expect(processStreamingMarkdown("Use `console.log")).toBe(
+        "Use `console.log`",
+      );
+    });
+
+    it("should not affect complete inline code", () => {
+      expect(processStreamingMarkdown("`complete code`")).toBe(
+        "`complete code`",
+      );
+      expect(processStreamingMarkdown("Use `code` here")).toBe(
+        "Use `code` here",
+      );
+    });
+
+    it("should handle multiple inline codes correctly", () => {
+      expect(processStreamingMarkdown("`code` and `more")).toBe(
+        "`code` and `more`",
+      );
+    });
+
+    it("should not confuse inline code with code blocks", () => {
+      expect(processStreamingMarkdown("```\nblock\n```\n`inline")).toBe(
+        "```\nblock\n```\n`inline`",
+      );
+    });
+  });
+
+  describe("Complex streaming scenarios", () => {
+    it("should handle progressive streaming of a heading", () => {
+      const steps = [
+        { input: "#", expected: "#" }, // # alone is not removed (needs space)
+        { input: "# ", expected: "" },
+        { input: "# H", expected: "# H" },
+        { input: "# Hello", expected: "# Hello" },
+      ];
+      steps.forEach(({ input, expected }) => {
+        expect(processStreamingMarkdown(input)).toBe(expected);
+      });
+    });
+
+    it("should handle progressive streaming of bold text", () => {
+      const steps = [
+        { input: "*", expected: "" },
+        { input: "**", expected: "" },
+        { input: "**b", expected: "**b**" },
+        { input: "**bold", expected: "**bold**" },
+        { input: "**bold**", expected: "**bold**" },
+      ];
+      steps.forEach(({ input, expected }) => {
+        expect(processStreamingMarkdown(input)).toBe(expected);
+      });
+    });
+
+    it("should handle multiline content with various patterns", () => {
+      const multiline = `# Title
+      
+This is a paragraph with **bold text** and *italic text*.
+
+* Item 1
+* Item 2
+* `;
+
+      const expected = `# Title
+      
+This is a paragraph with **bold text** and *italic text*.
+
+* Item 1
+* Item 2`;
+
+      expect(processStreamingMarkdown(multiline)).toBe(expected);
+    });
+
+    it("should only fix the last line", () => {
+      expect(processStreamingMarkdown("# Complete\n# Another\n# ")).toBe(
+        "# Complete\n# Another",
+      );
+      expect(processStreamingMarkdown("* Item 1\n* Item 2\n* ")).toBe(
+        "* Item 1\n* Item 2",
+      );
+    });
+
+    it("should handle mixed content correctly", () => {
+      const input = `# Header
+
+This has **bold** text and *italic* text.
+
+\`\`\`js
+const x = 42;
+\`\`\`
+
+Now some \`inline code\` and **unclosed bold`;
+
+      const expected = `# Header
+
+This has **bold** text and *italic* text.
+
+\`\`\`js
+const x = 42;
+\`\`\`
+
+Now some \`inline code\` and **unclosed bold**`;
+
+      expect(processStreamingMarkdown(input)).toBe(expected);
+    });
+  });
+
+  describe("Edge cases with escaping", () => {
+    it("should handle escaped asterisks (future enhancement)", () => {
+      // Note: Current implementation doesn't handle escaping
+      // This is a known limitation - escaped characters still trigger closing
+      expect(processStreamingMarkdown("Text \\*not italic")).toBe(
+        "Text \\*not italic*",
+      );
+    });
+
+    it("should handle escaped backticks (future enhancement)", () => {
+      // Note: Current implementation doesn't handle escaping
+      // This is a known limitation - escaped characters still trigger closing
+      expect(processStreamingMarkdown("Text \\`not code")).toBe(
+        "Text \\`not code`",
+      );
+    });
+  });
+
+  describe("Code block edge cases", () => {
+    it("should handle triple backticks in the middle of lines", () => {
+      expect(processStreamingMarkdown("Text ``` in middle")).toBe(
+        "Text ``` in middle\n```",
+      );
+      expect(processStreamingMarkdown("```\nText ``` in code\nmore")).toBe(
+        "```\nText ``` in code\nmore\n```",
+      );
+    });
+
+    it("should properly close code blocks with language specifiers", () => {
+      expect(processStreamingMarkdown("```typescript")).toBe(
+        "```typescript\n```",
+      );
+      expect(processStreamingMarkdown("```typescript\nconst x = 1")).toBe(
+        "```typescript\nconst x = 1\n```",
+      );
+    });
+
+    it("should remove a completely empty partial code block", () => {
+      expect(processStreamingMarkdown("```\n")).toBe("");
+    });
+  });
+});
+
+*/
--- a/app/ui/app/src/components/StreamingMarkdownContent.tsx
+++ b/app/ui/app/src/components/StreamingMarkdownContent.tsx
@@ -1,123 +1,66 @@
 import React from "react";
-import { Streamdown, defaultRemarkPlugins } from "streamdown";
+import Markdown from "react-markdown";
+import remarkGfm from "remark-gfm";
+import remarkMath from "remark-math";
+import rehypeRaw from "rehype-raw";
+import rehypeSanitize, { defaultSchema } from "rehype-sanitize";
+import rehypePrismPlus from "rehype-prism-plus";
+import rehypeKatex from "rehype-katex";
+import remarkStreamingMarkdown, {
+  type LastNodeInfo,
+} from "@/utils/remarkStreamingMarkdown";
+import type { PluggableList } from "unified";
 import remarkCitationParser from "@/utils/remarkCitationParser";
 import CopyButton from "./CopyButton";
-import type { BundledLanguage } from "shiki";
-import { highlighter } from "@/lib/highlighter";

 interface StreamingMarkdownContentProps {
  content: string;
  isStreaming?: boolean;
  size?: "sm" | "md" | "lg";
+  onLastNode?: (info: LastNodeInfo) => void;
  browserToolResult?: any; // TODO: proper type
 }

-// Helper to extract text from React nodes
-const extractText = (node: React.ReactNode): string => {
-  if (typeof node === "string") return node;
-  if (typeof node === "number") return String(node);
-  if (!node) return "";
-  if (React.isValidElement(node)) {
-    const props = node.props as any;
-    if (props?.children) {
-      return extractText(props.children as React.ReactNode);
-    }
-  }
-  if (Array.isArray(node)) {
-    return node.map(extractText).join("");
-  }
-  return "";
-};
-
 const CodeBlock = React.memo(
-  ({ children }: React.HTMLAttributes<HTMLPreElement>) => {
-    // Extract code and language from children
-    const codeElement = children as React.ReactElement<{
-      className?: string;
-      children: React.ReactNode;
-    }>;
-    const language =
-      codeElement.props.className?.replace(/language-/, "") || "";
-    const codeText = extractText(codeElement.props.children);
+  ({ children, className, ...props }: React.HTMLAttributes<HTMLPreElement>) => {
+    const extractText = React.useCallback((node: React.ReactNode): string => {
+      if (typeof node === "string") return node;
+      if (typeof node === "number") return String(node);
+      if (!node) return "";

-    // Synchronously highlight code using the pre-loaded highlighter
-    const tokens = React.useMemo(() => {
-      if (!highlighter) return null;
-
-      try {
-        return {
-          light: highlighter.codeToTokensBase(codeText, {
-            lang: language as BundledLanguage,
-            theme: "one-light" as any,
-          }),
-          dark: highlighter.codeToTokensBase(codeText, {
-            lang: language as BundledLanguage,
-            theme: "one-dark" as any,
-          }),
-        };
-      } catch (error) {
-        console.error("Failed to highlight code:", error);
-        return null;
+      if (React.isValidElement(node)) {
+        if (
+          node.props &&
+          typeof node.props === "object" &&
+          "children" in node.props
+        ) {
+          return extractText(node.props.children as React.ReactNode);
+        }
      }
-    }, [codeText, language]);
+
+      if (Array.isArray(node)) {
+        return node.map(extractText).join("");
+      }
+
+      return "";
+    }, []);
+
+    const language = className?.replace(/language-/, "") || "";

    return (
      <div className="relative bg-neutral-100 dark:bg-neutral-800 rounded-2xl overflow-hidden my-6">
-        <div className="flex select-none">
-          {language && (
-            <div className="text-[13px] text-neutral-500 dark:text-neutral-400 font-mono px-4 py-2">
-              {language}
-            </div>
-          )}
+        <div className="flex justify-between select-none">
+          <div className="text-[13px] text-neutral-500 dark:text-neutral-400 font-mono px-4 py-2">
+            {language}
+          </div>
          <CopyButton
-            content={codeText}
+            content={extractText(children)}
            showLabels={true}
-            className="copy-button text-neutral-500 dark:text-neutral-400 bg-neutral-100 dark:bg-neutral-800 ml-auto"
+            className="copy-button text-neutral-500 dark:text-neutral-400 bg-neutral-100 dark:bg-neutral-800"
          />
        </div>
-        {/* Light mode */}
-        <pre className="dark:hidden m-0 bg-neutral-100 text-sm overflow-x-auto p-4">
-          <code className="font-mono text-sm">
-            {tokens?.light
-              ? tokens.light.map((line: any, i: number) => (
-                  <React.Fragment key={i}>
-                    {line.map((token: any, j: number) => (
-                      <span
-                        key={j}
-                        style={{
-                          color: token.color,
-                        }}
-                      >
-                        {token.content}
-                      </span>
-                    ))}
-                    {i < tokens.light.length - 1 && "\n"}
-                  </React.Fragment>
-                ))
-              : codeText}
-          </code>
-        </pre>
-        {/* Dark mode */}
-        <pre className="hidden dark:block m-0 bg-neutral-800 text-sm overflow-x-auto p-4">
-          <code className="font-mono text-sm">
-            {tokens?.dark
-              ? tokens.dark.map((line: any, i: number) => (
-                  <React.Fragment key={i}>
-                    {line.map((token: any, j: number) => (
-                      <span
-                        key={j}
-                        style={{
-                          color: token.color,
-                        }}
-                      >
-                        {token.content}
-                      </span>
-                    ))}
-                    {i < tokens.dark.length - 1 && "\n"}
-                  </React.Fragment>
-                ))
-              : codeText}
-          </code>
+        <pre className={className} {...props}>
+          {children}
        </pre>
      </div>
    );
@@ -125,19 +68,65 @@ const CodeBlock = React.memo(
 );

 const StreamingMarkdownContent: React.FC<StreamingMarkdownContentProps> =
-  React.memo(({ content, isStreaming = false, size, browserToolResult }) => {
-    // Build the remark plugins array - keep default GFM and Math, add citations
-    const remarkPlugins = React.useMemo(() => {
-      return [
-        defaultRemarkPlugins.gfm,
-        defaultRemarkPlugins.math,
-        remarkCitationParser,
-      ];
-    }, []);
+  React.memo(
+    ({ content, isStreaming = false, size, onLastNode, browserToolResult }) => {
+      // Build the remark plugins array
+      const remarkPlugins = React.useMemo(() => {
+        const plugins: PluggableList = [
+          remarkGfm,
+          [remarkMath, { singleDollarTextMath: false }],
+          remarkCitationParser,
+        ];

-    return (
-      <div
-        className={`
+        // Add streaming plugin when in streaming mode
+        if (isStreaming) {
+          plugins.push([remarkStreamingMarkdown, { debug: true, onLastNode }]);
+        }
+
+        return plugins;
+      }, [isStreaming, onLastNode]);
+
+      // Create a custom sanitization schema that allows math elements
+      const sanitizeSchema = React.useMemo(() => {
+        return {
+          ...defaultSchema,
+          attributes: {
+            ...defaultSchema.attributes,
+            span: [
+              ...(defaultSchema.attributes?.span || []),
+              ["className", /^katex/],
+            ],
+            div: [
+              ...(defaultSchema.attributes?.div || []),
+              ["className", /^katex/],
+            ],
+            "ol-citation": ["cursor", "start", "end"],
+          },
+          tagNames: [
+            ...(defaultSchema.tagNames || []),
+            "math",
+            "mrow",
+            "mi",
+            "mo",
+            "mn",
+            "msup",
+            "msub",
+            "mfrac",
+            "mover",
+            "munder",
+            "msqrt",
+            "mroot",
+            "merror",
+            "mspace",
+            "mpadded",
+            "ol-citation",
+          ],
+        };
+      }, []);
+
+      return (
+        <div
+          className={`
          max-w-full
          ${size === "sm" ? "prose-sm" : size === "lg" ? "prose-lg" : ""}
          prose
@@ -155,27 +144,7 @@ const StreamingMarkdownContent: React.FC<StreamingMarkdownContentProps> =
          prose-pre:my-0
          prose-pre:max-w-full
          prose-pre:pt-1
-          [&_table]:border-collapse
-          [&_table]:w-full
-          [&_table]:border
-          [&_table]:border-neutral-200
-          [&_table]:rounded-lg
-          [&_table]:overflow-hidden
-          [&_th]:px-3
-          [&_th]:py-2
-          [&_th]:text-left
-          [&_th]:font-semibold
-          [&_th]:border-b
-          [&_th]:border-r
-          [&_th]:border-neutral-200
-          [&_th:last-child]:border-r-0
-          [&_td]:px-3
-          [&_td]:py-2
-          [&_td]:border-r
-          [&_td]:border-neutral-200
-          [&_td:last-child]:border-r-0
-          [&_tbody_tr:not(:last-child)_td]:border-b
-          [&_code:not(pre_code)]:text-neutral-700
+           [&_code:not(pre_code)]:text-neutral-700
          [&_code:not(pre_code)]:bg-neutral-100
          [&_code:not(pre_code)]:font-normal
          [&_code:not(pre_code)]:px-1.5
@@ -191,10 +160,6 @@ const StreamingMarkdownContent: React.FC<StreamingMarkdownContentProps> =
          dark:prose-strong:text-neutral-200
          dark:prose-pre:text-neutral-200
          dark:prose:pre:text-neutral-200
-          dark:[&_table]:border-neutral-700
-          dark:[&_thead]:bg-neutral-800
-          dark:[&_th]:border-neutral-700
-          dark:[&_td]:border-neutral-700
          dark:[&_code:not(pre_code)]:text-neutral-200
          dark:[&_code:not(pre_code)]:bg-neutral-800
          dark:[&_code:not(pre_code)]:font-normal
@@ -202,86 +167,104 @@ const StreamingMarkdownContent: React.FC<StreamingMarkdownContentProps> =
          dark:prose-li:marker:text-neutral-300
          break-words
        `}
-      >
-        <StreamingMarkdownErrorBoundary
-          content={content}
-          isStreaming={isStreaming}
        >
-          <Streamdown
-            parseIncompleteMarkdown={isStreaming}
-            isAnimating={isStreaming}
-            remarkPlugins={remarkPlugins}
-            controls={false}
-            components={{
-              pre: CodeBlock,
-              table: ({
-                children,
-                ...props
-              }: React.HTMLAttributes<HTMLTableElement>) => (
-                <div className="overflow-x-auto max-w-full">
-                  <table
-                    {...props}
-                    className="border-collapse w-full border border-neutral-200 dark:border-neutral-700 rounded-lg overflow-hidden"
-                  >
-                    {children}
-                  </table>
-                </div>
-              ),
-              // @ts-expect-error: custom citation type
-              "ol-citation": ({
-                cursor,
-              }: {
-                cursor: number;
-                start: number;
-                end: number;
-              }) => {
-                const pageStack = browserToolResult?.page_stack;
-                const hasValidPage = pageStack && cursor < pageStack.length;
-                const pageUrl = hasValidPage ? pageStack[cursor] : null;
-
-                const getPageTitle = (url: string) => {
-                  if (url.startsWith("search_results_")) {
-                    const searchTerm = url.substring("search_results_".length);
-                    return `Search: ${searchTerm}`;
-                  }
-                  try {
-                    const urlObj = new URL(url);
-                    return urlObj.hostname;
-                  } catch {
-                    return url;
-                  }
-                };
-
-                const citationElement = (
-                  <span className="text-xs text-neutral-500 dark:text-neutral-400 bg-neutral-100 dark:bg-neutral-800 rounded-full px-2 py-1 ml-1">
-                    [{cursor}]
-                  </span>
-                );
-
-                if (pageUrl && pageUrl.startsWith("http")) {
-                  return (
-                    <a
-                      href={pageUrl}
-                      target="_blank"
-                      rel="noopener noreferrer"
-                      className="inline-flex items-center hover:opacity-80 transition-opacity no-underline"
-                      title={getPageTitle(pageUrl)}
-                    >
-                      {citationElement}
-                    </a>
-                  );
-                }
-
-                return citationElement;
-              },
-            }}
+          <StreamingMarkdownErrorBoundary
+            content={content}
+            isStreaming={isStreaming}
          >
-            {content}
-          </Streamdown>
-        </StreamingMarkdownErrorBoundary>
-      </div>
-    );
-  });
+            <Markdown
+              remarkPlugins={remarkPlugins}
+              rehypePlugins={
+                [
+                  [rehypeRaw, { allowDangerousHtml: true }],
+                  [rehypeSanitize, sanitizeSchema],
+                  [rehypePrismPlus, { ignoreMissing: true }],
+                  [
+                    rehypeKatex,
+                    {
+                      errorColor: "#000000", // Black instead of red for errors
+                      strict: false, // Be more lenient with parsing
+                      throwOnError: false,
+                    },
+                  ],
+                ] as PluggableList
+              }
+              components={{
+                pre: CodeBlock,
+                table: ({
+                  children,
+                  ...props
+                }: React.HTMLAttributes<HTMLTableElement>) => (
+                  <div className="overflow-x-auto max-w-full">
+                    <table {...props}>{children}</table>
+                  </div>
+                ),
+                // @ts-expect-error: custom type
+                "ol-citation": ({
+                  cursor,
+                  // start,
+                  // end,
+                }: {
+                  cursor: number;
+                  start: number;
+                  end: number;
+                }) => {
+                  // Check if we have a page_stack and if the cursor is valid
+                  const pageStack = browserToolResult?.page_stack;
+                  const hasValidPage = pageStack && cursor < pageStack.length;
+                  const pageUrl = hasValidPage ? pageStack[cursor] : null;
+
+                  // Extract a readable title from the URL if possible
+                  const getPageTitle = (url: string) => {
+                    if (url.startsWith("search_results_")) {
+                      const searchTerm = url.substring(
+                        "search_results_".length,
+                      );
+                      return `Search: ${searchTerm}`;
+                    }
+                    // For regular URLs, try to extract domain or use full URL
+                    try {
+                      const urlObj = new URL(url);
+                      return urlObj.hostname;
+                    } catch {
+                      // If not a valid URL, return as is
+                      return url;
+                    }
+                  };
+
+                  const citationElement = (
+                    <span className="text-xs text-neutral-500 dark:text-neutral-400 bg-neutral-100 dark:bg-neutral-800 rounded-full px-2 py-1 ml-1">
+                      [{cursor}]
+                    </span>
+                  );
+
+                  // If we have a valid page URL, wrap in a link
+                  if (pageUrl && pageUrl.startsWith("http")) {
+                    return (
+                      <a
+                        href={pageUrl}
+                        target="_blank"
+                        rel="noopener noreferrer"
+                        className="inline-flex items-center hover:opacity-80 transition-opacity no-underline"
+                        title={getPageTitle(pageUrl)}
+                      >
+                        {citationElement}
+                      </a>
+                    );
+                  }
+
+                  // Otherwise, just return the citation without a link
+                  return citationElement;
+                },
+              }}
+            >
+              {content}
+            </Markdown>
+          </StreamingMarkdownErrorBoundary>
+        </div>
+      );
+    },
+  );

 interface StreamingMarkdownErrorBoundaryProps {
  content: string;
--- a/app/ui/app/src/components/Thinking.tsx
+++ b/app/ui/app/src/components/Thinking.tsx
@@ -50,33 +50,21 @@ export default function Thinking({
  // Position content to show bottom when collapsed
  useEffect(() => {
    if (isCollapsed && contentRef.current && wrapperRef.current) {
-      requestAnimationFrame(() => {
-        if (!contentRef.current || !wrapperRef.current) return;
-
-        const contentHeight = contentRef.current.scrollHeight;
-        const wrapperHeight = wrapperRef.current.clientHeight;
-        if (contentHeight > wrapperHeight) {
-          const translateY = -(contentHeight - wrapperHeight);
-          contentRef.current.style.transform = `translateY(${translateY}px)`;
-          setHasOverflow(true);
-        } else {
-          contentRef.current.style.transform = "translateY(0)";
-          setHasOverflow(false);
-        }
-      });
+      const contentHeight = contentRef.current.scrollHeight;
+      const wrapperHeight = wrapperRef.current.clientHeight;
+      if (contentHeight > wrapperHeight) {
+        const translateY = -(contentHeight - wrapperHeight);
+        contentRef.current.style.transform = `translateY(${translateY}px)`;
+        setHasOverflow(true);
+      } else {
+        setHasOverflow(false);
+      }
    } else if (contentRef.current) {
      contentRef.current.style.transform = "translateY(0)";
      setHasOverflow(false);
    }
  }, [thinking, isCollapsed]);

-  useEffect(() => {
-    if (activelyThinking && wrapperRef.current && !isCollapsed) {
-      // When expanded and actively thinking, scroll to bottom
-      wrapperRef.current.scrollTop = wrapperRef.current.scrollHeight;
-    }
-  }, [thinking, activelyThinking, isCollapsed]);
-
  const handleToggle = () => {
    setIsCollapsed(!isCollapsed);
    setHasUserInteracted(true);
@@ -85,9 +73,8 @@ export default function Thinking({
  // Calculate max height for smooth animations
  const getMaxHeight = () => {
    if (isCollapsed) {
-      return finishedThinking ? "0px" : "12rem";
+      return finishedThinking ? "0px" : "12rem"; // 8rem = 128px (same as max-h-32)
    }
-    // When expanded, use the content height or grow naturally
    return contentHeight ? `${contentHeight}px` : "none";
  };

@@ -144,11 +131,10 @@ export default function Thinking({
      </div>
      <div
        ref={wrapperRef}
-        className={`text-xs text-neutral-500 dark:text-neutral-500 rounded-md
-          transition-[max-height,opacity] duration-300 ease-in-out relative ml-6 mt-2
-          ${isCollapsed ? "overflow-hidden" : "overflow-y-auto"}`}
+        className={`text-xs text-neutral-500 dark:text-neutral-500 rounded-md overflow-hidden
+          transition-[max-height,opacity] duration-300 ease-in-out relative ml-6 mt-2`}
        style={{
-          maxHeight: isCollapsed ? getMaxHeight() : undefined,
+          maxHeight: getMaxHeight(),
          opacity: isCollapsed && finishedThinking ? 0 : 1,
        }}
      >
--- a/app/ui/app/src/hooks/useChats.ts
+++ b/app/ui/app/src/hooks/useChats.ts
@@ -7,7 +7,6 @@ import { createQueryBatcher } from "./useQueryBatcher";
 import { useRefetchModels } from "./useModels";
 import { useStreamingContext } from "@/contexts/StreamingContext";
 import { useSettings } from "./useSettings";
-import { getModelCapabilities } from "@/api";

 export const useChats = () => {
  return useQuery({
@@ -607,24 +606,6 @@ export const useSendMessage = (chatId: string) => {
              queryClient.setQueryData(["staleModels"], newStaleMap);

              queryClient.invalidateQueries({ queryKey: ["models"] });
-
-              // Fetch fresh capabilities for the downloaded model
-              getModelCapabilities(selectedModel.model)
-                .then((capabilities) => {
-                  queryClient.setQueryData(
-                    ["modelCapabilities", selectedModel.model],
-                    capabilities,
-                  );
-                })
-                .catch((error) => {
-                  console.error(
-                    "Failed to fetch capabilities after download:",
-                    error,
-                  );
-                  queryClient.invalidateQueries({
-                    queryKey: ["modelCapabilities", selectedModel.model],
-                  });
-                });
            }
            break;
          }
--- a/app/ui/app/src/hooks/useDownloadModel.ts
+++ b/app/ui/app/src/hooks/useDownloadModel.ts
@@ -0,0 +1,114 @@
+import { useMutation, useQueryClient } from "@tanstack/react-query";
+import { useState } from "react";
+import { pullModel } from "@/api";
+import { useSelectedModel } from "./useSelectedModel";
+import { useSettings } from "./useSettings";
+
+interface DownloadProgress {
+  status: string;
+  digest?: string;
+  total?: number;
+  completed?: number;
+  done?: boolean;
+}
+
+export function useDownloadModel(chatId?: string) {
+  const queryClient = useQueryClient();
+  const { selectedModel } = useSelectedModel(chatId);
+  const { setSettings } = useSettings();
+  const [downloadProgress, setDownloadProgress] =
+    useState<DownloadProgress | null>(null);
+  const [abortController, setAbortController] =
+    useState<AbortController | null>(null);
+  const [downloadingChatIds, setDownloadingChatIds] = useState<Set<string>>(
+    new Set(),
+  );
+
+  const mutation = useMutation({
+    mutationFn: async (modelName: string) => {
+      const controller = new AbortController();
+      setAbortController(controller);
+      setDownloadProgress({ status: "Starting download..." });
+      if (chatId) {
+        setDownloadingChatIds((prev) => new Set(prev).add(chatId));
+      }
+
+      try {
+        for await (const progress of pullModel(modelName, controller.signal)) {
+          setDownloadProgress(progress);
+
+          if (progress.status === "success") {
+            // Update selected model to indicate it's now available locally
+            if (selectedModel && selectedModel.model === modelName) {
+              setSettings({ SelectedModel: modelName });
+            }
+            // Invalidate models query to refresh the list
+            await queryClient.invalidateQueries({ queryKey: ["models"] });
+            break;
+          }
+        }
+      } finally {
+        setAbortController(null);
+        if (chatId) {
+          setDownloadingChatIds((prev) => {
+            const newSet = new Set(prev);
+            newSet.delete(chatId);
+            return newSet;
+          });
+        }
+      }
+    },
+    onSuccess: () => {
+      setDownloadProgress(null);
+      if (chatId) {
+        setDownloadingChatIds((prev) => {
+          const newSet = new Set(prev);
+          newSet.delete(chatId);
+          return newSet;
+        });
+      }
+    },
+    onError: (error: Error) => {
+      const status =
+        error.name === "AbortError" ? "Download cancelled" : "Download failed";
+      setDownloadProgress({ status, done: true });
+
+      // Clear error message after delay
+      const delay = error.name === "AbortError" ? 1500 : 3000;
+      setTimeout(() => {
+        setDownloadProgress(null);
+        if (chatId) {
+          setDownloadingChatIds((prev) => {
+            const newSet = new Set(prev);
+            newSet.delete(chatId);
+            return newSet;
+          });
+        }
+      }, delay);
+    },
+  });
+
+  const cancelDownload = () => {
+    if (abortController) {
+      abortController.abort();
+      setAbortController(null);
+      if (chatId) {
+        setDownloadingChatIds((prev) => {
+          const newSet = new Set(prev);
+          newSet.delete(chatId);
+          return newSet;
+        });
+      }
+    }
+  };
+
+  return {
+    downloadModel: mutation.mutate,
+    isDownloading:
+      mutation.isPending && chatId ? downloadingChatIds.has(chatId) : false,
+    downloadProgress:
+      chatId && downloadingChatIds.has(chatId) ? downloadProgress : null,
+    error: mutation.error,
+    cancelDownload,
+  };
+}
--- a/app/ui/app/src/hooks/useUser.ts
+++ b/app/ui/app/src/hooks/useUser.ts
@@ -1,20 +1,29 @@
 import { useQuery, useMutation, useQueryClient } from "@tanstack/react-query";
+import { useEffect, useState } from "react";
 import { fetchUser, fetchConnectUrl, disconnectUser } from "@/api";

 export function useUser() {
  const queryClient = useQueryClient();
+  const [initialDataLoaded, setInitialDataLoaded] = useState(false);
+
+  // Wait for initial data to be loaded
+  useEffect(() => {
+    const initialPromise = window.__initialUserDataPromise;
+    if (initialPromise) {
+      initialPromise.finally(() => {
+        setInitialDataLoaded(true);
+      });
+    } else {
+      setInitialDataLoaded(true);
+    }
+  }, []);

  const userQuery = useQuery({
    queryKey: ["user"],
-    queryFn: async () => {
-      const result = await fetchUser();
-      return result;
-    },
+    queryFn: () => fetchUser(),
    staleTime: 5 * 60 * 1000, // Consider data stale after 5 minutes
    gcTime: 10 * 60 * 1000, // Keep in cache for 10 minutes
-    retry: 10,
-    retryDelay: (attemptIndex) => Math.min(500 * attemptIndex, 2000),
-    refetchOnMount: true, // Always fetch when component mounts
+    initialData: null, // Start with null to prevent flashing
  });

  // Mutation to refresh user data
@@ -40,15 +49,14 @@ export function useUser() {
    },
  });

-  const isLoading = userQuery.isLoading || userQuery.isFetching;
-  const isAuthenticated = Boolean(userQuery.data?.name);
-
  return {
    user: userQuery.data,
-    isLoading,
+    isLoading:
+      !initialDataLoaded ||
+      (userQuery.isLoading && userQuery.data === undefined), // Show loading until initial data is loaded
    isError: userQuery.isError,
    error: userQuery.error,
-    isAuthenticated,
+    isAuthenticated: Boolean(userQuery.data?.name),
    refreshUser: refreshUser.mutate,
    isRefreshing: refreshUser.isPending,
    refetchUser: userQuery.refetch,
--- a/app/ui/app/src/index.css
+++ b/app/ui/app/src/index.css
@@ -16,6 +16,793 @@
    --text-color: #ffffff;
  }
 }
+@media (prefers-color-scheme: light) {
+  .prose {
+    /**
+    * One Light theme for prism.js
+    * Based on Atom's One Light theme: https://github.com/atom/atom/tree/master/packages/one-light-syntax
+    */
+
+    /**
+    * One Light colours (accurate as of commit eb064bf on 19 Feb 2021)
+    * From colors.less
+    * --mono-1: hsl(230, 8%, 24%);
+    * --mono-2: hsl(230, 6%, 44%);
+    * --mono-3: hsl(230, 4%, 64%)
+    * --hue-1: hsl(198, 99%, 37%);
+    * --hue-2: hsl(221, 87%, 60%);
+    * --hue-3: hsl(301, 63%, 40%);
+    * --hue-4: hsl(119, 34%, 47%);
+    * --hue-5: hsl(5, 74%, 59%);
+    * --hue-5-2: hsl(344, 84%, 43%);
+    * --hue-6: hsl(35, 99%, 36%);
+    * --hue-6-2: hsl(35, 99%, 40%);
+    * --syntax-fg: hsl(230, 8%, 24%);
+    * --syntax-bg: hsl(230, 1%, 98%);
+    * --syntax-gutter: hsl(230, 1%, 62%);
+    * --syntax-guide: hsla(230, 8%, 24%, 0.2);
+    * --syntax-accent: hsl(230, 100%, 66%);
+    * From syntax-variables.less
+    * --syntax-selection-color: hsl(230, 1%, 90%);
+    * --syntax-gutter-background-color-selected: hsl(230, 1%, 90%);
+    * --syntax-cursor-line: hsla(230, 8%, 24%, 0.05);
+    */
+
+    .token.comment,
+    .token.prolog,
+    .token.cdata {
+      color: hsl(230, 4%, 64%);
+    }
+
+    .token.doctype,
+    .token.punctuation,
+    .token.entity {
+      color: hsl(230, 8%, 24%);
+    }
+
+    .token.attr-name,
+    .token.class-name,
+    .token.boolean,
+    .token.constant,
+    .token.number,
+    .token.atrule {
+      color: hsl(35, 99%, 36%);
+    }
+
+    .token.keyword {
+      color: hsl(301, 63%, 40%);
+    }
+
+    .token.property,
+    .token.tag,
+    .token.symbol,
+    .token.deleted,
+    .token.important {
+      color: hsl(5, 74%, 59%);
+    }
+
+    .token.selector,
+    .token.string,
+    .token.char,
+    .token.builtin,
+    .token.inserted,
+    .token.regex,
+    .token.attr-value,
+    .token.attr-value > .token.punctuation {
+      color: hsl(119, 34%, 47%);
+    }
+
+    .token.variable,
+    .token.operator,
+    .token.function {
+      color: hsl(221, 87%, 60%);
+    }
+
+    .token.url {
+      color: hsl(198, 99%, 37%);
+    }
+
+    /* HTML overrides */
+    .token.attr-value > .token.punctuation.attr-equals,
+    .token.special-attr > .token.attr-value > .token.value.css {
+      color: hsl(230, 8%, 24%);
+    }
+
+    /* CSS overrides */
+    .language-css .token.selector {
+      color: hsl(5, 74%, 59%);
+    }
+
+    .language-css .token.property {
+      color: hsl(230, 8%, 24%);
+    }
+
+    .language-css .token.function,
+    .language-css .token.url > .token.function {
+      color: hsl(198, 99%, 37%);
+    }
+
+    .language-css .token.url > .token.string.url {
+      color: hsl(119, 34%, 47%);
+    }
+
+    .language-css .token.important,
+    .language-css .token.atrule .token.rule {
+      color: hsl(301, 63%, 40%);
+    }
+
+    /* JS overrides */
+    .language-javascript .token.operator {
+      color: hsl(301, 63%, 40%);
+    }
+
+    .language-javascript
+      .token.template-string
+      > .token.interpolation
+      > .token.interpolation-punctuation.punctuation {
+      color: hsl(344, 84%, 43%);
+    }
+
+    /* JSON overrides */
+    .language-json .token.operator {
+      color: hsl(230, 8%, 24%);
+    }
+
+    .language-json .token.null.keyword {
+      color: hsl(35, 99%, 36%);
+    }
+
+    /* MD overrides */
+    .language-markdown .token.url,
+    .language-markdown .token.url > .token.operator,
+    .language-markdown .token.url-reference.url > .token.string {
+      color: hsl(230, 8%, 24%);
+    }
+
+    .language-markdown .token.url > .token.content {
+      color: hsl(221, 87%, 60%);
+    }
+
+    .language-markdown .token.url > .token.url,
+    .language-markdown .token.url-reference.url {
+      color: hsl(198, 99%, 37%);
+    }
+
+    .language-markdown .token.blockquote.punctuation,
+    .language-markdown .token.hr.punctuation {
+      color: hsl(230, 4%, 64%);
+      font-style: italic;
+    }
+
+    .language-markdown .token.code-snippet {
+      color: hsl(119, 34%, 47%);
+    }
+
+    .language-markdown .token.bold .token.content {
+      color: hsl(35, 99%, 36%);
+    }
+
+    .language-markdown .token.italic .token.content {
+      color: hsl(301, 63%, 40%);
+    }
+
+    .language-markdown .token.strike .token.content,
+    .language-markdown .token.strike .token.punctuation,
+    .language-markdown .token.list.punctuation,
+    .language-markdown .token.title.important > .token.punctuation {
+      color: hsl(5, 74%, 59%);
+    }
+
+    /* General */
+    .token.bold {
+      font-weight: bold;
+    }
+
+    .token.comment,
+    .token.italic {
+      font-style: italic;
+    }
+
+    .token.entity {
+      cursor: help;
+    }
+
+    .token.namespace {
+      opacity: 0.8;
+    }
+
+    /* Plugin overrides */
+    /* Selectors should have higher specificity than those in the plugins' default stylesheets */
+
+    /* Show Invisibles plugin overrides */
+    .token.token.tab:not(:empty):before,
+    .token.token.cr:before,
+    .token.token.lf:before,
+    .token.token.space:before {
+      color: hsla(230, 8%, 24%, 0.2);
+    }
+
+    /* Toolbar plugin overrides */
+    /* Space out all buttons and move them away from the right edge of the code block */
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item {
+      margin-right: 0.4em;
+    }
+
+    /* Styling the buttons */
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item > button,
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item > a,
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item > span {
+      background: hsl(230, 1%, 90%);
+      color: hsl(230, 6%, 44%);
+      padding: 0.1em 0.4em;
+      border-radius: 0.3em;
+    }
+
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item > button:hover,
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item > button:focus,
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item > a:hover,
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item > a:focus,
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item > span:hover,
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item > span:focus {
+      background: hsl(230, 1%, 78%); /* custom: darken(--syntax-bg, 20%) */
+      color: hsl(230, 8%, 24%);
+    }
+
+    /* Line Highlight plugin overrides */
+    /* The highlighted line itself */
+    .line-highlight.line-highlight {
+      background: hsla(230, 8%, 24%, 0.05);
+    }
+
+    /* Default line numbers in Line Highlight plugin */
+    .line-highlight.line-highlight:before,
+    .line-highlight.line-highlight[data-end]:after {
+      background: hsl(230, 1%, 90%);
+      color: hsl(230, 8%, 24%);
+      padding: 0.1em 0.6em;
+      border-radius: 0.3em;
+      box-shadow: 0 2px 0 0 rgba(0, 0, 0, 0.2); /* same as Toolbar plugin default */
+    }
+
+    /* Hovering over a linkable line number (in the gutter area) */
+    /* Requires Line Numbers plugin as well */
+    pre[id].linkable-line-numbers.linkable-line-numbers
+      span.line-numbers-rows
+      > span:hover:before {
+      background-color: hsla(230, 8%, 24%, 0.05);
+    }
+
+    /* Line Numbers and Command Line plugins overrides */
+    /* Line separating gutter from coding area */
+    .line-numbers.line-numbers .line-numbers-rows,
+    .command-line .command-line-prompt {
+      border-right-color: hsla(230, 8%, 24%, 0.2);
+    }
+
+    /* Stuff in the gutter */
+    .line-numbers .line-numbers-rows > span:before,
+    .command-line .command-line-prompt > span:before {
+      color: hsl(230, 1%, 62%);
+    }
+
+    /* Match Braces plugin overrides */
+    /* Note: Outline colour is inherited from the braces */
+    .rainbow-braces .token.token.punctuation.brace-level-1,
+    .rainbow-braces .token.token.punctuation.brace-level-5,
+    .rainbow-braces .token.token.punctuation.brace-level-9 {
+      color: hsl(5, 74%, 59%);
+    }
+
+    .rainbow-braces .token.token.punctuation.brace-level-2,
+    .rainbow-braces .token.token.punctuation.brace-level-6,
+    .rainbow-braces .token.token.punctuation.brace-level-10 {
+      color: hsl(119, 34%, 47%);
+    }
+
+    .rainbow-braces .token.token.punctuation.brace-level-3,
+    .rainbow-braces .token.token.punctuation.brace-level-7,
+    .rainbow-braces .token.token.punctuation.brace-level-11 {
+      color: hsl(221, 87%, 60%);
+    }
+
+    .rainbow-braces .token.token.punctuation.brace-level-4,
+    .rainbow-braces .token.token.punctuation.brace-level-8,
+    .rainbow-braces .token.token.punctuation.brace-level-12 {
+      color: hsl(301, 63%, 40%);
+    }
+
+    /* Diff Highlight plugin overrides */
+    /* Taken from https://github.com/atom/github/blob/master/styles/variables.less */
+    pre.diff-highlight > code .token.token.deleted:not(.prefix),
+    pre > code.diff-highlight .token.token.deleted:not(.prefix) {
+      background-color: hsla(353, 100%, 66%, 0.15);
+    }
+
+    pre.diff-highlight > code .token.token.deleted:not(.prefix)::-moz-selection,
+    pre.diff-highlight
+      > code
+      .token.token.deleted:not(.prefix)
+      *::-moz-selection,
+    pre > code.diff-highlight .token.token.deleted:not(.prefix)::-moz-selection,
+    pre
+      > code.diff-highlight
+      .token.token.deleted:not(.prefix)
+      *::-moz-selection {
+      background-color: hsla(353, 95%, 66%, 0.25);
+    }
+
+    pre.diff-highlight > code .token.token.deleted:not(.prefix)::selection,
+    pre.diff-highlight > code .token.token.deleted:not(.prefix) *::selection,
+    pre > code.diff-highlight .token.token.deleted:not(.prefix)::selection,
+    pre > code.diff-highlight .token.token.deleted:not(.prefix) *::selection {
+      background-color: hsla(353, 95%, 66%, 0.25);
+    }
+
+    pre.diff-highlight > code .token.token.inserted:not(.prefix),
+    pre > code.diff-highlight .token.token.inserted:not(.prefix) {
+      background-color: hsla(137, 100%, 55%, 0.15);
+    }
+
+    pre.diff-highlight
+      > code
+      .token.token.inserted:not(.prefix)::-moz-selection,
+    pre.diff-highlight
+      > code
+      .token.token.inserted:not(.prefix)
+      *::-moz-selection,
+    pre
+      > code.diff-highlight
+      .token.token.inserted:not(.prefix)::-moz-selection,
+    pre
+      > code.diff-highlight
+      .token.token.inserted:not(.prefix)
+      *::-moz-selection {
+      background-color: hsla(135, 73%, 55%, 0.25);
+    }
+
+    pre.diff-highlight > code .token.token.inserted:not(.prefix)::selection,
+    pre.diff-highlight > code .token.token.inserted:not(.prefix) *::selection,
+    pre > code.diff-highlight .token.token.inserted:not(.prefix)::selection,
+    pre > code.diff-highlight .token.token.inserted:not(.prefix) *::selection {
+      background-color: hsla(135, 73%, 55%, 0.25);
+    }
+
+    /* Previewers plugin overrides */
+    /* Based on https://github.com/atom-community/atom-ide-datatip/blob/master/styles/atom-ide-datatips.less and https://github.com/atom/atom/blob/master/packages/one-light-ui */
+    /* Border around popup */
+    .prism-previewer.prism-previewer:before,
+    .prism-previewer-gradient.prism-previewer-gradient div {
+      border-color: hsl(0, 0, 95%);
+    }
+
+    /* Angle and time should remain as circles and are hence not included */
+    .prism-previewer-color.prism-previewer-color:before,
+    .prism-previewer-gradient.prism-previewer-gradient div,
+    .prism-previewer-easing.prism-previewer-easing:before {
+      border-radius: 0.3em;
+    }
+
+    /* Triangles pointing to the code */
+    .prism-previewer.prism-previewer:after {
+      border-top-color: hsl(0, 0, 95%);
+    }
+
+    .prism-previewer-flipped.prism-previewer-flipped.after {
+      border-bottom-color: hsl(0, 0, 95%);
+    }
+
+    /* Background colour within the popup */
+    .prism-previewer-angle.prism-previewer-angle:before,
+    .prism-previewer-time.prism-previewer-time:before,
+    .prism-previewer-easing.prism-previewer-easing {
+      background: hsl(0, 0%, 100%);
+    }
+
+    /* For angle, this is the positive area (eg. 90deg will display one quadrant in this colour) */
+    /* For time, this is the alternate colour */
+    .prism-previewer-angle.prism-previewer-angle circle,
+    .prism-previewer-time.prism-previewer-time circle {
+      stroke: hsl(230, 8%, 24%);
+      stroke-opacity: 1;
+    }
+
+    /* Stroke colours of the handle, direction point, and vector itself */
+    .prism-previewer-easing.prism-previewer-easing circle,
+    .prism-previewer-easing.prism-previewer-easing path,
+    .prism-previewer-easing.prism-previewer-easing line {
+      stroke: hsl(230, 8%, 24%);
+    }
+
+    /* Fill colour of the handle */
+    .prism-previewer-easing.prism-previewer-easing circle {
+      fill: transparent;
+    }
+  }
+}
+
+@media (prefers-color-scheme: dark) {
+  .prose {
+    .token.comment,
+    .token.prolog,
+    .token.cdata {
+      color: hsl(220, 10%, 40%);
+    }
+
+    .token.doctype,
+    .token.punctuation,
+    .token.entity {
+      color: hsl(220, 14%, 71%);
+    }
+
+    .token.attr-name,
+    .token.class-name,
+    .token.boolean,
+    .token.constant,
+    .token.number,
+    .token.atrule {
+      color: hsl(29, 54%, 61%);
+    }
+
+    .token.keyword {
+      color: hsl(286, 60%, 67%);
+    }
+
+    .token.property,
+    .token.tag,
+    .token.symbol,
+    .token.deleted,
+    .token.important {
+      color: hsl(355, 65%, 65%);
+    }
+
+    .token.selector,
+    .token.string,
+    .token.char,
+    .token.builtin,
+    .token.inserted,
+    .token.regex,
+    .token.attr-value,
+    .token.attr-value > .token.punctuation {
+      color: hsl(95, 38%, 62%);
+    }
+
+    .token.variable,
+    .token.operator,
+    .token.function {
+      color: hsl(207, 82%, 66%);
+    }
+
+    .token.url {
+      color: hsl(187, 47%, 55%);
+    }
+
+    /* HTML overrides */
+    .token.attr-value > .token.punctuation.attr-equals,
+    .token.special-attr > .token.attr-value > .token.value.css {
+      color: hsl(220, 14%, 71%);
+    }
+
+    /* CSS overrides */
+    .language-css .token.selector {
+      color: hsl(355, 65%, 65%);
+    }
+
+    .language-css .token.property {
+      color: hsl(220, 14%, 71%);
+    }
+
+    .language-css .token.function,
+    .language-css .token.url > .token.function {
+      color: hsl(187, 47%, 55%);
+    }
+
+    .language-css .token.url > .token.string.url {
+      color: hsl(95, 38%, 62%);
+    }
+
+    .language-css .token.important,
+    .language-css .token.atrule .token.rule {
+      color: hsl(286, 60%, 67%);
+    }
+
+    /* JS overrides */
+    .language-javascript .token.operator {
+      color: hsl(286, 60%, 67%);
+    }
+
+    .language-javascript
+      .token.template-string
+      > .token.interpolation
+      > .token.interpolation-punctuation.punctuation {
+      color: hsl(5, 48%, 51%);
+    }
+
+    /* JSON overrides */
+    .language-json .token.operator {
+      color: hsl(220, 14%, 71%);
+    }
+
+    .language-json .token.null.keyword {
+      color: hsl(29, 54%, 61%);
+    }
+
+    /* MD overrides */
+    .language-markdown .token.url,
+    .language-markdown .token.url > .token.operator,
+    .language-markdown .token.url-reference.url > .token.string {
+      color: hsl(220, 14%, 71%);
+    }
+
+    .language-markdown .token.url > .token.content {
+      color: hsl(207, 82%, 66%);
+    }
+
+    .language-markdown .token.url > .token.url,
+    .language-markdown .token.url-reference.url {
+      color: hsl(187, 47%, 55%);
+    }
+
+    .language-markdown .token.blockquote.punctuation,
+    .language-markdown .token.hr.punctuation {
+      color: hsl(220, 10%, 40%);
+      font-style: italic;
+    }
+
+    .language-markdown .token.code-snippet {
+      color: hsl(95, 38%, 62%);
+    }
+
+    .language-markdown .token.bold .token.content {
+      color: hsl(29, 54%, 61%);
+    }
+
+    .language-markdown .token.italic .token.content {
+      color: hsl(286, 60%, 67%);
+    }
+
+    .language-markdown .token.strike .token.content,
+    .language-markdown .token.strike .token.punctuation,
+    .language-markdown .token.list.punctuation,
+    .language-markdown .token.title.important > .token.punctuation {
+      color: hsl(355, 65%, 65%);
+    }
+
+    /* General */
+    .token.bold {
+      font-weight: bold;
+    }
+
+    .token.comment,
+    .token.italic {
+      font-style: italic;
+    }
+
+    .token.entity {
+      cursor: help;
+    }
+
+    .token.namespace {
+      opacity: 0.8;
+    }
+
+    /* Plugin overrides */
+    /* Selectors should have higher specificity than those in the plugins' default stylesheets */
+
+    /* Show Invisibles plugin overrides */
+    .token.token.tab:not(:empty):before,
+    .token.token.cr:before,
+    .token.token.lf:before,
+    .token.token.space:before {
+      color: hsla(220, 14%, 71%, 0.15);
+      text-shadow: none;
+    }
+
+    /* Toolbar plugin overrides */
+    /* Space out all buttons and move them away from the right edge of the code block */
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item {
+      margin-right: 0.4em;
+    }
+
+    /* Styling the buttons */
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item > button,
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item > a,
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item > span {
+      background: hsl(220, 13%, 26%);
+      color: hsl(220, 9%, 55%);
+      padding: 0.1em 0.4em;
+      border-radius: 0.3em;
+    }
+
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item > button:hover,
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item > button:focus,
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item > a:hover,
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item > a:focus,
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item > span:hover,
+    div.code-toolbar > .toolbar.toolbar > .toolbar-item > span:focus {
+      background: hsl(220, 13%, 28%);
+      color: hsl(220, 14%, 71%);
+    }
+
+    /* Line Highlight plugin overrides */
+    /* The highlighted line itself */
+    .line-highlight.line-highlight {
+      background: hsla(220, 100%, 80%, 0.04);
+    }
+
+    /* Default line numbers in Line Highlight plugin */
+    .line-highlight.line-highlight:before,
+    .line-highlight.line-highlight[data-end]:after {
+      background: hsl(220, 13%, 26%);
+      color: hsl(220, 14%, 71%);
+      padding: 0.1em 0.6em;
+      border-radius: 0.3em;
+      box-shadow: 0 2px 0 0 rgba(0, 0, 0, 0.2); /* same as Toolbar plugin default */
+    }
+
+    /* Hovering over a linkable line number (in the gutter area) */
+    /* Requires Line Numbers plugin as well */
+    pre[id].linkable-line-numbers.linkable-line-numbers
+      span.line-numbers-rows
+      > span:hover:before {
+      background-color: hsla(220, 100%, 80%, 0.04);
+    }
+
+    /* Line Numbers and Command Line plugins overrides */
+    /* Line separating gutter from coding area */
+    .line-numbers.line-numbers .line-numbers-rows,
+    .command-line .command-line-prompt {
+      border-right-color: hsla(220, 14%, 71%, 0.15);
+    }
+
+    /* Stuff in the gutter */
+    .line-numbers .line-numbers-rows > span:before,
+    .command-line .command-line-prompt > span:before {
+      color: hsl(220, 14%, 45%);
+    }
+
+    /* Match Braces plugin overrides */
+    /* Note: Outline colour is inherited from the braces */
+    .rainbow-braces .token.token.punctuation.brace-level-1,
+    .rainbow-braces .token.token.punctuation.brace-level-5,
+    .rainbow-braces .token.token.punctuation.brace-level-9 {
+      color: hsl(355, 65%, 65%);
+    }
+
+    .rainbow-braces .token.token.punctuation.brace-level-2,
+    .rainbow-braces .token.token.punctuation.brace-level-6,
+    .rainbow-braces .token.token.punctuation.brace-level-10 {
+      color: hsl(95, 38%, 62%);
+    }
+
+    .rainbow-braces .token.token.punctuation.brace-level-3,
+    .rainbow-braces .token.token.punctuation.brace-level-7,
+    .rainbow-braces .token.token.punctuation.brace-level-11 {
+      color: hsl(207, 82%, 66%);
+    }
+
+    .rainbow-braces .token.token.punctuation.brace-level-4,
+    .rainbow-braces .token.token.punctuation.brace-level-8,
+    .rainbow-braces .token.token.punctuation.brace-level-12 {
+      color: hsl(286, 60%, 67%);
+    }
+
+    /* Diff Highlight plugin overrides */
+    /* Taken from https://github.com/atom/github/blob/master/styles/variables.less */
+    pre.diff-highlight > code .token.token.deleted:not(.prefix),
+    pre > code.diff-highlight .token.token.deleted:not(.prefix) {
+      background-color: hsla(353, 100%, 66%, 0.15);
+    }
+
+    pre.diff-highlight > code .token.token.deleted:not(.prefix)::-moz-selection,
+    pre.diff-highlight
+      > code
+      .token.token.deleted:not(.prefix)
+      *::-moz-selection,
+    pre > code.diff-highlight .token.token.deleted:not(.prefix)::-moz-selection,
+    pre
+      > code.diff-highlight
+      .token.token.deleted:not(.prefix)
+      *::-moz-selection {
+      background-color: hsla(353, 95%, 66%, 0.25);
+    }
+
+    pre.diff-highlight > code .token.token.deleted:not(.prefix)::selection,
+    pre.diff-highlight > code .token.token.deleted:not(.prefix) *::selection,
+    pre > code.diff-highlight .token.token.deleted:not(.prefix)::selection,
+    pre > code.diff-highlight .token.token.deleted:not(.prefix) *::selection {
+      background-color: hsla(353, 95%, 66%, 0.25);
+    }
+
+    pre.diff-highlight > code .token.token.inserted:not(.prefix),
+    pre > code.diff-highlight .token.token.inserted:not(.prefix) {
+      background-color: hsla(137, 100%, 55%, 0.15);
+    }
+
+    pre.diff-highlight
+      > code
+      .token.token.inserted:not(.prefix)::-moz-selection,
+    pre.diff-highlight
+      > code
+      .token.token.inserted:not(.prefix)
+      *::-moz-selection,
+    pre
+      > code.diff-highlight
+      .token.token.inserted:not(.prefix)::-moz-selection,
+    pre
+      > code.diff-highlight
+      .token.token.inserted:not(.prefix)
+      *::-moz-selection {
+      background-color: hsla(135, 73%, 55%, 0.25);
+    }
+
+    pre.diff-highlight > code .token.token.inserted:not(.prefix)::selection,
+    pre.diff-highlight > code .token.token.inserted:not(.prefix) *::selection,
+    pre > code.diff-highlight .token.token.inserted:not(.prefix)::selection,
+    pre > code.diff-highlight .token.token.inserted:not(.prefix) *::selection {
+      background-color: hsla(135, 73%, 55%, 0.25);
+    }
+
+    /* Previewers plugin overrides */
+    /* Based on https://github.com/atom-community/atom-ide-datatip/blob/master/styles/atom-ide-datatips.less and https://github.com/atom/atom/blob/master/packages/one-dark-ui */
+    /* Border around popup */
+    .prism-previewer.prism-previewer:before,
+    .prism-previewer-gradient.prism-previewer-gradient div {
+      border-color: hsl(224, 13%, 17%);
+    }
+
+    /* Angle and time should remain as circles and are hence not included */
+    .prism-previewer-color.prism-previewer-color:before,
+    .prism-previewer-gradient.prism-previewer-gradient div,
+    .prism-previewer-easing.prism-previewer-easing:before {
+      border-radius: 0.3em;
+    }
+
+    /* Triangles pointing to the code */
+    .prism-previewer.prism-previewer:after {
+      border-top-color: hsl(224, 13%, 17%);
+    }
+
+    .prism-previewer-flipped.prism-previewer-flipped.after {
+      border-bottom-color: hsl(224, 13%, 17%);
+    }
+
+    /* Background colour within the popup */
+    .prism-previewer-angle.prism-previewer-angle:before,
+    .prism-previewer-time.prism-previewer-time:before,
+    .prism-previewer-easing.prism-previewer-easing {
+      background: hsl(219, 13%, 22%);
+    }
+
+    /* For angle, this is the positive area (eg. 90deg will display one quadrant in this colour) */
+    /* For time, this is the alternate colour */
+    .prism-previewer-angle.prism-previewer-angle circle,
+    .prism-previewer-time.prism-previewer-time circle {
+      stroke: hsl(220, 14%, 71%);
+      stroke-opacity: 1;
+    }
+
+    /* Stroke colours of the handle, direction point, and vector itself */
+    .prism-previewer-easing.prism-previewer-easing circle,
+    .prism-previewer-easing.prism-previewer-easing path,
+    .prism-previewer-easing.prism-previewer-easing line {
+      stroke: hsl(220, 14%, 71%);
+    }
+
+    /* Fill colour of the handle */
+    .prism-previewer-easing.prism-previewer-easing circle {
+      fill: transparent;
+    }
+  }
+}
+
+.prose pre {
+  contain: layout style;
+}
+
+/* Or more aggressively */
+.prose pre code {
+  contain: layout style paint;
+}

 /* messaging-style typing indicator animation */
@keyframes typing {
--- a/app/ui/app/src/lib/config.ts
+++ b/app/ui/app/src/lib/config.ts
@@ -1,13 +0,0 @@
-// API configuration
-const DEV_API_URL = "http://127.0.0.1:3001";
-
-// Base URL for fetch API calls (can be relative in production)
-export const API_BASE = import.meta.env.DEV ? DEV_API_URL : "";
-
-// Full host URL for Ollama client (needs full origin in production)
-export const OLLAMA_HOST = import.meta.env.DEV
-  ? DEV_API_URL
-  : window.location.origin;
-
-export const OLLAMA_DOT_COM =
-  import.meta.env.VITE_OLLAMA_DOT_COM_URL || "https://ollama.com";
--- a/app/ui/app/src/lib/highlighter.ts
+++ b/app/ui/app/src/lib/highlighter.ts
@@ -1,157 +0,0 @@
-import { createHighlighter } from "shiki";
-import type { ThemeRegistration } from "shiki";
-
-const oneLightTheme: ThemeRegistration = {
-  name: "one-light",
-  type: "light",
-  colors: {
-    "editor.background": "#fafafa",
-    "editor.foreground": "#383a42",
-  },
-  tokenColors: [
-    {
-      scope: ["comment", "punctuation.definition.comment"],
-      settings: { foreground: "#a0a1a7" },
-    },
-    {
-      scope: ["keyword", "storage.type", "storage.modifier"],
-      settings: { foreground: "#a626a4" },
-    },
-    { scope: ["string", "string.quoted"], settings: { foreground: "#50a14f" } },
-    {
-      scope: ["function", "entity.name.function", "support.function"],
-      settings: { foreground: "#4078f2" },
-    },
-    {
-      scope: [
-        "constant.numeric",
-        "constant.language",
-        "constant.character",
-        "number",
-      ],
-      settings: { foreground: "#c18401" },
-    },
-    {
-      scope: ["variable", "support.variable"],
-      settings: { foreground: "#e45649" },
-    },
-    {
-      scope: ["entity.name.tag", "entity.name.type", "entity.name.class"],
-      settings: { foreground: "#e45649" },
-    },
-    {
-      scope: ["entity.other.attribute-name"],
-      settings: { foreground: "#c18401" },
-    },
-    {
-      scope: ["keyword.operator", "operator"],
-      settings: { foreground: "#a626a4" },
-    },
-    { scope: ["punctuation"], settings: { foreground: "#383a42" } },
-    {
-      scope: ["markup.heading"],
-      settings: { foreground: "#e45649", fontStyle: "bold" },
-    },
-    {
-      scope: ["markup.bold"],
-      settings: { foreground: "#c18401", fontStyle: "bold" },
-    },
-    {
-      scope: ["markup.italic"],
-      settings: { foreground: "#a626a4", fontStyle: "italic" },
-    },
-  ],
-};
-
-const oneDarkTheme: ThemeRegistration = {
-  name: "one-dark",
-  type: "dark",
-  colors: {
-    "editor.background": "#282c34",
-    "editor.foreground": "#abb2bf",
-  },
-  tokenColors: [
-    {
-      scope: ["comment", "punctuation.definition.comment"],
-      settings: { foreground: "#5c6370" },
-    },
-    {
-      scope: ["keyword", "storage.type", "storage.modifier"],
-      settings: { foreground: "#c678dd" },
-    },
-    { scope: ["string", "string.quoted"], settings: { foreground: "#98c379" } },
-    {
-      scope: ["function", "entity.name.function", "support.function"],
-      settings: { foreground: "#61afef" },
-    },
-    {
-      scope: [
-        "constant.numeric",
-        "constant.language",
-        "constant.character",
-        "number",
-      ],
-      settings: { foreground: "#d19a66" },
-    },
-    {
-      scope: ["variable", "support.variable"],
-      settings: { foreground: "#e06c75" },
-    },
-    {
-      scope: ["entity.name.tag", "entity.name.type", "entity.name.class"],
-      settings: { foreground: "#e06c75" },
-    },
-    {
-      scope: ["entity.other.attribute-name"],
-      settings: { foreground: "#d19a66" },
-    },
-    {
-      scope: ["keyword.operator", "operator"],
-      settings: { foreground: "#c678dd" },
-    },
-    { scope: ["punctuation"], settings: { foreground: "#abb2bf" } },
-    {
-      scope: ["markup.heading"],
-      settings: { foreground: "#e06c75", fontStyle: "bold" },
-    },
-    {
-      scope: ["markup.bold"],
-      settings: { foreground: "#d19a66", fontStyle: "bold" },
-    },
-    {
-      scope: ["markup.italic"],
-      settings: { foreground: "#c678dd", fontStyle: "italic" },
-    },
-  ],
-};
-
-export let highlighter: Awaited<ReturnType<typeof createHighlighter>> | null =
-  null;
-
-export const highlighterPromise = createHighlighter({
-  themes: [oneLightTheme, oneDarkTheme],
-  langs: [
-    "javascript",
-    "typescript",
-    "python",
-    "bash",
-    "shell",
-    "json",
-    "html",
-    "css",
-    "tsx",
-    "jsx",
-    "go",
-    "rust",
-    "java",
-    "c",
-    "cpp",
-    "sql",
-    "swift",
-    "yaml",
-    "markdown",
-  ],
-}).then((h) => {
-  highlighter = h;
-  return h;
-});
--- a/app/ui/app/src/lib/ollama-client.ts
+++ b/app/ui/app/src/lib/ollama-client.ts
@@ -1,5 +1,4 @@
 import { Ollama } from "ollama/browser";
-import { OLLAMA_HOST } from "./config";

 let _ollamaClient: Ollama | null = null;

@@ -7,7 +6,7 @@ export const ollamaClient = new Proxy({} as Ollama, {
  get(_target, prop) {
    if (!_ollamaClient) {
      _ollamaClient = new Ollama({
-        host: OLLAMA_HOST,
+        host: window.location.origin,
      });
    }
    const value = _ollamaClient[prop as keyof Ollama];
--- a/app/ui/app/src/main.tsx
+++ b/app/ui/app/src/main.tsx
@@ -5,6 +5,13 @@ import { QueryClient, QueryClientProvider } from "@tanstack/react-query";
 import { routeTree } from "./routeTree.gen";
 import { fetchUser } from "./api";
 import { StreamingProvider } from "./contexts/StreamingContext";
+import { User } from "@/gotypes";
+
+declare global {
+  interface Window {
+    __initialUserDataPromise?: Promise<User | null>;
+  }
+}

 const queryClient = new QueryClient({
  defaultOptions: {
@@ -17,11 +24,27 @@ const queryClient = new QueryClient({
  },
 });

-fetchUser().then((userData) => {
-  if (userData) {
+// Track initial user data fetch
+let initialUserDataPromise: Promise<User | null> | null = null;
+
+// Initialize user data on app startup
+const initializeUserData = async () => {
+  try {
+    const userData = await fetchUser();
    queryClient.setQueryData(["user"], userData);
+    return userData;
+  } catch (error) {
+    console.error("Error initializing user data:", error);
+    queryClient.setQueryData(["user"], null);
+    return null;
  }
-});
+};
+
+// Start initialization immediately and track the promise
+initialUserDataPromise = initializeUserData();
+
+// Export the promise so hooks can await it
+window.__initialUserDataPromise = initialUserDataPromise;

 const router = createRouter({
  routeTree,
--- a/app/ui/app/src/utils/fileValidation.test.ts
+++ b/app/ui/app/src/utils/fileValidation.test.ts
@@ -1,97 +0,0 @@
-import { describe, it, expect } from "vitest";
-import { IMAGE_EXTENSIONS, validateFile } from "./fileValidation";
-
-describe("fileValidation", () => {
-  describe("IMAGE_EXTENSIONS", () => {
-    it("should include all supported image formats including WebP", () => {
-      expect(IMAGE_EXTENSIONS).toContain("png");
-      expect(IMAGE_EXTENSIONS).toContain("jpg");
-      expect(IMAGE_EXTENSIONS).toContain("jpeg");
-      expect(IMAGE_EXTENSIONS).toContain("webp");
-    });
-  });
-
-  describe("validateFile", () => {
-    const createMockFile = (
-      name: string,
-      size: number,
-      type: string,
-    ): File => {
-      const blob = new Blob(["test content"], { type });
-      return new File([blob], name, { type });
-    };
-
-    it("should accept WebP images when vision capability is enabled", () => {
-      const file = createMockFile("test.webp", 1024, "image/webp");
-      const result = validateFile(file, {
-        hasVisionCapability: true,
-      });
-      expect(result.valid).toBe(true);
-    });
-
-    it("should reject WebP images when vision capability is disabled", () => {
-      const file = createMockFile("test.webp", 1024, "image/webp");
-      const result = validateFile(file, {
-        hasVisionCapability: false,
-      });
-      expect(result.valid).toBe(false);
-      expect(result.error).toBe("This model does not support images");
-    });
-
-    it("should accept PNG images when vision capability is enabled", () => {
-      const file = createMockFile("test.png", 1024, "image/png");
-      const result = validateFile(file, {
-        hasVisionCapability: true,
-      });
-      expect(result.valid).toBe(true);
-    });
-
-    it("should accept JPEG images when vision capability is enabled", () => {
-      const file = createMockFile("test.jpg", 1024, "image/jpeg");
-      const result = validateFile(file, {
-        hasVisionCapability: true,
-      });
-      expect(result.valid).toBe(true);
-    });
-
-    it("should reject files that are too large", () => {
-      // Create a file with size property set correctly
-      const largeSize = 11 * 1024 * 1024; // 11MB
-      const content = new Uint8Array(largeSize);
-      const blob = new Blob([content], { type: "image/webp" });
-      const file = new File([blob], "large.webp", { type: "image/webp" });
-      
-      const result = validateFile(file, {
-        hasVisionCapability: true,
-        maxFileSize: 10, // 10MB limit
-      });
-      expect(result.valid).toBe(false);
-      expect(result.error).toBe("File too large");
-    });
-
-    it("should reject unsupported file types", () => {
-      const file = createMockFile("test.xyz", 1024, "application/xyz");
-      const result = validateFile(file, {
-        hasVisionCapability: true,
-      });
-      expect(result.valid).toBe(false);
-      expect(result.error).toBe("File type not supported");
-    });
-
-    it("should respect custom validators", () => {
-      const file = createMockFile("test.webp", 1024, "image/webp");
-      const result = validateFile(file, {
-        hasVisionCapability: true,
-        customValidator: () => ({
-          valid: false,
-          error: "Custom error",
-        }),
-      });
-      expect(result.valid).toBe(false);
-      expect(result.error).toBe("Custom error");
-    });
-  });
-
-  // Note: processFiles tests are skipped because FileReader is not available in the Node.js test environment
-  // These functions are tested in browser environment via integration tests
-});
--- a/app/ui/app/src/utils/fileValidation.ts
+++ b/app/ui/app/src/utils/fileValidation.ts
@@ -41,7 +41,7 @@ export const TEXT_FILE_EXTENSIONS = [
  "rtf",
 ];

-export const IMAGE_EXTENSIONS = ["png", "jpg", "jpeg", "webp"];
+export const IMAGE_EXTENSIONS = ["png", "jpg", "jpeg"];

 export interface FileValidationOptions {
  maxFileSize?: number; // in MB
--- a/app/ui/app/src/utils/processStreamingMarkdown.ts
+++ b/app/ui/app/src/utils/processStreamingMarkdown.ts
@@ -0,0 +1,24 @@
+import { remark } from "remark";
+import remarkStringify from "remark-stringify";
+import remarkStreamingMarkdown from "./remarkStreamingMarkdown";
+
+/**
+ * Process markdown content for streaming display using the remark plugin.
+ * This is primarily used for testing the remark plugin with string inputs/outputs.
+ */
+export function processStreamingMarkdown(content: string): string {
+  if (!content) return content;
+
+  const result = remark()
+    .use(remarkStreamingMarkdown, { debug: false })
+    .use(remarkStringify)
+    .processSync(content);
+
+  // remove trailing newline to keep tests cleaner
+  let output = result.toString();
+  if (output.endsWith("\n")) {
+    output = output.slice(0, -1);
+  }
+
+  return output;
+}
--- a/app/ui/app/src/utils/remarkStreamingMarkdown.ts
+++ b/app/ui/app/src/utils/remarkStreamingMarkdown.ts
@@ -0,0 +1,447 @@
+import { parents, type Proxy } from "unist-util-parents";
+import type { Plugin } from "unified";
+import type {
+  Emphasis,
+  Node,
+  Parent,
+  Root,
+  RootContent,
+  Text,
+  Strong,
+  PhrasingContent,
+  Paragraph,
+} from "mdast";
+import { u } from "unist-builder";
+
+declare module "unist" {
+  interface Node {
+    /** Added by `unist-util-parents` (or your own walk). */
+    parent?: Proxy & Parent;
+  }
+}
+
+// interface SimpleTextRule {
+//   pattern: RegExp;
+//   transform: (matches: RegExpExecArray[], lastNode: Proxy) => void;
+// }
+
+// const simpleTextRules: SimpleTextRule[] = [
+//   // TODO(drifkin): generalize this for `__`/`_`/`~~`/`~` etc.
+//   {
+//     pattern: /(\*\*)(?=\S|$)/g,
+//     transform: (matchesIterator, lastNode) => {
+//       const textNode = lastNode.node as Text;
+
+//       const matches = [...matchesIterator];
+//       const lastMatch = matches[matches.length - 1];
+//       const origValue = textNode.value;
+//       const start = lastMatch.index;
+//       const sep = lastMatch[1];
+
+//       const before = origValue.slice(0, start);
+//       const after = origValue.slice(start + sep.length);
+
+//       if (lastNode.parent) {
+//         const index = (lastNode.parent.node as Parent).children.indexOf(
+//           lastNode.node as RootContent,
+//         );
+//         const shouldRemove = before.length === 0;
+//         if (!shouldRemove) {
+//           textNode.value = before;
+//         }
+
+//         const newNode = u("strong", {
+//           children: [u("text", { value: after })],
+//         });
+//         (lastNode.parent.node as Parent).children.splice(
+//           index + (shouldRemove ? 0 : 1),
+//           shouldRemove ? 1 : 0,
+//           newNode,
+//         );
+//       }
+//     },
+//   },
+// ];
+
+interface Options {
+  debug?: boolean;
+  onLastNode?: (info: LastNodeInfo) => void;
+}
+
+export interface LastNodeInfo {
+  path: string[];
+  type: string;
+  value?: string;
+  lastChars?: string;
+  fullNode: Node;
+}
+
+/**
+ * Removes `child` from `parent` in-place.
+ * @returns `true` if the child was found and removed; `false` otherwise.
+ */
+export function removeChildFromParent(
+  child: RootContent,
+  parent: Node,
+): boolean {
+  if (!isParent(parent)) return false; // parent isn’t a Parent → nothing to do
+
+  const idx = parent.children.indexOf(child);
+  if (idx < 0) return false; // not a child → nothing to remove
+
+  parent.children.splice(idx, 1);
+  return true; // removal successful
+}
+
+/** Narrow a generic `Node` to a `Parent` (i.e. one that really has children). */
+function isParent(node: Node): node is Parent {
+  // A `Parent` always has a `children` array; make sure it's an array first.
+  return Array.isArray((node as Partial<Parent>).children);
+}
+
+/**
+ * Follow “last-child” pointers until you reach a leaf.
+ * Returns the right-most, deepest node in source order.
+ */
+export function findRightmostDeepestNode(root: Node): Node {
+  let current: Node = root;
+
+  // While the current node *is* a Parent and has at least one child…
+  while (isParent(current) && current.children.length > 0) {
+    const lastIndex = current.children.length - 1;
+    current = current.children[lastIndex];
+  }
+
+  return current; // Leaf: no further children
+}
+
+const remarkStreamingMarkdown: Plugin<[Options?], Root> = () => {
+  return (tree) => {
+    const treeWithParents = parents(tree);
+    const lastNode = findRightmostDeepestNode(treeWithParents) as Proxy;
+
+    const parentNode = lastNode.parent;
+    const grandparentNode = parentNode?.parent;
+
+    let ruleMatched = false;
+
+    // handling `* *` -> ``
+    //
+    // if the last node is part of a <list item (otherwise empty)> ->
+    // <list (otherwise empty)> -> <list item (last node, empty)>, then we need to
+    // remove everything up to and including the first list item. This happens
+    // when we have `* *`, which can become a bolded list item OR a horizontal
+    // line
+    if (
+      lastNode.type === "listItem" &&
+      parentNode &&
+      grandparentNode &&
+      parentNode.type === "list" &&
+      grandparentNode.type === "listItem" &&
+      parentNode.children.length === 1 &&
+      grandparentNode.children.length === 1
+    ) {
+      ruleMatched = true;
+      if (grandparentNode.parent) {
+        removeChildFromParent(
+          grandparentNode.node as RootContent,
+          grandparentNode.parent.node,
+        );
+      }
+      // Handle `*` -> ``:
+      //
+      // if the last node is just an empty list item, we need to remove it
+      // because it could become something else (e.g., a horizontal line)
+    } else if (
+      lastNode.type === "listItem" &&
+      parentNode &&
+      parentNode.type === "list"
+    ) {
+      ruleMatched = true;
+      removeChildFromParent(lastNode.node as RootContent, parentNode.node);
+    } else if (lastNode.type === "thematicBreak") {
+      ruleMatched = true;
+      const parent = lastNode.parent;
+      if (parent) {
+        removeChildFromParent(lastNode.node as RootContent, parent.node);
+      }
+    } else if (lastNode.type === "text") {
+      const textNode = lastNode.node as Text;
+      if (textNode.value.endsWith("**")) {
+        ruleMatched = true;
+        textNode.value = textNode.value.slice(0, -2);
+        // if there's a newline then a number, this is very very likely a
+        // numbered list item. Let's just hide it until the period comes (or
+        // other text disambiguates it)
+      } else {
+        const match = textNode.value.match(/^([0-9]+)$/m);
+        if (match) {
+          const number = match[1];
+          textNode.value = textNode.value.slice(0, -number.length - 1);
+          ruleMatched = true;
+          // if the text node is now empty, then we might want to remove other
+          // elements, like a now-empty containing paragraph, or a break that
+          // might disappear once more tokens come in
+          if (textNode.value.length === 0) {
+            if (
+              lastNode.parent?.type === "paragraph" &&
+              lastNode.parent.children.length === 1
+            ) {
+              // remove the whole paragraph if it's now empty (otherwise it'll
+              // cause an extra newline that might not last)
+              removeChildFromParent(
+                lastNode.parent.node as Paragraph,
+                lastNode.parent.parent?.node as Node,
+              );
+            } else {
+              const prev = prevSibling(lastNode);
+              if (prev?.type === "break") {
+                removeChildFromParent(
+                  prev.node as RootContent,
+                  lastNode.parent?.node as Node,
+                );
+                removeChildFromParent(
+                  lastNode.node as RootContent,
+                  lastNode.parent?.node as Node,
+                );
+              }
+            }
+          }
+        }
+      }
+    }
+
+    if (ruleMatched) {
+      return tree;
+    }
+
+    // we need to
+    // a case like
+    //     - *def `abc` [abc **def**](abc)*
+    // is pretty tricky, because if we land just after def, then we actually
+    // have two separate tags to process at two different parents. Maybe we
+    // need to keep iterating up until we find a paragraph, but process each
+    // parent on the way up. Hmm, well actually after `def` we won't even be a proper link yet
+    // TODO(drifkin): it's really if the last node's parent is a paragraph, for which the following is a sub-cas where the lastNode is a text node.
+    // And instead of just processing simple text rules, they need to operate on the whole paragraph
+    // like `**[abc](def)` needs to become `**[abc](def)**`
+
+    // if we're just text at the end, then we should remove some ambiguous characters
+
+    if (lastNode.parent) {
+      const didChange = processParent(lastNode.parent as Parent & Proxy);
+      if (didChange) {
+        // TODO(drifkin): need to fix up the tree, but not sure lastNode will still exist? Check all the transforms to see if it's safe to find the last node again
+        //
+        // need to regen the tree w/ parents since reparenting could've happened
+        // treeWithParents = parents(tree);
+      }
+    }
+
+    const grandparent = lastNode.parent?.parent;
+    // TODO(drifkin): let's go arbitrarily high up the tree, but limiting it
+    // to 2 levels for now until I think more about the stop condition
+    if (grandparent) {
+      processParent(grandparent as Parent & Proxy);
+    }
+
+    // console.log("ruleMatched", ruleMatched);
+
+    // } else if (lastNode.parent?.type === "paragraph") {
+    //   console.log("!!! paragraph");
+    //   console.log("lastNode.parent", lastNode.parent);
+
+    //   // Handle `**abc*` -> `**abc**`:
+    //   // We detect this when the last child is an emphasis node, and it's preceded by a text node that ends with `*`
+    //   const paragraph = lastNode.parent as Proxy & Paragraph;
+    //   if (paragraph.children.length >= 2) {
+    //     const lastChild = paragraph.children[paragraph.children.length - 1];
+    //     if (lastChild.type === "emphasis") {
+    //       const sibling = paragraph.children[paragraph.children.length - 2];
+    //       if (sibling.type === "text") {
+    //         const siblingText = sibling as Text & Proxy;
+    //         if (siblingText.value.endsWith("*")) {
+    //           ruleMatched = true;
+    //           const textNode = (lastNode as Proxy).node as Text;
+    //           textNode.value = textNode.value.slice(0, -1);
+    //           paragraph.node.type = "strong";
+    //         }
+    //       }
+    //     }
+    //   }
+    // } else if (lastNode.type === "text") {
+    //   // Handle `**abc*` -> `**abc**`:
+    //   //
+    //   // this gets parsed as a text node ending in `*` followed by an emphasis
+    //   // node. So if we're in text, we need to check if our parent is emphasis,
+    //   // and then get our parent's sibling before it and check if it ends with
+    //   // `*`
+    //   const parent = lastNode.parent;
+    //   if (parent && parent.type === "emphasis") {
+    //     const grandparent = parent.parent;
+    //     if (grandparent) {
+    //       const index = (grandparent.node as Parent).children.indexOf(
+    //         parent.node as RootContent,
+    //       );
+    //       if (index > 0) {
+    //         const prevNode = grandparent.children[index - 1];
+    //         if (
+    //           prevNode.type === "text" &&
+    //           (prevNode as Text).value.endsWith("*")
+    //         ) {
+    //           ruleMatched = true;
+    //           const textNode = (prevNode as Proxy).node as Text;
+    //           textNode.value = textNode.value.slice(0, -1);
+    //           parent.node.type = "strong";
+    //         }
+    //       }
+    //     }
+    // }
+
+    //   if (!ruleMatched) {
+    //     // if the last node is just text, then we process it in order to fix up certain unclosed items
+    //     // e.g., `**abc` -> `**abc**`
+    //     const textNode = lastNode.node as Text;
+    //     for (const rule of simpleTextRules) {
+    //       const matchesIterator = textNode.value.matchAll(rule.pattern);
+    //       const matches = [...matchesIterator];
+    //       if (matches.length > 0) {
+    //         rule.transform(matches, lastNode);
+    //         ruleMatched = true;
+    //         break;
+    //       }
+    //     }
+    //   }
+    // } else if (!ruleMatched) {
+    //   // console.log("no rule matched", lastNode);
+    // }
+
+    return tree;
+  };
+};
+
+function processParent(parent: Parent & Proxy): boolean {
+  if (parent.type === "emphasis") {
+    // Handle `**abc*` -> `**abc**`:
+    // We detect this when we end with an emphasis node, and it's preceded by
+    // a text node that ends with `*`
+    // TODO(drifkin): the last node can be more deeply nested (e.g., a code
+    // literal in a link), so we probably need to walk up the tree until we
+    // find an emphasis node or a block? For now we'll just go up one layer to
+    // catch the most common cases
+    const emphasisNode = parent as Emphasis & Proxy;
+    const grandparent = emphasisNode.parent;
+    if (grandparent) {
+      const indexOfEmphasisNode = (grandparent.node as Parent).children.indexOf(
+        emphasisNode.node as RootContent,
+      );
+      if (indexOfEmphasisNode >= 0) {
+        const nodeBefore = grandparent.children[indexOfEmphasisNode - 1] as
+          | (Node & Proxy)
+          | undefined;
+        if (nodeBefore?.type === "text") {
+          const textNode = nodeBefore.node as Text;
+          if (textNode.value.endsWith("*")) {
+            const strBefore = textNode.value.slice(0, -1);
+            textNode.value = strBefore;
+            const strongNode = u("strong", {
+              children: emphasisNode.children,
+            });
+            (grandparent.node as Parent).children.splice(
+              indexOfEmphasisNode,
+              1,
+              strongNode,
+            );
+            return true;
+          }
+        }
+      }
+    }
+  }
+
+  // Let's check if we have any bold items to close
+  for (let i = parent.children.length - 1; i >= 0; i--) {
+    const child = parent.children[i];
+    if (child.type === "text") {
+      const textNode = child as Text & Proxy;
+      const sep = "**";
+      const index = textNode.value.lastIndexOf(sep);
+      if (index >= 0) {
+        let isValidOpening = false;
+        if (index + sep.length < textNode.value.length) {
+          const charAfter = textNode.value[index + sep.length];
+          if (!isWhitespace(charAfter)) {
+            isValidOpening = true;
+          }
+        } else {
+          if (i < parent.children.length - 1) {
+            // TODO(drifkin): I'm not sure that this check is strict enough.
+            // We're trying to detect cases like `**[abc]()` where the char
+            // after the opening ** is indeed a non-whitespace character. We're
+            // using the heuristic that there's another item after the current
+            // one, but I'm not sure if that is good enough. In a well
+            // constructed tree, there aren't two text nodes in a row, so this
+            // _seems_ good, but I should think through it more
+            isValidOpening = true;
+          }
+        }
+
+        if (isValidOpening) {
+          // TODO(drifkin): close the bold
+          const strBefore = textNode.value.slice(0, index);
+          const strAfter = textNode.value.slice(index + sep.length);
+          (textNode.node as Text).value = strBefore;
+          // TODO(drifkin): the node above could be empty in which case we probably want to delete it
+          const children: PhrasingContent[] = [
+            ...(strAfter.length > 0 ? [u("text", { value: strAfter })] : []),
+          ];
+          const strongNode: Strong = u("strong", {
+            children,
+          });
+          const nodesAfter = (parent.node as Parent).children.splice(
+            i + 1,
+            parent.children.length - i - 1,
+            strongNode,
+          );
+          // TODO(drifkin): this cast seems iffy, should see if we can cast the
+          // parent instead, which would also help us check some of our
+          // assumptions
+          strongNode.children.push(...(nodesAfter as PhrasingContent[]));
+          return true;
+        }
+      }
+    }
+  }
+
+  return false;
+}
+
+function prevSibling(node: Node & Proxy): (Node & Proxy) | null {
+  const parent = node.parent;
+  if (parent) {
+    const index = parent.children.indexOf(node);
+    return parent.children[index - 1] as Node & Proxy;
+  }
+  return null;
+}
+
+function isWhitespace(str: string) {
+  return str.trim() === "";
+}
+
+// function debugPrintTreeNoPos(tree: Node) {
+//   console.log(
+//     JSON.stringify(
+//       tree,
+//       (key, value) => {
+//         if (key === "position") {
+//           return undefined;
+//         }
+//         return value;
+//       },
+//       2,
+//     ),
+//   );
+// }
+
+export default remarkStreamingMarkdown;
--- a/app/ui/responses/types.go
+++ b/app/ui/responses/types.go
@@ -101,14 +101,15 @@ type HealthResponse struct {
 }

 type User struct {
-	ID        string `json:"id"`
-	Email     string `json:"email"`
-	Name      string `json:"name"`
-	Bio       string `json:"bio,omitempty"`
-	AvatarURL string `json:"avatarurl,omitempty"`
-	FirstName string `json:"firstname,omitempty"`
-	LastName  string `json:"lastname,omitempty"`
-	Plan      string `json:"plan,omitempty"`
+	ID            string `json:"id"`
+	Name          string `json:"name"`
+	Email         string `json:"email"`
+	AvatarURL     string `json:"avatarURL"`
+	Plan          string `json:"plan"`
+	Bio           string `json:"bio"`
+	FirstName     string `json:"firstName"`
+	LastName      string `json:"lastName"`
+	OverThreshold bool   `json:"overThreshold"`
 }

 type Attachment struct {
--- a/app/ui/ui.go
+++ b/app/ui/ui.go
@@ -12,17 +12,18 @@ import (
 	"log/slog"
 	"net/http"
 	"net/http/httputil"
+	"net/url"
 	"os"
 	"runtime"
 	"runtime/debug"
 	"slices"
 	"strconv"
 	"strings"
-	"sync"
 	"time"

 	"github.com/google/uuid"
 	"github.com/ollama/ollama/api"
+	"github.com/ollama/ollama/app/auth"
 	"github.com/ollama/ollama/app/server"
 	"github.com/ollama/ollama/app/store"
 	"github.com/ollama/ollama/app/tools"
@@ -117,66 +118,40 @@ func (s *Server) log() *slog.Logger {

 // ollamaProxy creates a reverse proxy handler to the Ollama server
 func (s *Server) ollamaProxy() http.Handler {
-	var (
-		proxy   http.Handler
-		proxyMu sync.Mutex
-	)
+	ollamaHost := os.Getenv("OLLAMA_HOST")
+	if ollamaHost == "" {
+		ollamaHost = "http://127.0.0.1:11434"
+	}

-	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		proxyMu.Lock()
-		p := proxy
-		proxyMu.Unlock()
+	if !strings.HasPrefix(ollamaHost, "http://") && !strings.HasPrefix(ollamaHost, "https://") {
+		ollamaHost = "http://" + ollamaHost
+	}

-		if p == nil {
-			proxyMu.Lock()
-			if proxy == nil {
-				var err error
-				for i := range 2 {
-					if i > 0 {
-						s.log().Warn("ollama server not ready, retrying", "attempt", i+1)
-						time.Sleep(1 * time.Second)
-					}
+	target, err := url.Parse(ollamaHost)
+	if err != nil {
+		s.log().Error("failed to parse OLLAMA_HOST", "error", err, "host", ollamaHost)
+		return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+			http.Error(w, "failed to configure proxy", http.StatusInternalServerError)
+		})
+	}

-					err = WaitForServer(context.Background(), 10*time.Second)
-					if err == nil {
-						break
-					}
-				}
+	s.log().Info("configuring ollama proxy", "target", target.String())

-				if err != nil {
-					proxyMu.Unlock()
-					s.log().Error("ollama server not ready after retries", "error", err)
-					http.Error(w, "Ollama server is not ready", http.StatusServiceUnavailable)
-					return
-				}
+	proxy := httputil.NewSingleHostReverseProxy(target)

-				target := envconfig.Host()
-				s.log().Info("configuring ollama proxy", "target", target.String())
+	originalDirector := proxy.Director
+	proxy.Director = func(req *http.Request) {
+		originalDirector(req)
+		req.Host = target.Host
+		s.log().Debug("proxying request", "method", req.Method, "path", req.URL.Path, "target", target.Host)
+	}

-				newProxy := httputil.NewSingleHostReverseProxy(target)
+	proxy.ErrorHandler = func(w http.ResponseWriter, r *http.Request, err error) {
+		s.log().Error("proxy error", "error", err, "path", r.URL.Path, "target", target.String())
+		http.Error(w, "proxy error: "+err.Error(), http.StatusBadGateway)
+	}

-				originalDirector := newProxy.Director
-				newProxy.Director = func(req *http.Request) {
-					originalDirector(req)
-					req.Host = target.Host
-					s.log().Debug("proxying request", "method", req.Method, "path", req.URL.Path, "target", target.Host)
-				}
-
-				newProxy.ErrorHandler = func(w http.ResponseWriter, r *http.Request, err error) {
-					s.log().Error("proxy error", "error", err, "path", r.URL.Path, "target", target.String())
-					http.Error(w, "proxy error: "+err.Error(), http.StatusBadGateway)
-				}
-
-				proxy = newProxy
-				p = newProxy
-			} else {
-				p = proxy
-			}
-			proxyMu.Unlock()
-		}
-
-		p.ServeHTTP(w, r)
-	})
+	return proxy
 }

 type errHandlerFunc func(http.ResponseWriter, *http.Request) error
@@ -289,10 +264,11 @@ func (s *Server) Handler() http.Handler {
 	ollamaProxy := s.ollamaProxy()
 	mux.Handle("GET /api/tags", ollamaProxy)
 	mux.Handle("POST /api/show", ollamaProxy)
-	mux.Handle("GET /api/version", ollamaProxy)
-	mux.Handle("HEAD /api/version", ollamaProxy)
-	mux.Handle("POST /api/me", ollamaProxy)
-	mux.Handle("POST /api/signout", ollamaProxy)
+
+	mux.Handle("GET /api/v1/me", handle(s.me))
+	mux.Handle("POST /api/v1/disconnect", handle(s.disconnect))
+	mux.Handle("GET /api/v1/connect", handle(s.connectURL))
+	mux.Handle("GET /api/v1/health", handle(s.health))

 	// React app - catch all non-API routes and serve the React app
 	mux.Handle("GET /", s.appHandler())
@@ -362,7 +338,7 @@ func (s *Server) doSelfSigned(ctx context.Context, method, path string) (*http.R
 }

 // UserData fetches user data from ollama.com API for the current ollama key
-func (s *Server) UserData(ctx context.Context) (*api.UserResponse, error) {
+func (s *Server) UserData(ctx context.Context) (*responses.User, error) {
 	resp, err := s.doSelfSigned(ctx, http.MethodPost, "/api/me")
 	if err != nil {
 		return nil, fmt.Errorf("failed to call ollama.com/api/me: %w", err)
@@ -373,7 +349,7 @@ func (s *Server) UserData(ctx context.Context) (*api.UserResponse, error) {
 		return nil, fmt.Errorf("unexpected status code: %d", resp.StatusCode)
 	}

-	var user api.UserResponse
+	var user responses.User
 	if err := json.NewDecoder(resp.Body).Decode(&user); err != nil {
 		return nil, fmt.Errorf("failed to parse user response: %w", err)
 	}
@@ -392,27 +368,29 @@ func (s *Server) UserData(ctx context.Context) (*api.UserResponse, error) {
 	return &user, nil
 }

-// WaitForServer waits for the Ollama server to be ready
-func WaitForServer(ctx context.Context, timeout time.Duration) error {
-	deadline := time.Now().Add(timeout)
-	for time.Now().Before(deadline) {
+func waitForServer(ctx context.Context) error {
+	timeout := time.Now().Add(10 * time.Second)
+	// TODO: this avoids an error on first load of the app
+	// however we should either show a loading state or
+	// wait for the Ollama server to be ready before redirecting
+	for {
 		c, err := api.ClientFromEnvironment()
 		if err != nil {
 			return err
 		}
 		if _, err := c.Version(ctx); err == nil {
-			slog.Debug("ollama server is ready")
-			return nil
+			break
+		}
+		if time.Now().After(timeout) {
+			return fmt.Errorf("timeout waiting for Ollama server to be ready")
 		}
 		time.Sleep(10 * time.Millisecond)
 	}
-	return errors.New("timeout waiting for Ollama server to be ready")
+	return nil
 }

 func (s *Server) createChat(w http.ResponseWriter, r *http.Request) error {
-	if err := WaitForServer(r.Context(), 10*time.Second); err != nil {
-		return err
-	}
+	waitForServer(r.Context())

 	id, err := uuid.NewV7()
 	if err != nil {
@@ -997,7 +975,7 @@ func (s *Server) chat(w http.ResponseWriter, r *http.Request) error {
 				for _, toolCall := range res.Message.ToolCalls {
 					// continues loop as tools were executed
 					toolsExecuted = true
-					result, content, err := registry.Execute(ctx, toolCall.Function.Name, toolCall.Function.Arguments.ToMap())
+					result, content, err := registry.Execute(ctx, toolCall.Function.Name, toolCall.Function.Arguments)
 					if err != nil {
 						errContent := fmt.Sprintf("Error: %v", err)
 						toolErrMsg := store.NewMessage("tool", errContent, nil)
@@ -1460,6 +1438,129 @@ func (s *Server) settings(w http.ResponseWriter, r *http.Request) error {
 	})
 }

+func (s *Server) me(w http.ResponseWriter, r *http.Request) error {
+	if r.Method != http.MethodGet {
+		http.Error(w, "Method Not Allowed", http.StatusMethodNotAllowed)
+		return nil
+	}
+
+	user, err := s.UserData(r.Context())
+	if err != nil {
+		// If fetching from API fails, try to return cached user data if available
+		if cachedUser, cacheErr := s.Store.User(); cacheErr == nil && cachedUser != nil {
+			s.log().Info("API request failed, returning cached user data", "error", err)
+			responseUser := &responses.User{
+				Name:  cachedUser.Name,
+				Email: cachedUser.Email,
+				Plan:  cachedUser.Plan,
+			}
+			w.Header().Set("Content-Type", "application/json")
+			w.WriteHeader(http.StatusOK)
+			return json.NewEncoder(w).Encode(responseUser)
+		}
+
+		s.log().Error("failed to get user data", "error", err)
+		w.WriteHeader(http.StatusInternalServerError)
+		return json.NewEncoder(w).Encode(responses.Error{
+			Error: "failed to get user data",
+		})
+	}
+
+	w.Header().Set("Content-Type", "application/json")
+	w.WriteHeader(http.StatusOK)
+	return json.NewEncoder(w).Encode(user)
+}
+
+func (s *Server) disconnect(w http.ResponseWriter, r *http.Request) error {
+	if r.Method != http.MethodPost {
+		http.Error(w, "Method Not Allowed", http.StatusMethodNotAllowed)
+		return nil
+	}
+
+	if err := s.Store.ClearUser(); err != nil {
+		s.log().Warn("failed to clear cached user data", "error", err)
+	}
+
+	// Get the SSH public key to encode for the delete request
+	pubKey, err := ollamaAuth.GetPublicKey()
+	if err != nil {
+		s.log().Error("failed to get public key", "error", err)
+		w.WriteHeader(http.StatusInternalServerError)
+		return json.NewEncoder(w).Encode(responses.Error{
+			Error: "failed to get public key",
+		})
+	}
+
+	// Encode the key using base64 URL encoding
+	encodedKey := base64.RawURLEncoding.EncodeToString([]byte(pubKey))
+
+	// Call the /api/user/keys/{encodedKey} endpoint with DELETE
+	resp, err := s.doSelfSigned(r.Context(), http.MethodDelete, fmt.Sprintf("/api/user/keys/%s", encodedKey))
+	if err != nil {
+		s.log().Error("failed to call ollama.com/api/user/keys", "error", err)
+		w.WriteHeader(http.StatusInternalServerError)
+		return json.NewEncoder(w).Encode(responses.Error{
+			Error: "failed to disconnect from ollama.com",
+		})
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK {
+		s.log().Error("disconnect request failed", "status", resp.StatusCode)
+		w.WriteHeader(http.StatusInternalServerError)
+		return json.NewEncoder(w).Encode(responses.Error{
+			Error: "failed to disconnect from ollama.com",
+		})
+	}
+
+	w.Header().Set("Content-Type", "application/json")
+	w.WriteHeader(http.StatusOK)
+	return json.NewEncoder(w).Encode(map[string]string{"status": "disconnected"})
+}
+
+func (s *Server) connectURL(w http.ResponseWriter, r *http.Request) error {
+	if r.Method != http.MethodGet {
+		http.Error(w, "Method Not Allowed", http.StatusMethodNotAllowed)
+		return nil
+	}
+
+	connectURL, err := auth.BuildConnectURL(OllamaDotCom)
+	if err != nil {
+		s.log().Error("failed to build connect URL", "error", err)
+		w.WriteHeader(http.StatusInternalServerError)
+		return json.NewEncoder(w).Encode(responses.Error{
+			Error: "failed to build connect URL",
+		})
+	}
+
+	w.Header().Set("Content-Type", "application/json")
+	w.WriteHeader(http.StatusOK)
+	return json.NewEncoder(w).Encode(map[string]string{
+		"connect_url": connectURL,
+	})
+}
+
+func (s *Server) health(w http.ResponseWriter, r *http.Request) error {
+	if r.Method != http.MethodGet {
+		http.Error(w, "Method Not Allowed", http.StatusMethodNotAllowed)
+		return nil
+	}
+
+	healthy := false
+	c, err := api.ClientFromEnvironment()
+	if err == nil {
+		if _, err := c.Version(r.Context()); err == nil {
+			healthy = true
+		}
+	}
+
+	w.Header().Set("Content-Type", "application/json")
+	w.WriteHeader(http.StatusOK)
+	return json.NewEncoder(w).Encode(responses.HealthResponse{
+		Healthy: healthy,
+	})
+}
+
 func (s *Server) getInferenceCompute(w http.ResponseWriter, r *http.Request) error {
 	ctx, cancel := context.WithTimeout(r.Context(), 500*time.Millisecond)
 	defer cancel()
@@ -1558,13 +1659,13 @@ func convertToOllamaTool(toolSchema map[string]any) api.Tool {

 	tool.Function.Parameters.Type = "object"
 	tool.Function.Parameters.Required = []string{}
-	tool.Function.Parameters.Properties = api.NewToolPropertiesMap()
+	tool.Function.Parameters.Properties = make(map[string]api.ToolProperty)

 	if schemaProps, ok := toolSchema["schema"].(map[string]any); ok {
 		tool.Function.Parameters.Type = getStringFromMap(schemaProps, "type", "object")

 		if props, ok := schemaProps["properties"].(map[string]any); ok {
-			tool.Function.Parameters.Properties = api.NewToolPropertiesMap()
+			tool.Function.Parameters.Properties = make(map[string]api.ToolProperty)

 			for propName, propDef := range props {
 				if propMap, ok := propDef.(map[string]any); ok {
@@ -1572,7 +1673,7 @@ func convertToOllamaTool(toolSchema map[string]any) api.Tool {
 						Type:        api.PropertyType{getStringFromMap(propMap, "type", "string")},
 						Description: getStringFromMap(propMap, "description", ""),
 					}
-					tool.Function.Parameters.Properties.Set(propName, prop)
+					tool.Function.Parameters.Properties[propName] = prop
 				}
 			}
 		}
@@ -1604,7 +1705,7 @@ func getStringFromMap(m map[string]any, key, defaultValue string) string {
 // isImageAttachment checks if a filename is an image file
 func isImageAttachment(filename string) bool {
 	ext := strings.ToLower(filename)
-	return strings.HasSuffix(ext, ".png") || strings.HasSuffix(ext, ".jpg") || strings.HasSuffix(ext, ".jpeg") || strings.HasSuffix(ext, ".webp")
+	return strings.HasSuffix(ext, ".png") || strings.HasSuffix(ext, ".jpg") || strings.HasSuffix(ext, ".jpeg")
 }

 // ptr is a convenience function for &literal
@@ -1693,14 +1794,13 @@ func (s *Server) buildChatRequest(chat *store.Chat, model string, think any, ava

 	var thinkValue *api.ThinkValue
 	if think != nil {
-		// Only set Think if it's actually requesting thinking
 		if boolValue, ok := think.(bool); ok {
-			if boolValue {
-				thinkValue = &api.ThinkValue{Value: boolValue}
+			thinkValue = &api.ThinkValue{
+				Value: boolValue,
 			}
 		} else if stringValue, ok := think.(string); ok {
-			if stringValue != "" && stringValue != "none" {
-				thinkValue = &api.ThinkValue{Value: stringValue}
+			thinkValue = &api.ThinkValue{
+				Value: stringValue,
 			}
 		}
 	}
--- a/app/wintray/eventloop.go
+++ b/app/wintray/eventloop.go
@@ -158,16 +158,16 @@ func (t *winTray) wndProc(hWnd windows.Handle, message uint32, wParam, lParam ui
 	case uint32(UI_REQUEST_MSG_ID):
 		// Requests for the UI must always come from the main event thread
 		l := int(wParam)
-		path := unsafe.String((*byte)(unsafe.Pointer(lParam)), l) //nolint:govet,gosec
+		path := unsafe.String((*byte)(unsafe.Pointer(lParam)), l)
 		t.app.UIRun(path)
 	case WM_COPYDATA:
 		// Handle URL scheme requests from other instances
 		if lParam != 0 {
-			cds := (*COPYDATASTRUCT)(unsafe.Pointer(lParam)) //nolint:govet,gosec
-			if cds.DwData == 1 {                             // Our identifier for URL scheme messages
+			cds := (*COPYDATASTRUCT)(unsafe.Pointer(lParam))
+			if cds.DwData == 1 { // Our identifier for URL scheme messages
 				// Convert the data back to string
 				data := make([]byte, cds.CbData)
-				copy(data, (*[1 << 30]byte)(unsafe.Pointer(cds.LpData))[:cds.CbData:cds.CbData]) //nolint:govet,gosec
+				copy(data, (*[1 << 30]byte)(unsafe.Pointer(cds.LpData))[:cds.CbData:cds.CbData])
 				urlScheme := string(data)
 				handleURLSchemeRequest(urlScheme)
 				lResult = 1 // Return non-zero to indicate success
--- a/cmd/background_unix.go
+++ b/cmd/background_unix.go
@@ -1,13 +0,0 @@
-//go:build !windows
-
-package cmd
-
-import "syscall"
-
-// backgroundServerSysProcAttr returns SysProcAttr for running the server in the background on Unix.
-// Setpgid prevents the server from being killed when the parent process exits.
-func backgroundServerSysProcAttr() *syscall.SysProcAttr {
-	return &syscall.SysProcAttr{
-		Setpgid: true,
-	}
-}
--- a/cmd/background_windows.go
+++ b/cmd/background_windows.go
@@ -1,12 +0,0 @@
-package cmd
-
-import "syscall"
-
-// backgroundServerSysProcAttr returns SysProcAttr for running the server in the background on Windows.
-// CREATE_NO_WINDOW (0x08000000) prevents a console window from appearing.
-func backgroundServerSysProcAttr() *syscall.SysProcAttr {
-	return &syscall.SysProcAttr{
-		CreationFlags: 0x08000000,
-		HideWindow:    true,
-	}
-}
--- a/cmd/bench/README.md
+++ b/cmd/bench/README.md
@@ -1,115 +0,0 @@
-Ollama Benchmark Tool
---------------------
-
-A Go-based command-line tool for benchmarking Ollama models with configurable parameters and multiple output formats.
-
-## Features
-
- * Benchmark multiple models in a single run
- * Support for both text and image prompts
- * Configurable generation parameters (temperature, max tokens, seed, etc.)
- * Supports benchstat and CSV output formats
- * Detailed performance metrics (prefill, generate, load, total durations)
-
-## Building from Source
-
-```
-go build -o ollama-bench bench.go
-./ollama-bench -model gpt-oss:20b -epochs 6 -format csv
-```
-
-Using Go Run (without building)
-
-```
-go run bench.go -model gpt-oss:20b -epochs 3
-```
-
-## Usage
-
-### Basic Example
-
-```
-./ollama-bench -model gemma3 -epochs 6
-```
-
-### Benchmark Multiple Models
-
-```
-./ollama-bench -model gemma3,gemma3n -epochs 6 -max-tokens 100 -p "Write me a short story" | tee gemma.bench
-benchstat -col /name gemma.bench
-```
-
-### With Image Prompt
-
-```
-./ollama-bench -model qwen3-vl -image photo.jpg -epochs 6 -max-tokens 100 -p "Describe this image"
-```
-
-### Advanced Example
-
-```
-./ollama-bench -model llama3 -epochs 10 -temperature 0.7 -max-tokens 500 -seed 42 -format csv -output results.csv
-```
-
-## Command Line Options
-
-| Option  	| Description | Default |
-|----------|-------------|---------|
-| -model	| Comma-separated list of models to benchmark	| (required)		|
-| -epochs	| Number of iterations per model		| 1			|
-| -max-tokens	| Maximum tokens for model response		| 0 (unlimited)		|
-| -temperature	| Temperature parameter				| 0.0			|
-| -seed		| Random seed					| 0 (random)		|
-| -timeout	| Timeout in seconds				| 300			|
-| -p		| Prompt text					| "Write a long story."	|
-| -image	| Image file to include in prompt		| 			|
-| -k		| Keep-alive duration in seconds		| 0			|
-| -format	| Output format (benchstat, csv)		| benchstat		|
-| -output	| Output file for results			| "" (stdout)		|
-| -v		| Verbose mode					| false			|
-| -debug	| Show debug information			| false			|
-
-## Output Formats
-
-### Markdown Format
-
-The default markdown format is suitable for copying and pasting into a GitHub issue and will look like:
-```
- Model | Step | Count | Duration | nsPerToken | tokensPerSec |
-|-------|------|-------|----------|------------|--------------|
-| gpt-oss:20b | prefill | 124 | 30.006458ms | 241987.56 | 4132.44 |
-| gpt-oss:20b | generate | 200 | 2.646843954s | 13234219.77 | 75.56 |
-| gpt-oss:20b | load | 1 | 121.674208ms | - | - |
-| gpt-oss:20b | total | 1 | 2.861047625s | - | - |
-```
-
-### Benchstat Format
-
-Compatible with Go's benchstat tool for statistical analysis:
-
-```
-BenchmarkModel/name=gpt-oss:20b/step=prefill 128 78125.00 ns/token 12800.00 token/sec
-BenchmarkModel/name=gpt-oss:20b/step=generate 512 19531.25 ns/token 51200.00 token/sec
-BenchmarkModel/name=gpt-oss:20b/step=load 1 1500000000 ns/request
-```
-
-### CSV Format
-
-Machine-readable comma-separated values:
-
-```
-NAME,STEP,COUNT,NS_PER_COUNT,TOKEN_PER_SEC
-gpt-oss:20b,prefill,128,78125.00,12800.00
-gpt-oss:20b,generate,512,19531.25,51200.00
-gpt-oss:20b,load,1,1500000000,0
-```
-
-## Metrics Explained
-
-The tool reports four types of metrics for each model:
-
- * prefill: Time spent processing the prompt
- * generate: Time spent generating the response
- * load: Model loading time (one-time cost)
- * total: Total request duration
-
--- a/cmd/bench/bench.go
+++ b/cmd/bench/bench.go
@@ -1,321 +0,0 @@
-package main
-
-import (
-	"cmp"
-	"context"
-	"flag"
-	"fmt"
-	"io"
-	"os"
-	"runtime"
-	"slices"
-	"strings"
-	"sync"
-	"time"
-
-	"github.com/ollama/ollama/api"
-)
-
-type flagOptions struct {
-	models      *string
-	epochs      *int
-	maxTokens   *int
-	temperature *float64
-	seed        *int
-	timeout     *int
-	prompt      *string
-	imageFile   *string
-	keepAlive   *float64
-	format      *string
-	outputFile  *string
-	debug       *bool
-	verbose     *bool
-}
-
-type Metrics struct {
-	Model    string
-	Step     string
-	Count    int
-	Duration time.Duration
-}
-
-var once sync.Once
-
-const DefaultPrompt = `Please write a descriptive story about a llama named Alonso who grows up to be President of the Land of Llamas. Include details about Alonso's childhood, adolescent years, and how he grew up to be a political mover and shaker. Write the story with a sense of whimsy.`
-
-func OutputMetrics(w io.Writer, format string, metrics []Metrics, verbose bool) {
-	switch format {
-	case "benchstat":
-		if verbose {
-			printHeader := func() {
-				fmt.Fprintf(w, "sysname: %s\n", runtime.GOOS)
-				fmt.Fprintf(w, "machine: %s\n", runtime.GOARCH)
-			}
-			once.Do(printHeader)
-		}
-		for _, m := range metrics {
-			if m.Step == "generate" || m.Step == "prefill" {
-				if m.Count > 0 {
-					nsPerToken := float64(m.Duration.Nanoseconds()) / float64(m.Count)
-					tokensPerSec := float64(m.Count) / (float64(m.Duration.Nanoseconds()) + 1e-12) * 1e9
-
-					fmt.Fprintf(w, "BenchmarkModel/name=%s/step=%s %d %.2f ns/token %.2f token/sec\n",
-						m.Model, m.Step, m.Count, nsPerToken, tokensPerSec)
-				} else {
-					fmt.Fprintf(w, "BenchmarkModel/name=%s/step=%s %d 0 ns/token 0 token/sec\n",
-						m.Model, m.Step, m.Count)
-				}
-			} else {
-				var suffix string
-				if m.Step == "load" {
-					suffix = "/step=load"
-				}
-				fmt.Fprintf(w, "BenchmarkModel/name=%s%s 1 %d ns/request\n",
-					m.Model, suffix, m.Duration.Nanoseconds())
-			}
-		}
-	case "csv":
-		printHeader := func() {
-			headings := []string{"NAME", "STEP", "COUNT", "NS_PER_COUNT", "TOKEN_PER_SEC"}
-			fmt.Fprintln(w, strings.Join(headings, ","))
-		}
-		once.Do(printHeader)
-
-		for _, m := range metrics {
-			if m.Step == "generate" || m.Step == "prefill" {
-				var nsPerToken float64
-				var tokensPerSec float64
-				if m.Count > 0 {
-					nsPerToken = float64(m.Duration.Nanoseconds()) / float64(m.Count)
-					tokensPerSec = float64(m.Count) / (float64(m.Duration.Nanoseconds()) + 1e-12) * 1e9
-				}
-				fmt.Fprintf(w, "%s,%s,%d,%.2f,%.2f\n", m.Model, m.Step, m.Count, nsPerToken, tokensPerSec)
-			} else {
-				fmt.Fprintf(w, "%s,%s,1,%d,0\n", m.Model, m.Step, m.Duration.Nanoseconds())
-			}
-		}
-	case "markdown":
-		printHeader := func() {
-			fmt.Fprintln(w, "| Model | Step | Count | Duration | nsPerToken | tokensPerSec |")
-			fmt.Fprintln(w, "|-------|------|-------|----------|------------|--------------|")
-		}
-		once.Do(printHeader)
-
-		for _, m := range metrics {
-			var nsPerToken, tokensPerSec float64
-			var nsPerTokenStr, tokensPerSecStr string
-
-			if m.Step == "generate" || m.Step == "prefill" {
-				nsPerToken = float64(m.Duration.Nanoseconds()) / float64(m.Count)
-				tokensPerSec = float64(m.Count) / (float64(m.Duration.Nanoseconds()) + 1e-12) * 1e9
-				nsPerTokenStr = fmt.Sprintf("%.2f", nsPerToken)
-				tokensPerSecStr = fmt.Sprintf("%.2f", tokensPerSec)
-			} else {
-				nsPerTokenStr = "-"
-				tokensPerSecStr = "-"
-			}
-
-			fmt.Fprintf(w, "| %s | %s | %d | %v | %s | %s |\n",
-				m.Model, m.Step, m.Count, m.Duration, nsPerTokenStr, tokensPerSecStr)
-		}
-	default:
-		fmt.Fprintf(os.Stderr, "Unknown output format '%s'\n", format)
-	}
-}
-
-func BenchmarkChat(fOpt flagOptions) error {
-	models := strings.Split(*fOpt.models, ",")
-
-	// todo - add multi-image support
-	var imgData api.ImageData
-	var err error
-	if *fOpt.imageFile != "" {
-		imgData, err = readImage(*fOpt.imageFile)
-		if err != nil {
-			fmt.Fprintf(os.Stderr, "ERROR: Couldn't read image '%s': %v\n", *fOpt.imageFile, err)
-			return err
-		}
-	}
-
-	if *fOpt.debug && imgData != nil {
-		fmt.Fprintf(os.Stderr, "Read file '%s'\n", *fOpt.imageFile)
-	}
-
-	client, err := api.ClientFromEnvironment()
-	if err != nil {
-		fmt.Fprintf(os.Stderr, "ERROR: Couldn't create ollama client: %v\n", err)
-		return err
-	}
-
-	var out io.Writer = os.Stdout
-	if fOpt.outputFile != nil && *fOpt.outputFile != "" {
-		f, err := os.OpenFile(*fOpt.outputFile, os.O_CREATE|os.O_WRONLY, 0o644)
-		if err != nil {
-			fmt.Fprintf(os.Stderr, "ERROR: cannot open output file %s: %v\n", *fOpt.outputFile, err)
-			return err
-		}
-		defer f.Close()
-		out = f
-	}
-
-	for _, model := range models {
-		for range *fOpt.epochs {
-			options := make(map[string]interface{})
-			if *fOpt.maxTokens > 0 {
-				options["num_predict"] = *fOpt.maxTokens
-			}
-			options["temperature"] = *fOpt.temperature
-			if fOpt.seed != nil && *fOpt.seed > 0 {
-				options["seed"] = *fOpt.seed
-			}
-
-			var keepAliveDuration *api.Duration
-			if *fOpt.keepAlive > 0 {
-				duration := api.Duration{Duration: time.Duration(*fOpt.keepAlive * float64(time.Second))}
-				keepAliveDuration = &duration
-			}
-
-			req := &api.ChatRequest{
-				Model: model,
-				Messages: []api.Message{
-					{
-						Role:    "user",
-						Content: *fOpt.prompt,
-					},
-				},
-				Options:   options,
-				KeepAlive: keepAliveDuration,
-			}
-
-			if imgData != nil {
-				req.Messages[0].Images = []api.ImageData{imgData}
-			}
-
-			var responseMetrics *api.Metrics
-
-			ctx, cancel := context.WithTimeout(context.Background(), time.Duration(*fOpt.timeout)*time.Second)
-			defer cancel()
-
-			err = client.Chat(ctx, req, func(resp api.ChatResponse) error {
-				if *fOpt.debug {
-					fmt.Fprintf(os.Stderr, "%s", cmp.Or(resp.Message.Thinking, resp.Message.Content))
-				}
-
-				if resp.Done {
-					responseMetrics = &resp.Metrics
-				}
-				return nil
-			})
-
-			if *fOpt.debug {
-				fmt.Fprintln(os.Stderr)
-			}
-
-			if err != nil {
-				if ctx.Err() == context.DeadlineExceeded {
-					fmt.Fprintf(os.Stderr, "ERROR: Chat request timed out with model '%s' after %vs\n", model, 1)
-					continue
-				}
-				fmt.Fprintf(os.Stderr, "ERROR: Couldn't chat with model '%s': %v\n", model, err)
-				continue
-			}
-
-			if responseMetrics == nil {
-				fmt.Fprintf(os.Stderr, "ERROR: No metrics received for model '%s'\n", model)
-				continue
-			}
-
-			metrics := []Metrics{
-				{
-					Model:    model,
-					Step:     "prefill",
-					Count:    responseMetrics.PromptEvalCount,
-					Duration: responseMetrics.PromptEvalDuration,
-				},
-				{
-					Model:    model,
-					Step:     "generate",
-					Count:    responseMetrics.EvalCount,
-					Duration: responseMetrics.EvalDuration,
-				},
-				{
-					Model:    model,
-					Step:     "load",
-					Count:    1,
-					Duration: responseMetrics.LoadDuration,
-				},
-				{
-					Model:    model,
-					Step:     "total",
-					Count:    1,
-					Duration: responseMetrics.TotalDuration,
-				},
-			}
-
-			OutputMetrics(out, *fOpt.format, metrics, *fOpt.verbose)
-
-			if *fOpt.keepAlive > 0 {
-				time.Sleep(time.Duration(*fOpt.keepAlive*float64(time.Second)) + 200*time.Millisecond)
-			}
-		}
-	}
-
-	return nil
-}
-
-func readImage(filePath string) (api.ImageData, error) {
-	file, err := os.Open(filePath)
-	if err != nil {
-		return nil, err
-	}
-	defer file.Close()
-
-	data, err := io.ReadAll(file)
-	if err != nil {
-		return nil, err
-	}
-
-	return api.ImageData(data), nil
-}
-
-func main() {
-	fOpt := flagOptions{
-		models:      flag.String("model", "", "Model to benchmark"),
-		epochs:      flag.Int("epochs", 6, "Number of epochs (iterations) per model"),
-		maxTokens:   flag.Int("max-tokens", 200, "Maximum tokens for model response"),
-		temperature: flag.Float64("temperature", 0, "Temperature parameter"),
-		seed:        flag.Int("seed", 0, "Random seed"),
-		timeout:     flag.Int("timeout", 60*5, "Timeout in seconds (default 300s)"),
-		prompt:      flag.String("p", DefaultPrompt, "Prompt to use"),
-		imageFile:   flag.String("image", "", "Filename for an image to include"),
-		keepAlive:   flag.Float64("k", 0, "Keep alive duration in seconds"),
-		format:      flag.String("format", "markdown", "Output format [benchstat|csv] (default benchstat)"),
-		outputFile:  flag.String("output", "", "Output file for results (stdout if empty)"),
-		verbose:     flag.Bool("v", false, "Show system information"),
-		debug:       flag.Bool("debug", false, "Show debug information"),
-	}
-
-	flag.Usage = func() {
-		fmt.Fprintf(os.Stderr, "Usage: %s [OPTIONS]\n\n", os.Args[0])
-		fmt.Fprintf(os.Stderr, "Description:\n")
-		fmt.Fprintf(os.Stderr, "  Model benchmarking tool with configurable parameters\n\n")
-		fmt.Fprintf(os.Stderr, "Options:\n")
-		flag.PrintDefaults()
-		fmt.Fprintf(os.Stderr, "\nExamples:\n")
-		fmt.Fprintf(os.Stderr, "  bench -model gpt-oss:20b -epochs 3 -temperature 0.7\n")
-	}
-	flag.Parse()
-
-	if !slices.Contains([]string{"markdown", "benchstat", "csv"}, *fOpt.format) {
-		fmt.Fprintf(os.Stderr, "ERROR: Unknown format '%s'\n", *fOpt.format)
-		os.Exit(1)
-	}
-
-	if len(*fOpt.models) == 0 {
-		fmt.Fprintf(os.Stderr, "ERROR: No model(s) specified to benchmark.\n")
-		flag.Usage()
-		return
-	}
-
-	BenchmarkChat(fOpt)
-}
--- a/cmd/bench/bench_test.go
+++ b/cmd/bench/bench_test.go
@@ -1,463 +0,0 @@
-package main
-
-import (
-	"bytes"
-	"crypto/rand"
-	"encoding/json"
-	"io"
-	"net/http"
-	"net/http/httptest"
-	"os"
-	"strings"
-	"testing"
-	"time"
-
-	"github.com/ollama/ollama/api"
-)
-
-func createTestFlagOptions() flagOptions {
-	models := "test-model"
-	format := "benchstat"
-	epochs := 1
-	maxTokens := 100
-	temperature := 0.7
-	seed := 42
-	timeout := 30
-	prompt := "test prompt"
-	imageFile := ""
-	keepAlive := 5.0
-	verbose := false
-	debug := false
-
-	return flagOptions{
-		models:      &models,
-		format:      &format,
-		epochs:      &epochs,
-		maxTokens:   &maxTokens,
-		temperature: &temperature,
-		seed:        &seed,
-		timeout:     &timeout,
-		prompt:      &prompt,
-		imageFile:   &imageFile,
-		keepAlive:   &keepAlive,
-		verbose:     &verbose,
-		debug:       &debug,
-	}
-}
-
-func captureOutput(f func()) string {
-	oldStdout := os.Stdout
-	oldStderr := os.Stderr
-	defer func() {
-		os.Stdout = oldStdout
-		os.Stderr = oldStderr
-	}()
-
-	r, w, _ := os.Pipe()
-	os.Stdout = w
-	os.Stderr = w
-
-	f()
-
-	w.Close()
-	var buf bytes.Buffer
-	io.Copy(&buf, r)
-	return buf.String()
-}
-
-func createMockOllamaServer(t *testing.T, responses []api.ChatResponse) *httptest.Server {
-	return httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		if r.URL.Path != "/api/chat" {
-			t.Errorf("Expected path /api/chat, got %s", r.URL.Path)
-			http.Error(w, "Not found", http.StatusNotFound)
-			return
-		}
-
-		if r.Method != "POST" {
-			t.Errorf("Expected POST method, got %s", r.Method)
-			http.Error(w, "Method not allowed", http.StatusMethodNotAllowed)
-			return
-		}
-
-		w.Header().Set("Content-Type", "application/json")
-		w.WriteHeader(http.StatusOK)
-
-		for _, resp := range responses {
-			jsonData, err := json.Marshal(resp)
-			if err != nil {
-				t.Errorf("Failed to marshal response: %v", err)
-				return
-			}
-			w.Write(jsonData)
-			w.Write([]byte("\n"))
-			if f, ok := w.(http.Flusher); ok {
-				f.Flush()
-			}
-			time.Sleep(10 * time.Millisecond) // Simulate some delay
-		}
-	}))
-}
-
-func TestBenchmarkChat_Success(t *testing.T) {
-	fOpt := createTestFlagOptions()
-
-	mockResponses := []api.ChatResponse{
-		{
-			Model: "test-model",
-			Message: api.Message{
-				Role:    "assistant",
-				Content: "test response part 1",
-			},
-			Done: false,
-		},
-		{
-			Model: "test-model",
-			Message: api.Message{
-				Role:    "assistant",
-				Content: "test response part 2",
-			},
-			Done: true,
-			Metrics: api.Metrics{
-				PromptEvalCount:    10,
-				PromptEvalDuration: 100 * time.Millisecond,
-				EvalCount:          50,
-				EvalDuration:       500 * time.Millisecond,
-				TotalDuration:      600 * time.Millisecond,
-				LoadDuration:       50 * time.Millisecond,
-			},
-		},
-	}
-
-	server := createMockOllamaServer(t, mockResponses)
-	defer server.Close()
-
-	t.Setenv("OLLAMA_HOST", server.URL)
-
-	output := captureOutput(func() {
-		err := BenchmarkChat(fOpt)
-		if err != nil {
-			t.Errorf("Expected no error, got %v", err)
-		}
-	})
-
-	if !strings.Contains(output, "BenchmarkModel/name=test-model/step=prefill") {
-		t.Errorf("Expected output to contain prefill metrics, got: %s", output)
-	}
-	if !strings.Contains(output, "BenchmarkModel/name=test-model/step=generate") {
-		t.Errorf("Expected output to contain generate metrics, got: %s", output)
-	}
-	if !strings.Contains(output, "ns/token") {
-		t.Errorf("Expected output to contain ns/token metric, got: %s", output)
-	}
-}
-
-func TestBenchmarkChat_ServerError(t *testing.T) {
-	fOpt := createTestFlagOptions()
-
-	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		http.Error(w, "Internal server error", http.StatusInternalServerError)
-	}))
-	defer server.Close()
-
-	t.Setenv("OLLAMA_HOST", server.URL)
-
-	output := captureOutput(func() {
-		err := BenchmarkChat(fOpt)
-		if err != nil {
-			t.Errorf("Expected error to be handled internally, got returned error: %v", err)
-		}
-	})
-
-	if !strings.Contains(output, "ERROR: Couldn't chat with model") {
-		t.Errorf("Expected error message about chat failure, got: %s", output)
-	}
-}
-
-func TestBenchmarkChat_Timeout(t *testing.T) {
-	fOpt := createTestFlagOptions()
-	shortTimeout := 1 // Very short timeout
-	fOpt.timeout = &shortTimeout
-
-	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		// Simulate a long delay that will cause timeout
-		time.Sleep(2 * time.Second)
-
-		w.Header().Set("Content-Type", "application/json")
-		response := api.ChatResponse{
-			Model: "test-model",
-			Message: api.Message{
-				Role:    "assistant",
-				Content: "test response",
-			},
-			Done: true,
-			Metrics: api.Metrics{
-				PromptEvalCount:    10,
-				PromptEvalDuration: 100 * time.Millisecond,
-				EvalCount:          50,
-				EvalDuration:       500 * time.Millisecond,
-				TotalDuration:      600 * time.Millisecond,
-				LoadDuration:       50 * time.Millisecond,
-			},
-		}
-		jsonData, _ := json.Marshal(response)
-		w.Write(jsonData)
-	}))
-	defer server.Close()
-
-	t.Setenv("OLLAMA_HOST", server.URL)
-
-	output := captureOutput(func() {
-		err := BenchmarkChat(fOpt)
-		if err != nil {
-			t.Errorf("Expected timeout to be handled internally, got returned error: %v", err)
-		}
-	})
-
-	if !strings.Contains(output, "ERROR: Chat request timed out") {
-		t.Errorf("Expected timeout error message, got: %s", output)
-	}
-}
-
-func TestBenchmarkChat_NoMetrics(t *testing.T) {
-	fOpt := createTestFlagOptions()
-
-	mockResponses := []api.ChatResponse{
-		{
-			Model: "test-model",
-			Message: api.Message{
-				Role:    "assistant",
-				Content: "test response",
-			},
-			Done: false, // Never sends Done=true
-		},
-	}
-
-	server := createMockOllamaServer(t, mockResponses)
-	defer server.Close()
-
-	t.Setenv("OLLAMA_HOST", server.URL)
-
-	output := captureOutput(func() {
-		err := BenchmarkChat(fOpt)
-		if err != nil {
-			t.Errorf("Expected no error, got %v", err)
-		}
-	})
-
-	if !strings.Contains(output, "ERROR: No metrics received") {
-		t.Errorf("Expected no metrics error message, got: %s", output)
-	}
-}
-
-func TestBenchmarkChat_MultipleModels(t *testing.T) {
-	fOpt := createTestFlagOptions()
-	models := "model1,model2"
-	epochs := 2
-	fOpt.models = &models
-	fOpt.epochs = &epochs
-
-	callCount := 0
-	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		callCount++
-
-		w.Header().Set("Content-Type", "application/json")
-
-		var req api.ChatRequest
-		body, _ := io.ReadAll(r.Body)
-		json.Unmarshal(body, &req)
-
-		response := api.ChatResponse{
-			Model: req.Model,
-			Message: api.Message{
-				Role:    "assistant",
-				Content: "test response for " + req.Model,
-			},
-			Done: true,
-			Metrics: api.Metrics{
-				PromptEvalCount:    10,
-				PromptEvalDuration: 100 * time.Millisecond,
-				EvalCount:          50,
-				EvalDuration:       500 * time.Millisecond,
-				TotalDuration:      600 * time.Millisecond,
-				LoadDuration:       50 * time.Millisecond,
-			},
-		}
-		jsonData, _ := json.Marshal(response)
-		w.Write(jsonData)
-	}))
-	defer server.Close()
-
-	t.Setenv("OLLAMA_HOST", server.URL)
-
-	output := captureOutput(func() {
-		err := BenchmarkChat(fOpt)
-		if err != nil {
-			t.Errorf("Expected no error, got %v", err)
-		}
-	})
-
-	// Should be called 4 times (2 models × 2 epochs)
-	if callCount != 4 {
-		t.Errorf("Expected 4 API calls, got %d", callCount)
-	}
-
-	if !strings.Contains(output, "BenchmarkModel/name=model1") || !strings.Contains(output, "BenchmarkModel/name=model2") {
-		t.Errorf("Expected output for both models, got: %s", output)
-	}
-}
-
-func TestBenchmarkChat_WithImage(t *testing.T) {
-	fOpt := createTestFlagOptions()
-
-	tmpfile, err := os.CreateTemp(t.TempDir(), "testimage")
-	if err != nil {
-		t.Fatalf("Failed to create temp file: %v", err)
-	}
-	defer os.Remove(tmpfile.Name())
-
-	content := []byte("fake image data")
-	if _, err := tmpfile.Write(content); err != nil {
-		t.Fatalf("Failed to write to temp file: %v", err)
-	}
-	tmpfile.Close()
-
-	tmpfileName := tmpfile.Name()
-	fOpt.imageFile = &tmpfileName
-
-	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		// Verify the request contains image data
-		var req api.ChatRequest
-		body, _ := io.ReadAll(r.Body)
-		json.Unmarshal(body, &req)
-
-		if len(req.Messages) == 0 || len(req.Messages[0].Images) == 0 {
-			t.Error("Expected request to contain images")
-		}
-
-		w.Header().Set("Content-Type", "application/json")
-		response := api.ChatResponse{
-			Model: "test-model",
-			Message: api.Message{
-				Role:    "assistant",
-				Content: "test response with image",
-			},
-			Done: true,
-			Metrics: api.Metrics{
-				PromptEvalCount:    10,
-				PromptEvalDuration: 100 * time.Millisecond,
-				EvalCount:          50,
-				EvalDuration:       500 * time.Millisecond,
-				TotalDuration:      600 * time.Millisecond,
-				LoadDuration:       50 * time.Millisecond,
-			},
-		}
-		jsonData, _ := json.Marshal(response)
-		w.Write(jsonData)
-	}))
-	defer server.Close()
-
-	t.Setenv("OLLAMA_HOST", server.URL)
-
-	output := captureOutput(func() {
-		err := BenchmarkChat(fOpt)
-		if err != nil {
-			t.Errorf("Expected no error, got %v", err)
-		}
-	})
-
-	if !strings.Contains(output, "BenchmarkModel/name=test-model") {
-		t.Errorf("Expected benchmark output, got: %s", output)
-	}
-}
-
-func TestBenchmarkChat_ImageError(t *testing.T) {
-	randFileName := func() string {
-		const charset = "abcdefghijklmnopqrstuvwxyz0123456789"
-		const length = 8
-
-		result := make([]byte, length)
-		rand.Read(result) // Fill with random bytes
-
-		for i := range result {
-			result[i] = charset[result[i]%byte(len(charset))]
-		}
-
-		return string(result) + ".txt"
-	}
-
-	fOpt := createTestFlagOptions()
-	imageFile := randFileName()
-	fOpt.imageFile = &imageFile
-
-	output := captureOutput(func() {
-		err := BenchmarkChat(fOpt)
-		if err == nil {
-			t.Error("Expected error from image reading, got nil")
-		}
-	})
-
-	if !strings.Contains(output, "ERROR: Couldn't read image") {
-		t.Errorf("Expected image read error message, got: %s", output)
-	}
-}
-
-func TestReadImage_Success(t *testing.T) {
-	tmpfile, err := os.CreateTemp(t.TempDir(), "testimage")
-	if err != nil {
-		t.Fatalf("Failed to create temp file: %v", err)
-	}
-	defer os.Remove(tmpfile.Name())
-
-	content := []byte("fake image data")
-	if _, err := tmpfile.Write(content); err != nil {
-		t.Fatalf("Failed to write to temp file: %v", err)
-	}
-	tmpfile.Close()
-
-	imgData, err := readImage(tmpfile.Name())
-	if err != nil {
-		t.Errorf("Expected no error, got %v", err)
-	}
-
-	if imgData == nil {
-		t.Error("Expected image data, got nil")
-	}
-
-	expected := api.ImageData(content)
-	if string(imgData) != string(expected) {
-		t.Errorf("Expected image data %v, got %v", expected, imgData)
-	}
-}
-
-func TestReadImage_FileNotFound(t *testing.T) {
-	imgData, err := readImage("nonexistentfile.jpg")
-	if err == nil {
-		t.Error("Expected error for non-existent file, got nil")
-	}
-	if imgData != nil {
-		t.Error("Expected nil image data for non-existent file")
-	}
-}
-
-func TestOptionsMapCreation(t *testing.T) {
-	fOpt := createTestFlagOptions()
-
-	options := make(map[string]interface{})
-	if *fOpt.maxTokens > 0 {
-		options["num_predict"] = *fOpt.maxTokens
-	}
-	options["temperature"] = *fOpt.temperature
-	if fOpt.seed != nil && *fOpt.seed > 0 {
-		options["seed"] = *fOpt.seed
-	}
-
-	if options["num_predict"] != *fOpt.maxTokens {
-		t.Errorf("Expected num_predict %d, got %v", *fOpt.maxTokens, options["num_predict"])
-	}
-	if options["temperature"] != *fOpt.temperature {
-		t.Errorf("Expected temperature %f, got %v", *fOpt.temperature, options["temperature"])
-	}
-	if options["seed"] != *fOpt.seed {
-		t.Errorf("Expected seed %d, got %v", *fOpt.seed, options["seed"])
-	}
-}
--- a/cmd/cmd.go
+++ b/cmd/cmd.go
@@ -15,7 +15,6 @@ import (
 	"net"
 	"net/http"
 	"os"
-	"os/exec"
 	"os/signal"
 	"path/filepath"
 	"runtime"
@@ -30,15 +29,12 @@ import (
 	"github.com/containerd/console"
 	"github.com/mattn/go-runewidth"
 	"github.com/olekukonko/tablewriter"
-	"github.com/pkg/browser"
 	"github.com/spf13/cobra"
 	"golang.org/x/crypto/ssh"
 	"golang.org/x/sync/errgroup"
 	"golang.org/x/term"

 	"github.com/ollama/ollama/api"
-	"github.com/ollama/ollama/cmd/config"
-	"github.com/ollama/ollama/cmd/tui"
 	"github.com/ollama/ollama/envconfig"
 	"github.com/ollama/ollama/format"
 	"github.com/ollama/ollama/parser"
@@ -49,56 +45,9 @@ import (
 	"github.com/ollama/ollama/types/model"
 	"github.com/ollama/ollama/types/syncmap"
 	"github.com/ollama/ollama/version"
-	xcmd "github.com/ollama/ollama/x/cmd"
-	"github.com/ollama/ollama/x/create"
-	xcreateclient "github.com/ollama/ollama/x/create/client"
-	"github.com/ollama/ollama/x/imagegen"
 )

-func init() {
-	// Override default selectors to use Bubbletea TUI instead of raw terminal I/O.
-	config.DefaultSingleSelector = func(title string, items []config.ModelItem) (string, error) {
-		tuiItems := make([]tui.SelectItem, len(items))
-		for i, item := range items {
-			tuiItems[i] = tui.SelectItem{Name: item.Name, Description: item.Description, Recommended: item.Recommended}
-		}
-		result, err := tui.SelectSingle(title, tuiItems)
-		if errors.Is(err, tui.ErrCancelled) {
-			return "", config.ErrCancelled
-		}
-		return result, err
-	}
-
-	config.DefaultMultiSelector = func(title string, items []config.ModelItem, preChecked []string) ([]string, error) {
-		tuiItems := make([]tui.SelectItem, len(items))
-		for i, item := range items {
-			tuiItems[i] = tui.SelectItem{Name: item.Name, Description: item.Description, Recommended: item.Recommended}
-		}
-		result, err := tui.SelectMultiple(title, tuiItems, preChecked)
-		if errors.Is(err, tui.ErrCancelled) {
-			return nil, config.ErrCancelled
-		}
-		return result, err
-	}
-
-	config.DefaultSignIn = func(modelName, signInURL string) (string, error) {
-		userName, err := tui.RunSignIn(modelName, signInURL)
-		if errors.Is(err, tui.ErrCancelled) {
-			return "", config.ErrCancelled
-		}
-		return userName, err
-	}
-
-	config.DefaultConfirmPrompt = func(prompt string) (bool, error) {
-		ok, err := tui.RunConfirm(prompt)
-		if errors.Is(err, tui.ErrCancelled) {
-			return false, config.ErrCancelled
-		}
-		return ok, err
-	}
-}
-
-const ConnectInstructions = "If your browser did not open, navigate to:\n    %s\n\n"
+const ConnectInstructions = "To sign in, navigate to:\n    %s\n\n"

 // ensureThinkingSupport emits a warning if the model does not advertise thinking support
 func ensureThinkingSupport(ctx context.Context, client *api.Client, name string) {
@@ -141,88 +90,11 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 	p := progress.NewProgress(os.Stderr)
 	defer p.Stop()

-	// Validate model name early to fail fast
-	modelName := args[0]
-	name := model.ParseName(modelName)
-	if !name.IsValid() {
-		return fmt.Errorf("invalid model name: %s", modelName)
-	}
-
-	// Check for --experimental flag for safetensors model creation
-	experimental, _ := cmd.Flags().GetBool("experimental")
-	if experimental {
-		// Get Modelfile content - either from -f flag or default to "FROM ."
-		var reader io.Reader
-		filename, err := getModelfileName(cmd)
-		if os.IsNotExist(err) || filename == "" {
-			// No Modelfile specified or found - use default
-			reader = strings.NewReader("FROM .\n")
-		} else if err != nil {
-			return err
-		} else {
-			f, err := os.Open(filename)
-			if err != nil {
-				return err
-			}
-			defer f.Close()
-			reader = f
-		}
-
-		// Parse the Modelfile
-		modelfile, err := parser.ParseFile(reader)
-		if err != nil {
-			return fmt.Errorf("failed to parse Modelfile: %w", err)
-		}
-
-		// Extract FROM path and configuration
-		var modelDir string
-		mfConfig := &xcreateclient.ModelfileConfig{}
-
-		for _, cmd := range modelfile.Commands {
-			switch cmd.Name {
-			case "model":
-				modelDir = cmd.Args
-			case "template":
-				mfConfig.Template = cmd.Args
-			case "system":
-				mfConfig.System = cmd.Args
-			case "license":
-				mfConfig.License = cmd.Args
-			}
-		}
-
-		if modelDir == "" {
-			modelDir = "."
-		}
-
-		// Resolve relative paths based on Modelfile location
-		if !filepath.IsAbs(modelDir) && filename != "" {
-			modelDir = filepath.Join(filepath.Dir(filename), modelDir)
-		}
-
-		quantize, _ := cmd.Flags().GetString("quantize")
-		return xcreateclient.CreateModel(xcreateclient.CreateOptions{
-			ModelName: modelName,
-			ModelDir:  modelDir,
-			Quantize:  quantize,
-			Modelfile: mfConfig,
-		}, p)
-	}
-
 	var reader io.Reader

 	filename, err := getModelfileName(cmd)
 	if os.IsNotExist(err) {
 		if filename == "" {
-			// No Modelfile found - check if current directory is an image gen model
-			if create.IsTensorModelDir(".") {
-				quantize, _ := cmd.Flags().GetString("quantize")
-				return xcreateclient.CreateModel(xcreateclient.CreateOptions{
-					ModelName: modelName,
-					ModelDir:  ".",
-					Quantize:  quantize,
-				}, p)
-			}
 			reader = strings.NewReader("FROM .\n")
 		} else {
 			return errModelfileNotFound
@@ -254,7 +126,7 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 	}
 	spinner.Stop()

-	req.Model = modelName
+	req.Model = args[0]
 	quantize, _ := cmd.Flags().GetString("quantize")
 	if quantize != "" {
 		req.Quantize = quantize
@@ -412,25 +284,14 @@ func loadOrUnloadModel(cmd *cobra.Command, opts *runOptions) error {
 		return err
 	} else if info.RemoteHost != "" {
 		// Cloud model, no need to load/unload
-
-		isCloud := strings.HasPrefix(info.RemoteHost, "https://ollama.com")
-
-		// Check if user is signed in for ollama.com cloud models
-		if isCloud {
-			if _, err := client.Whoami(cmd.Context()); err != nil {
-				return err
-			}
-		}
-
 		if opts.ShowConnect {
 			p.StopAndClear()
-			if isCloud {
+			if strings.HasPrefix(info.RemoteHost, "https://ollama.com") {
 				fmt.Fprintf(os.Stderr, "Connecting to '%s' on 'ollama.com' ⚡\n", info.RemoteModel)
 			} else {
 				fmt.Fprintf(os.Stderr, "Connecting to '%s' on '%s'\n", info.RemoteModel, info.RemoteHost)
 			}
 		}
-
 		return nil
 	}

@@ -595,7 +456,6 @@ func RunHandler(cmd *cobra.Command, args []string) error {
 	}

 	name := args[0]
-
 	info, err := func() (*api.ShowResponse, error) {
 		showReq := &api.ShowRequest{Name: name}
 		info, err := client.Show(cmd.Context(), showReq)
@@ -657,19 +517,6 @@ func RunHandler(cmd *cobra.Command, args []string) error {
 		return generateEmbedding(cmd, name, opts.Prompt, opts.KeepAlive, truncate, dimensions)
 	}

-	// Check if this is an image generation model
-	if slices.Contains(info.Capabilities, model.CapabilityImage) {
-		if opts.Prompt == "" && !interactive {
-			return errors.New("image generation models require a prompt. Usage: ollama run " + name + " \"your prompt here\"")
-		}
-		return imagegen.RunCLI(cmd, name, opts.Prompt, interactive, opts.KeepAlive)
-	}
-
-	// Check for experimental flag
-	isExperimental, _ := cmd.Flags().GetBool("experimental")
-	yoloMode, _ := cmd.Flags().GetBool("experimental-yolo")
-	enableWebsearch, _ := cmd.Flags().GetBool("experimental-websearch")
-
 	if interactive {
 		if err := loadOrUnloadModel(cmd, &opts); err != nil {
 			var sErr api.AuthorizationError
@@ -696,11 +543,6 @@ func RunHandler(cmd *cobra.Command, args []string) error {
 			}
 		}

-		// Use experimental agent loop with tools
-		if isExperimental {
-			return xcmd.GenerateInteractive(cmd, opts.Model, opts.WordWrap, opts.Options, opts.Think, opts.HideThinking, opts.KeepAlive, yoloMode, enableWebsearch)
-		}
-
 		return generateInteractive(cmd, opts)
 	}
 	return generate(cmd, opts)
@@ -720,7 +562,6 @@ func SigninHandler(cmd *cobra.Command, args []string) error {
 			fmt.Println()

 			if aErr.SigninURL != "" {
-				_ = browser.OpenURL(aErr.SigninURL)
 				fmt.Printf(ConnectInstructions, aErr.SigninURL)
 			}
 			return nil
@@ -805,11 +646,7 @@ func PushHandler(cmd *cobra.Command, args []string) error {

 			bar, ok := bars[resp.Digest]
 			if !ok {
-				msg := resp.Status
-				if msg == "" {
-					msg = fmt.Sprintf("pushing %s...", resp.Digest[7:19])
-				}
-				bar = progress.NewBar(msg, resp.Total, resp.Completed)
+				bar = progress.NewBar(fmt.Sprintf("pushing %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
 				bars[resp.Digest] = bar
 				p.Add(resp.Digest, bar)
 			}
@@ -958,11 +795,11 @@ func DeleteHandler(cmd *cobra.Command, args []string) error {
 	for _, arg := range args {
 		// Unload the model if it's running before deletion
 		if err := loadOrUnloadModel(cmd, &runOptions{
-			Model:     arg,
+			Model:     args[0],
 			KeepAlive: &api.Duration{Duration: 0},
 		}); err != nil {
 			if !strings.Contains(strings.ToLower(err.Error()), "not found") {
-				fmt.Fprintf(os.Stderr, "Warning: unable to stop model '%s'\n", arg)
+				fmt.Fprintf(os.Stderr, "Warning: unable to stop model '%s'\n", args[0])
 			}
 		}

@@ -1077,10 +914,8 @@ func showInfo(resp *api.ShowResponse, verbose bool, w io.Writer) error {
 		}

 		if resp.ModelInfo != nil {
-			arch, _ := resp.ModelInfo["general.architecture"].(string)
-			if arch != "" {
-				rows = append(rows, []string{"", "architecture", arch})
-			}
+			arch := resp.ModelInfo["general.architecture"].(string)
+			rows = append(rows, []string{"", "architecture", arch})

 			var paramStr string
 			if resp.Details.ParameterSize != "" {
@@ -1090,9 +925,7 @@ func showInfo(resp *api.ShowResponse, verbose bool, w io.Writer) error {
 					paramStr = format.HumanNumber(uint64(f))
 				}
 			}
-			if paramStr != "" {
-				rows = append(rows, []string{"", "parameters", paramStr})
-			}
+			rows = append(rows, []string{"", "parameters", paramStr})

 			if v, ok := resp.ModelInfo[fmt.Sprintf("%s.context_length", arch)]; ok {
 				if f, ok := v.(float64); ok {
@@ -1110,9 +943,6 @@ func showInfo(resp *api.ShowResponse, verbose bool, w io.Writer) error {
 			rows = append(rows, []string{"", "parameters", resp.Details.ParameterSize})
 		}
 		rows = append(rows, []string{"", "quantization", resp.Details.QuantizationLevel})
-		if resp.Requires != "" {
-			rows = append(rows, []string{"", "requires", resp.Requires})
-		}
 		return
 	})

@@ -1600,7 +1430,7 @@ func chat(cmd *cobra.Command, opts runOptions) (*api.Message, error) {
 		latest.Summary()
 	}

-	return &api.Message{Role: role, Thinking: thinkingContent.String(), Content: fullResponse.String()}, nil
+	return &api.Message{Role: role, Content: fullResponse.String()}, nil
 }

 func generate(cmd *cobra.Command, opts runOptions) error {
@@ -1808,7 +1638,7 @@ func checkServerHeartbeat(cmd *cobra.Command, _ []string) error {
 			return err
 		}
 		if err := startApp(cmd.Context(), client); err != nil {
-			return err
+			return fmt.Errorf("ollama server not responding - %w", err)
 		}
 	}
 	return nil
@@ -1849,197 +1679,6 @@ Environment Variables:
 	cmd.SetUsageTemplate(cmd.UsageTemplate() + envUsage)
 }

-// ensureServerRunning checks if the ollama server is running and starts it in the background if not.
-func ensureServerRunning(ctx context.Context) error {
-	client, err := api.ClientFromEnvironment()
-	if err != nil {
-		return err
-	}
-
-	// Check if server is already running
-	if err := client.Heartbeat(ctx); err == nil {
-		return nil // server is already running
-	}
-
-	// Server not running, start it in the background
-	exe, err := os.Executable()
-	if err != nil {
-		return fmt.Errorf("could not find executable: %w", err)
-	}
-
-	serverCmd := exec.CommandContext(ctx, exe, "serve")
-	serverCmd.Env = os.Environ()
-	serverCmd.SysProcAttr = backgroundServerSysProcAttr()
-	if err := serverCmd.Start(); err != nil {
-		return fmt.Errorf("failed to start server: %w", err)
-	}
-
-	// Wait for the server to be ready
-	for {
-		time.Sleep(500 * time.Millisecond)
-		if err := client.Heartbeat(ctx); err == nil {
-			return nil // server has started
-		}
-	}
-}
-
-// runInteractiveTUI runs the main interactive TUI menu.
-func runInteractiveTUI(cmd *cobra.Command) {
-	// Ensure the server is running before showing the TUI
-	if err := ensureServerRunning(cmd.Context()); err != nil {
-		fmt.Fprintf(os.Stderr, "Error starting server: %v\n", err)
-		return
-	}
-
-	// Selector adapters for tui
-	singleSelector := func(title string, items []config.ModelItem) (string, error) {
-		tuiItems := make([]tui.SelectItem, len(items))
-		for i, item := range items {
-			tuiItems[i] = tui.SelectItem{Name: item.Name, Description: item.Description, Recommended: item.Recommended}
-		}
-		result, err := tui.SelectSingle(title, tuiItems)
-		if errors.Is(err, tui.ErrCancelled) {
-			return "", config.ErrCancelled
-		}
-		return result, err
-	}
-
-	multiSelector := func(title string, items []config.ModelItem, preChecked []string) ([]string, error) {
-		tuiItems := make([]tui.SelectItem, len(items))
-		for i, item := range items {
-			tuiItems[i] = tui.SelectItem{Name: item.Name, Description: item.Description, Recommended: item.Recommended}
-		}
-		result, err := tui.SelectMultiple(title, tuiItems, preChecked)
-		if errors.Is(err, tui.ErrCancelled) {
-			return nil, config.ErrCancelled
-		}
-		return result, err
-	}
-
-	for {
-		result, err := tui.Run()
-		if err != nil {
-			fmt.Fprintf(os.Stderr, "Error: %v\n", err)
-			return
-		}
-
-		runModel := func(modelName string) {
-			client, err := api.ClientFromEnvironment()
-			if err != nil {
-				fmt.Fprintf(os.Stderr, "Error: %v\n", err)
-				return
-			}
-			if err := config.ShowOrPull(cmd.Context(), client, modelName); err != nil {
-				if errors.Is(err, config.ErrCancelled) {
-					return
-				}
-				fmt.Fprintf(os.Stderr, "Error: %v\n", err)
-				return
-			}
-			_ = config.SetLastModel(modelName)
-			opts := runOptions{
-				Model:       modelName,
-				WordWrap:    os.Getenv("TERM") == "xterm-256color",
-				Options:     map[string]any{},
-				ShowConnect: true,
-			}
-			if err := loadOrUnloadModel(cmd, &opts); err != nil {
-				fmt.Fprintf(os.Stderr, "Error loading model: %v\n", err)
-				return
-			}
-			if err := generateInteractive(cmd, opts); err != nil {
-				fmt.Fprintf(os.Stderr, "Error running model: %v\n", err)
-			}
-		}
-
-		launchIntegration := func(name string) bool {
-			// If not configured or model no longer exists, prompt for model selection
-			configuredModel := config.IntegrationModel(name)
-			if configuredModel == "" || !config.ModelExists(cmd.Context(), configuredModel) {
-				err := config.ConfigureIntegrationWithSelectors(cmd.Context(), name, singleSelector, multiSelector)
-				if errors.Is(err, config.ErrCancelled) {
-					return false // Return to main menu
-				}
-				if err != nil {
-					fmt.Fprintf(os.Stderr, "Error configuring %s: %v\n", name, err)
-					return true
-				}
-			}
-			if err := config.LaunchIntegration(name); err != nil {
-				fmt.Fprintf(os.Stderr, "Error launching %s: %v\n", name, err)
-			}
-			return true
-		}
-
-		switch result.Selection {
-		case tui.SelectionNone:
-			// User quit
-			return
-		case tui.SelectionRunModel:
-			_ = config.SetLastSelection("run")
-			if modelName := config.LastModel(); modelName != "" {
-				runModel(modelName)
-			} else {
-				modelName, err := config.SelectModelWithSelector(cmd.Context(), singleSelector)
-				if errors.Is(err, config.ErrCancelled) {
-					continue // Return to main menu
-				}
-				if err != nil {
-					fmt.Fprintf(os.Stderr, "Error selecting model: %v\n", err)
-					continue
-				}
-				runModel(modelName)
-			}
-		case tui.SelectionChangeRunModel:
-			_ = config.SetLastSelection("run")
-			// Use model from modal if selected, otherwise show picker
-			modelName := result.Model
-			if modelName == "" {
-				var err error
-				modelName, err = config.SelectModelWithSelector(cmd.Context(), singleSelector)
-				if errors.Is(err, config.ErrCancelled) {
-					continue // Return to main menu
-				}
-				if err != nil {
-					fmt.Fprintf(os.Stderr, "Error selecting model: %v\n", err)
-					continue
-				}
-			}
-			runModel(modelName)
-		case tui.SelectionIntegration:
-			_ = config.SetLastSelection(result.Integration)
-			if !launchIntegration(result.Integration) {
-				continue // Return to main menu
-			}
-		case tui.SelectionChangeIntegration:
-			_ = config.SetLastSelection(result.Integration)
-			// Use model from modal if selected, otherwise show picker
-			if result.Model != "" {
-				// Model already selected from modal - save and launch
-				if err := config.SaveIntegrationModel(result.Integration, result.Model); err != nil {
-					fmt.Fprintf(os.Stderr, "Error saving config: %v\n", err)
-					continue
-				}
-				if err := config.LaunchIntegrationWithModel(result.Integration, result.Model); err != nil {
-					fmt.Fprintf(os.Stderr, "Error launching %s: %v\n", result.Integration, err)
-				}
-			} else {
-				err := config.ConfigureIntegrationWithSelectors(cmd.Context(), result.Integration, singleSelector, multiSelector)
-				if errors.Is(err, config.ErrCancelled) {
-					continue // Return to main menu
-				}
-				if err != nil {
-					fmt.Fprintf(os.Stderr, "Error configuring %s: %v\n", result.Integration, err)
-					continue
-				}
-				if err := config.LaunchIntegration(result.Integration); err != nil {
-					fmt.Fprintf(os.Stderr, "Error launching %s: %v\n", result.Integration, err)
-				}
-			}
-		}
-	}
-}
-
 func NewCLI() *cobra.Command {
 	log.SetFlags(log.LstdFlags | log.Lshortfile)
 	cobra.EnableCommandSorting = false
@@ -2062,31 +1701,22 @@ func NewCLI() *cobra.Command {
 				return
 			}

-			runInteractiveTUI(cmd)
+			cmd.Print(cmd.UsageString())
 		},
 	}

 	rootCmd.Flags().BoolP("version", "v", false, "Show version information")
-	rootCmd.Flags().Bool("verbose", false, "Show timings for response")
-	rootCmd.Flags().Bool("nowordwrap", false, "Don't wrap words to the next line automatically")

 	createCmd := &cobra.Command{
-		Use:   "create MODEL",
-		Short: "Create a model",
-		Args:  cobra.ExactArgs(1),
-		PreRunE: func(cmd *cobra.Command, args []string) error {
-			// Skip server check for experimental mode (writes directly to disk)
-			if experimental, _ := cmd.Flags().GetBool("experimental"); experimental {
-				return nil
-			}
-			return checkServerHeartbeat(cmd, args)
-		},
-		RunE: CreateHandler,
+		Use:     "create MODEL",
+		Short:   "Create a model",
+		Args:    cobra.ExactArgs(1),
+		PreRunE: checkServerHeartbeat,
+		RunE:    CreateHandler,
 	}

 	createCmd.Flags().StringP("file", "f", "", "Name of the Modelfile (default \"Modelfile\")")
 	createCmd.Flags().StringP("quantize", "q", "", "Quantize model to this level (e.g. q4_K_M)")
-	createCmd.Flags().Bool("experimental", false, "Enable experimental safetensors model creation")

 	showCmd := &cobra.Command{
 		Use:     "show MODEL",
@@ -2121,12 +1751,6 @@ func NewCLI() *cobra.Command {
 	runCmd.Flags().Bool("hidethinking", false, "Hide thinking output (if provided)")
 	runCmd.Flags().Bool("truncate", false, "For embedding models: truncate inputs exceeding context length (default: true). Set --truncate=false to error instead")
 	runCmd.Flags().Int("dimensions", 0, "Truncate output embeddings to specified dimension (embedding models only)")
-	runCmd.Flags().Bool("experimental", false, "Enable experimental agent loop with tools")
-	runCmd.Flags().Bool("experimental-yolo", false, "Skip all tool approval prompts (use with caution)")
-	runCmd.Flags().Bool("experimental-websearch", false, "Enable web search tool in experimental mode")
-
-	// Image generation flags (width, height, steps, seed, etc.)
-	imagegen.RegisterFlags(runCmd)

 	stopCmd := &cobra.Command{
 		Use:     "stop MODEL",
@@ -2139,7 +1763,7 @@ func NewCLI() *cobra.Command {
 	serveCmd := &cobra.Command{
 		Use:     "serve",
 		Aliases: []string{"start"},
-		Short:   "Start Ollama",
+		Short:   "Start ollama",
 		Args:    cobra.ExactArgs(0),
 		RunE:    RunServer,
 	}
@@ -2172,15 +1796,6 @@ func NewCLI() *cobra.Command {
 		RunE:    SigninHandler,
 	}

-	loginCmd := &cobra.Command{
-		Use:     "login",
-		Short:   "Sign in to ollama.com",
-		Hidden:  true,
-		Args:    cobra.ExactArgs(0),
-		PreRunE: checkServerHeartbeat,
-		RunE:    SigninHandler,
-	}
-
 	signoutCmd := &cobra.Command{
 		Use:     "signout",
 		Short:   "Sign out from ollama.com",
@@ -2189,15 +1804,6 @@ func NewCLI() *cobra.Command {
 		RunE:    SignoutHandler,
 	}

-	logoutCmd := &cobra.Command{
-		Use:     "logout",
-		Short:   "Sign out from ollama.com",
-		Hidden:  true,
-		Args:    cobra.ExactArgs(0),
-		PreRunE: checkServerHeartbeat,
-		RunE:    SignoutHandler,
-	}
-
 	listCmd := &cobra.Command{
 		Use:     "list",
 		Aliases: []string{"ls"},
@@ -2259,7 +1865,6 @@ func NewCLI() *cobra.Command {
 	} {
 		switch cmd {
 		case runCmd:
-			imagegen.AppendFlagsDocs(cmd)
 			appendEnvDocs(cmd, []envconfig.EnvVar{envVars["OLLAMA_HOST"], envVars["OLLAMA_NOHISTORY"]})
 		case serveCmd:
 			appendEnvDocs(cmd, []envconfig.EnvVar{
@@ -2294,15 +1899,12 @@ func NewCLI() *cobra.Command {
 		pullCmd,
 		pushCmd,
 		signinCmd,
-		loginCmd,
 		signoutCmd,
-		logoutCmd,
 		listCmd,
 		psCmd,
 		copyCmd,
 		deleteCmd,
 		runnerCmd,
-		config.LaunchCmd(checkServerHeartbeat, runInteractiveTUI),
 	)

 	return rootCmd
--- a/cmd/cmd_test.go
+++ b/cmd/cmd_test.go
@@ -3,7 +3,6 @@ package cmd
 import (
 	"bytes"
 	"encoding/json"
-	"errors"
 	"fmt"
 	"io"
 	"net/http"
@@ -292,31 +291,6 @@ Weigh anchor!
 			t.Errorf("unexpected output (-want +got):\n%s", diff)
 		}
 	})
-
-	t.Run("min version", func(t *testing.T) {
-		var b bytes.Buffer
-		if err := showInfo(&api.ShowResponse{
-			Details: api.ModelDetails{
-				Family:            "test",
-				ParameterSize:     "7B",
-				QuantizationLevel: "FP16",
-			},
-			Requires: "0.14.0",
-		}, false, &b); err != nil {
-			t.Fatal(err)
-		}
-
-		expect := `  Model
-    architecture    test      
-    parameters      7B        
-    quantization    FP16      
-    requires        0.14.0    
-
-`
-		if diff := cmp.Diff(expect, b.String()); diff != "" {
-			t.Errorf("unexpected output (-want +got):\n%s", diff)
-		}
-	})
 }

 func TestDeleteHandler(t *testing.T) {
@@ -1548,79 +1522,6 @@ func TestRunOptions_Copy_ThinkValueVariants(t *testing.T) {
 	}
 }

-func TestShowInfoImageGen(t *testing.T) {
-	var b bytes.Buffer
-	err := showInfo(&api.ShowResponse{
-		Details: api.ModelDetails{
-			Family:            "ZImagePipeline",
-			ParameterSize:     "10.3B",
-			QuantizationLevel: "Q8",
-		},
-		Capabilities: []model.Capability{model.CapabilityImage},
-		Requires:     "0.14.0",
-	}, false, &b)
-	if err != nil {
-		t.Fatal(err)
-	}
-
-	expect := "  Model\n" +
-		"    architecture    ZImagePipeline    \n" +
-		"    parameters      10.3B             \n" +
-		"    quantization    Q8                \n" +
-		"    requires        0.14.0            \n" +
-		"\n" +
-		"  Capabilities\n" +
-		"    image    \n" +
-		"\n"
-	if diff := cmp.Diff(expect, b.String()); diff != "" {
-		t.Errorf("unexpected output (-want +got):\n%s", diff)
-	}
-}
-
-func TestPushProgressMessage(t *testing.T) {
-	tests := []struct {
-		name    string
-		status  string
-		digest  string
-		wantMsg string
-	}{
-		{
-			name:    "uses status when provided",
-			status:  "uploading model",
-			digest:  "sha256:abc123456789def",
-			wantMsg: "uploading model",
-		},
-		{
-			name:    "falls back to digest when status empty",
-			status:  "",
-			digest:  "sha256:abc123456789def",
-			wantMsg: "pushing abc123456789...",
-		},
-		{
-			name:    "handles short digest gracefully",
-			status:  "",
-			digest:  "sha256:abc",
-			wantMsg: "pushing sha256:abc...",
-		},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			msg := tt.status
-			if msg == "" {
-				if len(tt.digest) >= 19 {
-					msg = fmt.Sprintf("pushing %s...", tt.digest[7:19])
-				} else {
-					msg = fmt.Sprintf("pushing %s...", tt.digest)
-				}
-			}
-			if msg != tt.wantMsg {
-				t.Errorf("got %q, want %q", msg, tt.wantMsg)
-			}
-		})
-	}
-}
-
 func TestRunOptions_Copy_Independence(t *testing.T) {
 	// Test that modifications to original don't affect copy
 	originalThink := &api.ThinkValue{Value: "original"}
@@ -1660,103 +1561,3 @@ func TestRunOptions_Copy_Independence(t *testing.T) {
 		t.Error("Copy Think should not be affected by original modification")
 	}
 }
-
-func TestLoadOrUnloadModel_CloudModelAuth(t *testing.T) {
-	tests := []struct {
-		name          string
-		remoteHost    string
-		whoamiStatus  int
-		whoamiResp    any
-		expectedError string
-	}{
-		{
-			name:         "ollama.com cloud model - user signed in",
-			remoteHost:   "https://ollama.com",
-			whoamiStatus: http.StatusOK,
-			whoamiResp:   api.UserResponse{Name: "testuser"},
-		},
-		{
-			name:         "ollama.com cloud model - user not signed in",
-			remoteHost:   "https://ollama.com",
-			whoamiStatus: http.StatusUnauthorized,
-			whoamiResp: map[string]string{
-				"error":      "unauthorized",
-				"signin_url": "https://ollama.com/signin",
-			},
-			expectedError: "unauthorized",
-		},
-		{
-			name:         "non-ollama.com remote - no auth check",
-			remoteHost:   "https://other-remote.com",
-			whoamiStatus: http.StatusUnauthorized, // should not be called
-			whoamiResp:   nil,
-		},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			whoamiCalled := false
-			mockServer := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-				switch r.URL.Path {
-				case "/api/show":
-					w.Header().Set("Content-Type", "application/json")
-					if err := json.NewEncoder(w).Encode(api.ShowResponse{
-						RemoteHost:  tt.remoteHost,
-						RemoteModel: "test-model",
-					}); err != nil {
-						http.Error(w, err.Error(), http.StatusInternalServerError)
-					}
-				case "/api/me":
-					whoamiCalled = true
-					w.Header().Set("Content-Type", "application/json")
-					w.WriteHeader(tt.whoamiStatus)
-					if tt.whoamiResp != nil {
-						if err := json.NewEncoder(w).Encode(tt.whoamiResp); err != nil {
-							http.Error(w, err.Error(), http.StatusInternalServerError)
-						}
-					}
-				default:
-					http.NotFound(w, r)
-				}
-			}))
-			defer mockServer.Close()
-
-			t.Setenv("OLLAMA_HOST", mockServer.URL)
-
-			cmd := &cobra.Command{}
-			cmd.SetContext(t.Context())
-
-			opts := &runOptions{
-				Model:       "test-cloud-model",
-				ShowConnect: false,
-			}
-
-			err := loadOrUnloadModel(cmd, opts)
-
-			if strings.HasPrefix(tt.remoteHost, "https://ollama.com") {
-				if !whoamiCalled {
-					t.Error("expected whoami to be called for ollama.com cloud model")
-				}
-			} else {
-				if whoamiCalled {
-					t.Error("whoami should not be called for non-ollama.com remote")
-				}
-			}
-
-			if tt.expectedError != "" {
-				if err == nil {
-					t.Errorf("expected error containing %q, got nil", tt.expectedError)
-				} else {
-					var authErr api.AuthorizationError
-					if !errors.As(err, &authErr) {
-						t.Errorf("expected AuthorizationError, got %T: %v", err, err)
-					}
-				}
-			} else {
-				if err != nil {
-					t.Errorf("expected no error, got %v", err)
-				}
-			}
-		})
-	}
-}
--- a/cmd/config/claude.go
+++ b/cmd/config/claude.go
@@ -1,192 +0,0 @@
-package config
-
-import (
-	"context"
-	"fmt"
-	"os"
-	"os/exec"
-	"path/filepath"
-	"runtime"
-
-	"github.com/ollama/ollama/api"
-	"github.com/ollama/ollama/envconfig"
-)
-
-// Claude implements Runner and AliasConfigurer for Claude Code integration
-type Claude struct{}
-
-// Compile-time check that Claude implements AliasConfigurer
-var _ AliasConfigurer = (*Claude)(nil)
-
-func (c *Claude) String() string { return "Claude Code" }
-
-func (c *Claude) args(model string, extra []string) []string {
-	var args []string
-	if model != "" {
-		args = append(args, "--model", model)
-	}
-	args = append(args, extra...)
-	return args
-}
-
-func (c *Claude) findPath() (string, error) {
-	if p, err := exec.LookPath("claude"); err == nil {
-		return p, nil
-	}
-	home, err := os.UserHomeDir()
-	if err != nil {
-		return "", err
-	}
-	name := "claude"
-	if runtime.GOOS == "windows" {
-		name = "claude.exe"
-	}
-	fallback := filepath.Join(home, ".claude", "local", name)
-	if _, err := os.Stat(fallback); err != nil {
-		return "", err
-	}
-	return fallback, nil
-}
-
-func (c *Claude) Run(model string, args []string) error {
-	claudePath, err := c.findPath()
-	if err != nil {
-		return fmt.Errorf("claude is not installed, install from https://code.claude.com/docs/en/quickstart")
-	}
-
-	cmd := exec.Command(claudePath, c.args(model, args)...)
-	cmd.Stdin = os.Stdin
-	cmd.Stdout = os.Stdout
-	cmd.Stderr = os.Stderr
-
-	env := append(os.Environ(),
-		"ANTHROPIC_BASE_URL="+envconfig.Host().String(),
-		"ANTHROPIC_API_KEY=",
-		"ANTHROPIC_AUTH_TOKEN=ollama",
-	)
-
-	env = append(env, c.modelEnvVars(model)...)
-
-	cmd.Env = env
-	return cmd.Run()
-}
-
-// modelEnvVars returns Claude Code env vars that route all model tiers through Ollama.
-func (c *Claude) modelEnvVars(model string) []string {
-	primary := model
-	fast := model
-	if cfg, err := loadIntegration("claude"); err == nil && cfg.Aliases != nil {
-		if p := cfg.Aliases["primary"]; p != "" {
-			primary = p
-		}
-		if f := cfg.Aliases["fast"]; f != "" {
-			fast = f
-		}
-	}
-	return []string{
-		"ANTHROPIC_DEFAULT_OPUS_MODEL=" + primary,
-		"ANTHROPIC_DEFAULT_SONNET_MODEL=" + primary,
-		"ANTHROPIC_DEFAULT_HAIKU_MODEL=" + fast,
-		"CLAUDE_CODE_SUBAGENT_MODEL=" + primary,
-	}
-}
-
-// ConfigureAliases sets up model aliases for Claude Code.
-// model: the model to use (if empty, user will be prompted to select)
-// aliases: existing alias configuration to preserve/update
-// Cloud-only: subagent routing (fast model) is gated to cloud models only until
-// there is a better strategy for prompt caching on local models.
-func (c *Claude) ConfigureAliases(ctx context.Context, model string, existingAliases map[string]string, force bool) (map[string]string, bool, error) {
-	aliases := make(map[string]string)
-	for k, v := range existingAliases {
-		aliases[k] = v
-	}
-
-	if model != "" {
-		aliases["primary"] = model
-	}
-
-	if !force && aliases["primary"] != "" {
-		client, _ := api.ClientFromEnvironment()
-		if isCloudModel(ctx, client, aliases["primary"]) {
-			if isCloudModel(ctx, client, aliases["fast"]) {
-				return aliases, false, nil
-			}
-		} else {
-			delete(aliases, "fast")
-			return aliases, false, nil
-		}
-	}
-
-	items, existingModels, cloudModels, client, err := listModels(ctx)
-	if err != nil {
-		return nil, false, err
-	}
-
-	fmt.Fprintf(os.Stderr, "\n%sModel Configuration%s\n\n", ansiBold, ansiReset)
-
-	if aliases["primary"] == "" || force {
-		primary, err := DefaultSingleSelector("Select model:", items)
-		if err != nil {
-			return nil, false, err
-		}
-		if err := pullIfNeeded(ctx, client, existingModels, primary); err != nil {
-			return nil, false, err
-		}
-		if err := ensureAuth(ctx, client, cloudModels, []string{primary}); err != nil {
-			return nil, false, err
-		}
-		aliases["primary"] = primary
-	}
-
-	if isCloudModel(ctx, client, aliases["primary"]) {
-		if aliases["fast"] == "" || !isCloudModel(ctx, client, aliases["fast"]) {
-			aliases["fast"] = aliases["primary"]
-		}
-	} else {
-		delete(aliases, "fast")
-	}
-
-	return aliases, true, nil
-}
-
-// SetAliases syncs the configured aliases to the Ollama server using prefix matching.
-// Cloud-only: for local models (fast is empty), we delete any existing aliases to
-// prevent stale routing to a previous cloud model.
-func (c *Claude) SetAliases(ctx context.Context, aliases map[string]string) error {
-	client, err := api.ClientFromEnvironment()
-	if err != nil {
-		return err
-	}
-
-	prefixes := []string{"claude-sonnet-", "claude-haiku-"}
-
-	if aliases["fast"] == "" {
-		for _, prefix := range prefixes {
-			_ = client.DeleteAliasExperimental(ctx, &api.AliasDeleteRequest{Alias: prefix})
-		}
-		return nil
-	}
-
-	prefixAliases := map[string]string{
-		"claude-sonnet-": aliases["primary"],
-		"claude-haiku-":  aliases["fast"],
-	}
-
-	var errs []string
-	for prefix, target := range prefixAliases {
-		req := &api.AliasRequest{
-			Alias:          prefix,
-			Target:         target,
-			PrefixMatching: true,
-		}
-		if err := client.SetAliasExperimental(ctx, req); err != nil {
-			errs = append(errs, prefix)
-		}
-	}
-
-	if len(errs) > 0 {
-		return fmt.Errorf("failed to set aliases: %v", errs)
-	}
-	return nil
-}
--- a/cmd/config/claude_test.go
+++ b/cmd/config/claude_test.go
@@ -1,198 +0,0 @@
-package config
-
-import (
-	"os"
-	"path/filepath"
-	"runtime"
-	"slices"
-	"strings"
-	"testing"
-)
-
-func TestClaudeIntegration(t *testing.T) {
-	c := &Claude{}
-
-	t.Run("String", func(t *testing.T) {
-		if got := c.String(); got != "Claude Code" {
-			t.Errorf("String() = %q, want %q", got, "Claude Code")
-		}
-	})
-
-	t.Run("implements Runner", func(t *testing.T) {
-		var _ Runner = c
-	})
-}
-
-func TestClaudeFindPath(t *testing.T) {
-	c := &Claude{}
-
-	t.Run("finds claude in PATH", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		name := "claude"
-		if runtime.GOOS == "windows" {
-			name = "claude.exe"
-		}
-		fakeBin := filepath.Join(tmpDir, name)
-		os.WriteFile(fakeBin, []byte("#!/bin/sh\n"), 0o755)
-		t.Setenv("PATH", tmpDir)
-
-		got, err := c.findPath()
-		if err != nil {
-			t.Fatalf("unexpected error: %v", err)
-		}
-		if got != fakeBin {
-			t.Errorf("findPath() = %q, want %q", got, fakeBin)
-		}
-	})
-
-	t.Run("falls back to ~/.claude/local/claude", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-		t.Setenv("PATH", t.TempDir()) // empty dir, no claude binary
-
-		name := "claude"
-		if runtime.GOOS == "windows" {
-			name = "claude.exe"
-		}
-		fallback := filepath.Join(tmpDir, ".claude", "local", name)
-		os.MkdirAll(filepath.Dir(fallback), 0o755)
-		os.WriteFile(fallback, []byte("#!/bin/sh\n"), 0o755)
-
-		got, err := c.findPath()
-		if err != nil {
-			t.Fatalf("unexpected error: %v", err)
-		}
-		if got != fallback {
-			t.Errorf("findPath() = %q, want %q", got, fallback)
-		}
-	})
-
-	t.Run("returns error when neither PATH nor fallback exists", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-		t.Setenv("PATH", t.TempDir()) // empty dir, no claude binary
-
-		_, err := c.findPath()
-		if err == nil {
-			t.Fatal("expected error, got nil")
-		}
-	})
-}
-
-func TestClaudeArgs(t *testing.T) {
-	c := &Claude{}
-
-	tests := []struct {
-		name  string
-		model string
-		args  []string
-		want  []string
-	}{
-		{"with model", "llama3.2", nil, []string{"--model", "llama3.2"}},
-		{"empty model", "", nil, nil},
-		{"with model and verbose", "llama3.2", []string{"--verbose"}, []string{"--model", "llama3.2", "--verbose"}},
-		{"empty model with help", "", []string{"--help"}, []string{"--help"}},
-		{"with allowed tools", "llama3.2", []string{"--allowedTools", "Read,Write,Bash"}, []string{"--model", "llama3.2", "--allowedTools", "Read,Write,Bash"}},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			got := c.args(tt.model, tt.args)
-			if !slices.Equal(got, tt.want) {
-				t.Errorf("args(%q, %v) = %v, want %v", tt.model, tt.args, got, tt.want)
-			}
-		})
-	}
-}
-
-func TestClaudeModelEnvVars(t *testing.T) {
-	c := &Claude{}
-
-	envMap := func(envs []string) map[string]string {
-		m := make(map[string]string)
-		for _, e := range envs {
-			k, v, _ := strings.Cut(e, "=")
-			m[k] = v
-		}
-		return m
-	}
-
-	t.Run("falls back to model param when no aliases saved", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		got := envMap(c.modelEnvVars("llama3.2"))
-		if got["ANTHROPIC_DEFAULT_OPUS_MODEL"] != "llama3.2" {
-			t.Errorf("OPUS = %q, want llama3.2", got["ANTHROPIC_DEFAULT_OPUS_MODEL"])
-		}
-		if got["ANTHROPIC_DEFAULT_SONNET_MODEL"] != "llama3.2" {
-			t.Errorf("SONNET = %q, want llama3.2", got["ANTHROPIC_DEFAULT_SONNET_MODEL"])
-		}
-		if got["ANTHROPIC_DEFAULT_HAIKU_MODEL"] != "llama3.2" {
-			t.Errorf("HAIKU = %q, want llama3.2", got["ANTHROPIC_DEFAULT_HAIKU_MODEL"])
-		}
-		if got["CLAUDE_CODE_SUBAGENT_MODEL"] != "llama3.2" {
-			t.Errorf("SUBAGENT = %q, want llama3.2", got["CLAUDE_CODE_SUBAGENT_MODEL"])
-		}
-	})
-
-	t.Run("uses primary alias for opus sonnet and subagent", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		saveIntegration("claude", []string{"qwen3:8b"})
-		saveAliases("claude", map[string]string{"primary": "qwen3:8b"})
-
-		got := envMap(c.modelEnvVars("qwen3:8b"))
-		if got["ANTHROPIC_DEFAULT_OPUS_MODEL"] != "qwen3:8b" {
-			t.Errorf("OPUS = %q, want qwen3:8b", got["ANTHROPIC_DEFAULT_OPUS_MODEL"])
-		}
-		if got["ANTHROPIC_DEFAULT_SONNET_MODEL"] != "qwen3:8b" {
-			t.Errorf("SONNET = %q, want qwen3:8b", got["ANTHROPIC_DEFAULT_SONNET_MODEL"])
-		}
-		if got["ANTHROPIC_DEFAULT_HAIKU_MODEL"] != "qwen3:8b" {
-			t.Errorf("HAIKU = %q, want qwen3:8b (no fast alias)", got["ANTHROPIC_DEFAULT_HAIKU_MODEL"])
-		}
-		if got["CLAUDE_CODE_SUBAGENT_MODEL"] != "qwen3:8b" {
-			t.Errorf("SUBAGENT = %q, want qwen3:8b", got["CLAUDE_CODE_SUBAGENT_MODEL"])
-		}
-	})
-
-	t.Run("uses fast alias for haiku", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		saveIntegration("claude", []string{"llama3.2:70b"})
-		saveAliases("claude", map[string]string{
-			"primary": "llama3.2:70b",
-			"fast":    "llama3.2:8b",
-		})
-
-		got := envMap(c.modelEnvVars("llama3.2:70b"))
-		if got["ANTHROPIC_DEFAULT_OPUS_MODEL"] != "llama3.2:70b" {
-			t.Errorf("OPUS = %q, want llama3.2:70b", got["ANTHROPIC_DEFAULT_OPUS_MODEL"])
-		}
-		if got["ANTHROPIC_DEFAULT_SONNET_MODEL"] != "llama3.2:70b" {
-			t.Errorf("SONNET = %q, want llama3.2:70b", got["ANTHROPIC_DEFAULT_SONNET_MODEL"])
-		}
-		if got["ANTHROPIC_DEFAULT_HAIKU_MODEL"] != "llama3.2:8b" {
-			t.Errorf("HAIKU = %q, want llama3.2:8b", got["ANTHROPIC_DEFAULT_HAIKU_MODEL"])
-		}
-		if got["CLAUDE_CODE_SUBAGENT_MODEL"] != "llama3.2:70b" {
-			t.Errorf("SUBAGENT = %q, want llama3.2:70b", got["CLAUDE_CODE_SUBAGENT_MODEL"])
-		}
-	})
-
-	t.Run("alias primary overrides model param", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		saveIntegration("claude", []string{"saved-model"})
-		saveAliases("claude", map[string]string{"primary": "saved-model"})
-
-		got := envMap(c.modelEnvVars("different-model"))
-		if got["ANTHROPIC_DEFAULT_OPUS_MODEL"] != "saved-model" {
-			t.Errorf("OPUS = %q, want saved-model", got["ANTHROPIC_DEFAULT_OPUS_MODEL"])
-		}
-	})
-}
--- a/cmd/config/codex.go
+++ b/cmd/config/codex.go
@@ -1,62 +0,0 @@
-package config
-
-import (
-	"fmt"
-	"os"
-	"os/exec"
-	"strings"
-
-	"golang.org/x/mod/semver"
-)
-
-// Codex implements Runner for Codex integration
-type Codex struct{}
-
-func (c *Codex) String() string { return "Codex" }
-
-func (c *Codex) args(model string, extra []string) []string {
-	args := []string{"--oss"}
-	if model != "" {
-		args = append(args, "-m", model)
-	}
-	args = append(args, extra...)
-	return args
-}
-
-func (c *Codex) Run(model string, args []string) error {
-	if err := checkCodexVersion(); err != nil {
-		return err
-	}
-
-	cmd := exec.Command("codex", c.args(model, args)...)
-	cmd.Stdin = os.Stdin
-	cmd.Stdout = os.Stdout
-	cmd.Stderr = os.Stderr
-	return cmd.Run()
-}
-
-func checkCodexVersion() error {
-	if _, err := exec.LookPath("codex"); err != nil {
-		return fmt.Errorf("codex is not installed, install with: npm install -g @openai/codex")
-	}
-
-	out, err := exec.Command("codex", "--version").Output()
-	if err != nil {
-		return fmt.Errorf("failed to get codex version: %w", err)
-	}
-
-	// Parse output like "codex-cli 0.87.0"
-	fields := strings.Fields(strings.TrimSpace(string(out)))
-	if len(fields) < 2 {
-		return fmt.Errorf("unexpected codex version output: %s", string(out))
-	}
-
-	version := "v" + fields[len(fields)-1]
-	minVersion := "v0.81.0"
-
-	if semver.Compare(version, minVersion) < 0 {
-		return fmt.Errorf("codex version %s is too old, minimum required is %s, update with: npm update -g @openai/codex", fields[len(fields)-1], "0.81.0")
-	}
-
-	return nil
-}
--- a/cmd/config/codex_test.go
+++ b/cmd/config/codex_test.go
@@ -1,31 +0,0 @@
-package config
-
-import (
-	"slices"
-	"testing"
-)
-
-func TestCodexArgs(t *testing.T) {
-	c := &Codex{}
-
-	tests := []struct {
-		name  string
-		model string
-		args  []string
-		want  []string
-	}{
-		{"with model", "llama3.2", nil, []string{"--oss", "-m", "llama3.2"}},
-		{"empty model", "", nil, []string{"--oss"}},
-		{"with model and profile", "qwen3-coder", []string{"-p", "myprofile"}, []string{"--oss", "-m", "qwen3-coder", "-p", "myprofile"}},
-		{"with sandbox flag", "llama3.2", []string{"--sandbox", "workspace-write"}, []string{"--oss", "-m", "llama3.2", "--sandbox", "workspace-write"}},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			got := c.args(tt.model, tt.args)
-			if !slices.Equal(got, tt.want) {
-				t.Errorf("args(%q, %v) = %v, want %v", tt.model, tt.args, got, tt.want)
-			}
-		})
-	}
-}
--- a/cmd/config/config.go
+++ b/cmd/config/config.go
@@ -1,271 +0,0 @@
-// Package config provides integration configuration for external coding tools
-// (Claude Code, Codex, Droid, OpenCode) to use Ollama models.
-package config
-
-import (
-	"context"
-	"encoding/json"
-	"errors"
-	"fmt"
-	"os"
-	"path/filepath"
-	"strings"
-
-	"github.com/ollama/ollama/api"
-)
-
-type integration struct {
-	Models  []string          `json:"models"`
-	Aliases map[string]string `json:"aliases,omitempty"`
-}
-
-type config struct {
-	Integrations  map[string]*integration `json:"integrations"`
-	LastModel     string                  `json:"last_model,omitempty"`
-	LastSelection string                  `json:"last_selection,omitempty"` // "run" or integration name
-}
-
-func configPath() (string, error) {
-	home, err := os.UserHomeDir()
-	if err != nil {
-		return "", err
-	}
-	return filepath.Join(home, ".ollama", "config.json"), nil
-}
-
-func legacyConfigPath() (string, error) {
-	home, err := os.UserHomeDir()
-	if err != nil {
-		return "", err
-	}
-	return filepath.Join(home, ".ollama", "config", "config.json"), nil
-}
-
-// migrateConfig moves the config from the legacy path to ~/.ollama/config.json
-func migrateConfig() (bool, error) {
-	oldPath, err := legacyConfigPath()
-	if err != nil {
-		return false, err
-	}
-
-	oldData, err := os.ReadFile(oldPath)
-	if err != nil {
-		if os.IsNotExist(err) {
-			return false, nil
-		}
-		return false, err
-	}
-
-	var js json.RawMessage
-	if err := json.Unmarshal(oldData, &js); err != nil {
-		return false, nil
-	}
-
-	newPath, err := configPath()
-	if err != nil {
-		return false, err
-	}
-
-	if err := os.MkdirAll(filepath.Dir(newPath), 0o755); err != nil {
-		return false, err
-	}
-	if err := os.WriteFile(newPath, oldData, 0o644); err != nil {
-		return false, fmt.Errorf("write new config: %w", err)
-	}
-
-	_ = os.Remove(oldPath)
-	_ = os.Remove(filepath.Dir(oldPath)) // clean up empty directory
-
-	return true, nil
-}
-
-func load() (*config, error) {
-	path, err := configPath()
-	if err != nil {
-		return nil, err
-	}
-
-	data, err := os.ReadFile(path)
-	if err != nil && os.IsNotExist(err) {
-		if migrated, merr := migrateConfig(); merr == nil && migrated {
-			data, err = os.ReadFile(path)
-		}
-	}
-	if err != nil {
-		if os.IsNotExist(err) {
-			return &config{Integrations: make(map[string]*integration)}, nil
-		}
-		return nil, err
-	}
-
-	var cfg config
-	if err := json.Unmarshal(data, &cfg); err != nil {
-		return nil, fmt.Errorf("failed to parse config: %w, at: %s", err, path)
-	}
-	if cfg.Integrations == nil {
-		cfg.Integrations = make(map[string]*integration)
-	}
-	return &cfg, nil
-}
-
-func save(cfg *config) error {
-	path, err := configPath()
-	if err != nil {
-		return err
-	}
-
-	if err := os.MkdirAll(filepath.Dir(path), 0o755); err != nil {
-		return err
-	}
-
-	data, err := json.MarshalIndent(cfg, "", "  ")
-	if err != nil {
-		return err
-	}
-
-	return writeWithBackup(path, data)
-}
-
-func saveIntegration(appName string, models []string) error {
-	if appName == "" {
-		return errors.New("app name cannot be empty")
-	}
-
-	cfg, err := load()
-	if err != nil {
-		return err
-	}
-
-	key := strings.ToLower(appName)
-	existing := cfg.Integrations[key]
-	var aliases map[string]string
-	if existing != nil && existing.Aliases != nil {
-		aliases = existing.Aliases
-	}
-
-	cfg.Integrations[key] = &integration{
-		Models:  models,
-		Aliases: aliases,
-	}
-
-	return save(cfg)
-}
-
-// IntegrationModel returns the first configured model for an integration, or empty string if not configured.
-func IntegrationModel(appName string) string {
-	ic, err := loadIntegration(appName)
-	if err != nil || len(ic.Models) == 0 {
-		return ""
-	}
-	return ic.Models[0]
-}
-
-// LastModel returns the last model that was run, or empty string if none.
-func LastModel() string {
-	cfg, err := load()
-	if err != nil {
-		return ""
-	}
-	return cfg.LastModel
-}
-
-// SetLastModel saves the last model that was run.
-func SetLastModel(model string) error {
-	cfg, err := load()
-	if err != nil {
-		return err
-	}
-	cfg.LastModel = model
-	return save(cfg)
-}
-
-// LastSelection returns the last menu selection ("run" or integration name), or empty string if none.
-func LastSelection() string {
-	cfg, err := load()
-	if err != nil {
-		return ""
-	}
-	return cfg.LastSelection
-}
-
-// SetLastSelection saves the last menu selection ("run" or integration name).
-func SetLastSelection(selection string) error {
-	cfg, err := load()
-	if err != nil {
-		return err
-	}
-	cfg.LastSelection = selection
-	return save(cfg)
-}
-
-// ModelExists checks if a model exists on the Ollama server.
-func ModelExists(ctx context.Context, name string) bool {
-	if name == "" {
-		return false
-	}
-	client, err := api.ClientFromEnvironment()
-	if err != nil {
-		return false
-	}
-	models, err := client.List(ctx)
-	if err != nil {
-		return false
-	}
-	for _, m := range models.Models {
-		if m.Name == name || strings.HasPrefix(m.Name, name+":") {
-			return true
-		}
-	}
-	return false
-}
-
-func loadIntegration(appName string) (*integration, error) {
-	cfg, err := load()
-	if err != nil {
-		return nil, err
-	}
-
-	ic, ok := cfg.Integrations[strings.ToLower(appName)]
-	if !ok {
-		return nil, os.ErrNotExist
-	}
-
-	return ic, nil
-}
-
-func saveAliases(appName string, aliases map[string]string) error {
-	if appName == "" {
-		return errors.New("app name cannot be empty")
-	}
-
-	cfg, err := load()
-	if err != nil {
-		return err
-	}
-
-	key := strings.ToLower(appName)
-	existing := cfg.Integrations[key]
-	if existing == nil {
-		existing = &integration{}
-	}
-
-	// Replace aliases entirely (not merge) so deletions are persisted
-	existing.Aliases = aliases
-
-	cfg.Integrations[key] = existing
-	return save(cfg)
-}
-
-func listIntegrations() ([]integration, error) {
-	cfg, err := load()
-	if err != nil {
-		return nil, err
-	}
-
-	result := make([]integration, 0, len(cfg.Integrations))
-	for _, ic := range cfg.Integrations {
-		result = append(result, *ic)
-	}
-
-	return result, nil
-}
--- a/cmd/config/config_cloud_test.go
+++ b/cmd/config/config_cloud_test.go
@@ -1,677 +0,0 @@
-package config
-
-import (
-	"context"
-	"errors"
-	"os"
-	"path/filepath"
-	"testing"
-)
-
-func TestSetAliases_CloudModel(t *testing.T) {
-	// Test the SetAliases logic by checking the alias map behavior
-	aliases := map[string]string{
-		"primary": "kimi-k2.5:cloud",
-		"fast":    "kimi-k2.5:cloud",
-	}
-
-	// Verify fast is set (cloud model behavior)
-	if aliases["fast"] == "" {
-		t.Error("cloud model should have fast alias set")
-	}
-	if aliases["fast"] != aliases["primary"] {
-		t.Errorf("fast should equal primary for auto-set, got fast=%q primary=%q", aliases["fast"], aliases["primary"])
-	}
-}
-
-func TestSetAliases_LocalModel(t *testing.T) {
-	aliases := map[string]string{
-		"primary": "llama3.2:latest",
-	}
-	// Simulate local model behavior: fast should be empty
-	delete(aliases, "fast")
-
-	if aliases["fast"] != "" {
-		t.Error("local model should have empty fast alias")
-	}
-}
-
-func TestSaveAliases_ReplacesNotMerges(t *testing.T) {
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	// First save with both primary and fast
-	initial := map[string]string{
-		"primary": "cloud-model",
-		"fast":    "cloud-model",
-	}
-	if err := saveAliases("claude", initial); err != nil {
-		t.Fatalf("failed to save initial aliases: %v", err)
-	}
-
-	// Verify both are saved
-	loaded, err := loadIntegration("claude")
-	if err != nil {
-		t.Fatalf("failed to load: %v", err)
-	}
-	if loaded.Aliases["fast"] != "cloud-model" {
-		t.Errorf("expected fast=cloud-model, got %q", loaded.Aliases["fast"])
-	}
-
-	// Now save without fast (simulating switch to local model)
-	updated := map[string]string{
-		"primary": "local-model",
-		// fast intentionally missing
-	}
-	if err := saveAliases("claude", updated); err != nil {
-		t.Fatalf("failed to save updated aliases: %v", err)
-	}
-
-	// Verify fast is GONE (not merged/preserved)
-	loaded, err = loadIntegration("claude")
-	if err != nil {
-		t.Fatalf("failed to load after update: %v", err)
-	}
-	if loaded.Aliases["fast"] != "" {
-		t.Errorf("fast should be removed after saving without it, got %q", loaded.Aliases["fast"])
-	}
-	if loaded.Aliases["primary"] != "local-model" {
-		t.Errorf("primary should be updated to local-model, got %q", loaded.Aliases["primary"])
-	}
-}
-
-func TestSaveAliases_PreservesModels(t *testing.T) {
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	// First save integration with models
-	if err := saveIntegration("claude", []string{"model1", "model2"}); err != nil {
-		t.Fatalf("failed to save integration: %v", err)
-	}
-
-	// Then update aliases
-	aliases := map[string]string{"primary": "new-model"}
-	if err := saveAliases("claude", aliases); err != nil {
-		t.Fatalf("failed to save aliases: %v", err)
-	}
-
-	// Verify models are preserved
-	loaded, err := loadIntegration("claude")
-	if err != nil {
-		t.Fatalf("failed to load: %v", err)
-	}
-	if len(loaded.Models) != 2 || loaded.Models[0] != "model1" {
-		t.Errorf("models should be preserved, got %v", loaded.Models)
-	}
-}
-
-// TestSaveAliases_EmptyMap clears all aliases
-func TestSaveAliases_EmptyMap(t *testing.T) {
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	// Save with aliases
-	if err := saveAliases("claude", map[string]string{"primary": "model", "fast": "model"}); err != nil {
-		t.Fatalf("failed to save: %v", err)
-	}
-
-	// Save empty map
-	if err := saveAliases("claude", map[string]string{}); err != nil {
-		t.Fatalf("failed to save empty: %v", err)
-	}
-
-	loaded, err := loadIntegration("claude")
-	if err != nil {
-		t.Fatalf("failed to load: %v", err)
-	}
-	if len(loaded.Aliases) != 0 {
-		t.Errorf("aliases should be empty, got %v", loaded.Aliases)
-	}
-}
-
-// TestSaveAliases_NilMap handles nil gracefully
-func TestSaveAliases_NilMap(t *testing.T) {
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	// Save with aliases first
-	if err := saveAliases("claude", map[string]string{"primary": "model"}); err != nil {
-		t.Fatalf("failed to save: %v", err)
-	}
-
-	// Save nil map - should clear aliases
-	if err := saveAliases("claude", nil); err != nil {
-		t.Fatalf("failed to save nil: %v", err)
-	}
-
-	loaded, err := loadIntegration("claude")
-	if err != nil {
-		t.Fatalf("failed to load: %v", err)
-	}
-	if len(loaded.Aliases) > 0 {
-		t.Errorf("aliases should be nil or empty, got %v", loaded.Aliases)
-	}
-}
-
-// TestSaveAliases_EmptyAppName returns error
-func TestSaveAliases_EmptyAppName(t *testing.T) {
-	err := saveAliases("", map[string]string{"primary": "model"})
-	if err == nil {
-		t.Error("expected error for empty app name")
-	}
-}
-
-func TestSaveAliases_CaseInsensitive(t *testing.T) {
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	if err := saveAliases("Claude", map[string]string{"primary": "model1"}); err != nil {
-		t.Fatalf("failed to save: %v", err)
-	}
-
-	// Load with different case
-	loaded, err := loadIntegration("claude")
-	if err != nil {
-		t.Fatalf("failed to load: %v", err)
-	}
-	if loaded.Aliases["primary"] != "model1" {
-		t.Errorf("expected primary=model1, got %q", loaded.Aliases["primary"])
-	}
-
-	// Update with different case
-	if err := saveAliases("CLAUDE", map[string]string{"primary": "model2"}); err != nil {
-		t.Fatalf("failed to update: %v", err)
-	}
-
-	loaded, err = loadIntegration("claude")
-	if err != nil {
-		t.Fatalf("failed to load after update: %v", err)
-	}
-	if loaded.Aliases["primary"] != "model2" {
-		t.Errorf("expected primary=model2, got %q", loaded.Aliases["primary"])
-	}
-}
-
-// TestSaveAliases_CreatesIntegration creates integration if it doesn't exist
-func TestSaveAliases_CreatesIntegration(t *testing.T) {
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	// Save aliases for non-existent integration
-	if err := saveAliases("newintegration", map[string]string{"primary": "model"}); err != nil {
-		t.Fatalf("failed to save: %v", err)
-	}
-
-	loaded, err := loadIntegration("newintegration")
-	if err != nil {
-		t.Fatalf("failed to load: %v", err)
-	}
-	if loaded.Aliases["primary"] != "model" {
-		t.Errorf("expected primary=model, got %q", loaded.Aliases["primary"])
-	}
-}
-
-func TestConfigureAliases_AliasMap(t *testing.T) {
-	t.Run("cloud model auto-sets fast to primary", func(t *testing.T) {
-		aliases := make(map[string]string)
-		aliases["primary"] = "cloud-model"
-
-		// Simulate cloud model behavior
-		isCloud := true
-		if isCloud {
-			if aliases["fast"] == "" {
-				aliases["fast"] = aliases["primary"]
-			}
-		}
-
-		if aliases["fast"] != "cloud-model" {
-			t.Errorf("expected fast=cloud-model, got %q", aliases["fast"])
-		}
-	})
-
-	t.Run("cloud model preserves custom fast", func(t *testing.T) {
-		aliases := map[string]string{
-			"primary": "cloud-model",
-			"fast":    "custom-fast-model",
-		}
-
-		// Simulate cloud model behavior - should preserve existing fast
-		isCloud := true
-		if isCloud {
-			if aliases["fast"] == "" {
-				aliases["fast"] = aliases["primary"]
-			}
-		}
-
-		if aliases["fast"] != "custom-fast-model" {
-			t.Errorf("expected fast=custom-fast-model (preserved), got %q", aliases["fast"])
-		}
-	})
-
-	t.Run("local model clears fast", func(t *testing.T) {
-		aliases := map[string]string{
-			"primary": "local-model",
-			"fast":    "should-be-cleared",
-		}
-
-		// Simulate local model behavior
-		isCloud := false
-		if !isCloud {
-			delete(aliases, "fast")
-		}
-
-		if aliases["fast"] != "" {
-			t.Errorf("expected fast to be cleared, got %q", aliases["fast"])
-		}
-	})
-
-	t.Run("switching cloud to local clears fast", func(t *testing.T) {
-		// Start with cloud config
-		aliases := map[string]string{
-			"primary": "cloud-model",
-			"fast":    "cloud-model",
-		}
-
-		// Switch to local
-		aliases["primary"] = "local-model"
-		isCloud := false
-		if !isCloud {
-			delete(aliases, "fast")
-		}
-
-		if aliases["fast"] != "" {
-			t.Errorf("fast should be cleared when switching to local, got %q", aliases["fast"])
-		}
-		if aliases["primary"] != "local-model" {
-			t.Errorf("primary should be updated, got %q", aliases["primary"])
-		}
-	})
-
-	t.Run("switching local to cloud sets fast", func(t *testing.T) {
-		// Start with local config (no fast)
-		aliases := map[string]string{
-			"primary": "local-model",
-		}
-
-		// Switch to cloud
-		aliases["primary"] = "cloud-model"
-		isCloud := true
-		if isCloud {
-			if aliases["fast"] == "" {
-				aliases["fast"] = aliases["primary"]
-			}
-		}
-
-		if aliases["fast"] != "cloud-model" {
-			t.Errorf("fast should be set when switching to cloud, got %q", aliases["fast"])
-		}
-	})
-}
-
-func TestSetAliases_PrefixMapping(t *testing.T) {
-	// This tests the expected mapping without needing a real client
-	aliases := map[string]string{
-		"primary": "my-cloud-model",
-		"fast":    "my-fast-model",
-	}
-
-	expectedMappings := map[string]string{
-		"claude-sonnet-": aliases["primary"],
-		"claude-haiku-":  aliases["fast"],
-	}
-
-	if expectedMappings["claude-sonnet-"] != "my-cloud-model" {
-		t.Errorf("claude-sonnet- should map to primary")
-	}
-	if expectedMappings["claude-haiku-"] != "my-fast-model" {
-		t.Errorf("claude-haiku- should map to fast")
-	}
-}
-
-func TestSetAliases_LocalDeletesPrefixes(t *testing.T) {
-	aliases := map[string]string{
-		"primary": "local-model",
-		// fast is empty/missing - indicates local model
-	}
-
-	prefixesToDelete := []string{"claude-sonnet-", "claude-haiku-"}
-
-	// Verify the logic: when fast is empty, we should delete
-	if aliases["fast"] != "" {
-		t.Error("fast should be empty for local model")
-	}
-
-	// Verify we have the right prefixes to delete
-	if len(prefixesToDelete) != 2 {
-		t.Errorf("expected 2 prefixes to delete, got %d", len(prefixesToDelete))
-	}
-}
-
-// TestAtomicUpdate_ServerFailsConfigNotSaved simulates atomic update behavior
-func TestAtomicUpdate_ServerFailsConfigNotSaved(t *testing.T) {
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	// Simulate: server fails, config should NOT be saved
-	serverErr := errors.New("server unavailable")
-
-	if serverErr == nil {
-		t.Error("config should NOT be saved when server fails")
-	}
-}
-
-// TestAtomicUpdate_ServerSucceedsConfigSaved simulates successful atomic update
-func TestAtomicUpdate_ServerSucceedsConfigSaved(t *testing.T) {
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	// Simulate: server succeeds, config should be saved
-	var serverErr error
-	if serverErr != nil {
-		t.Fatal("server should succeed")
-	}
-
-	if err := saveAliases("claude", map[string]string{"primary": "model"}); err != nil {
-		t.Fatalf("saveAliases failed: %v", err)
-	}
-
-	// Verify it was actually saved
-	loaded, err := loadIntegration("claude")
-	if err != nil {
-		t.Fatalf("failed to load: %v", err)
-	}
-	if loaded.Aliases["primary"] != "model" {
-		t.Errorf("expected primary=model, got %q", loaded.Aliases["primary"])
-	}
-}
-
-func TestConfigFile_PreservesUnknownFields(t *testing.T) {
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	// Write config with extra fields
-	configPath := filepath.Join(tmpDir, ".ollama", "config.json")
-	os.MkdirAll(filepath.Dir(configPath), 0o755)
-
-	// Note: Our config struct only has Integrations, so top-level unknown fields
-	// won't be preserved by our current implementation. This test documents that.
-	initialConfig := `{
-  "integrations": {
-    "claude": {
-      "models": ["model1"],
-      "aliases": {"primary": "model1"},
-      "unknownField": "should be lost"
-    }
-  },
-  "topLevelUnknown": "will be lost"
-}`
-	os.WriteFile(configPath, []byte(initialConfig), 0o644)
-
-	// Update aliases
-	if err := saveAliases("claude", map[string]string{"primary": "model2"}); err != nil {
-		t.Fatalf("failed to save: %v", err)
-	}
-
-	// Read raw file to check
-	data, _ := os.ReadFile(configPath)
-	content := string(data)
-
-	// models should be preserved
-	if !contains(content, "model1") {
-		t.Error("models should be preserved")
-	}
-
-	// primary should be updated
-	if !contains(content, "model2") {
-		t.Error("primary should be updated to model2")
-	}
-}
-
-func contains(s, substr string) bool {
-	return len(s) >= len(substr) && (s == substr || len(s) > 0 && containsHelper(s, substr))
-}
-
-func containsHelper(s, substr string) bool {
-	for i := 0; i <= len(s)-len(substr); i++ {
-		if s[i:i+len(substr)] == substr {
-			return true
-		}
-	}
-	return false
-}
-
-func TestClaudeImplementsAliasConfigurer(t *testing.T) {
-	c := &Claude{}
-	var _ AliasConfigurer = c // Compile-time check
-}
-
-func TestModelNameEdgeCases(t *testing.T) {
-	testCases := []struct {
-		name  string
-		model string
-	}{
-		{"simple", "llama3.2"},
-		{"with tag", "llama3.2:latest"},
-		{"with cloud tag", "kimi-k2.5:cloud"},
-		{"with namespace", "library/llama3.2"},
-		{"with dots", "glm-4.7-flash"},
-		{"with numbers", "qwen3:8b"},
-	}
-
-	for _, tc := range testCases {
-		t.Run(tc.name, func(t *testing.T) {
-			tmpDir := t.TempDir()
-			setTestHome(t, tmpDir)
-
-			aliases := map[string]string{"primary": tc.model}
-			if err := saveAliases("claude", aliases); err != nil {
-				t.Fatalf("failed to save model %q: %v", tc.model, err)
-			}
-
-			loaded, err := loadIntegration("claude")
-			if err != nil {
-				t.Fatalf("failed to load: %v", err)
-			}
-			if loaded.Aliases["primary"] != tc.model {
-				t.Errorf("expected primary=%q, got %q", tc.model, loaded.Aliases["primary"])
-			}
-		})
-	}
-}
-
-func TestSwitchingScenarios(t *testing.T) {
-	t.Run("cloud to local removes fast", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		// Initial cloud config
-		if err := saveAliases("claude", map[string]string{
-			"primary": "cloud-model",
-			"fast":    "cloud-model",
-		}); err != nil {
-			t.Fatal(err)
-		}
-
-		// Switch to local (no fast)
-		if err := saveAliases("claude", map[string]string{
-			"primary": "local-model",
-		}); err != nil {
-			t.Fatal(err)
-		}
-
-		loaded, _ := loadIntegration("claude")
-		if loaded.Aliases["fast"] != "" {
-			t.Errorf("fast should be removed, got %q", loaded.Aliases["fast"])
-		}
-		if loaded.Aliases["primary"] != "local-model" {
-			t.Errorf("primary should be local-model, got %q", loaded.Aliases["primary"])
-		}
-	})
-
-	t.Run("local to cloud adds fast", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		// Initial local config
-		if err := saveAliases("claude", map[string]string{
-			"primary": "local-model",
-		}); err != nil {
-			t.Fatal(err)
-		}
-
-		// Switch to cloud (with fast)
-		if err := saveAliases("claude", map[string]string{
-			"primary": "cloud-model",
-			"fast":    "cloud-model",
-		}); err != nil {
-			t.Fatal(err)
-		}
-
-		loaded, _ := loadIntegration("claude")
-		if loaded.Aliases["fast"] != "cloud-model" {
-			t.Errorf("fast should be cloud-model, got %q", loaded.Aliases["fast"])
-		}
-	})
-
-	t.Run("cloud to different cloud updates both", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		// Initial cloud config
-		if err := saveAliases("claude", map[string]string{
-			"primary": "cloud-model-1",
-			"fast":    "cloud-model-1",
-		}); err != nil {
-			t.Fatal(err)
-		}
-
-		// Switch to different cloud
-		if err := saveAliases("claude", map[string]string{
-			"primary": "cloud-model-2",
-			"fast":    "cloud-model-2",
-		}); err != nil {
-			t.Fatal(err)
-		}
-
-		loaded, _ := loadIntegration("claude")
-		if loaded.Aliases["primary"] != "cloud-model-2" {
-			t.Errorf("primary should be cloud-model-2, got %q", loaded.Aliases["primary"])
-		}
-		if loaded.Aliases["fast"] != "cloud-model-2" {
-			t.Errorf("fast should be cloud-model-2, got %q", loaded.Aliases["fast"])
-		}
-	})
-}
-
-func TestToolCapabilityFiltering(t *testing.T) {
-	t.Run("all models checked for tool capability", func(t *testing.T) {
-		// Both cloud and local models are checked for tool capability via Show API
-		// Only models with "tools" in capabilities are included
-		m := modelInfo{Name: "tool-model", Remote: false, ToolCapable: true}
-		if !m.ToolCapable {
-			t.Error("tool capable model should be marked as such")
-		}
-	})
-
-	t.Run("modelInfo includes ToolCapable field", func(t *testing.T) {
-		m := modelInfo{Name: "test", Remote: true, ToolCapable: true}
-		if !m.ToolCapable {
-			t.Error("ToolCapable field should be accessible")
-		}
-	})
-}
-
-func TestIsCloudModel_RequiresClient(t *testing.T) {
-	t.Run("nil client always returns false", func(t *testing.T) {
-		// isCloudModel now only uses Show API, no suffix detection
-		if isCloudModel(context.Background(), nil, "model:cloud") {
-			t.Error("nil client should return false regardless of suffix")
-		}
-		if isCloudModel(context.Background(), nil, "local-model") {
-			t.Error("nil client should return false")
-		}
-	})
-}
-
-func TestModelsAndAliasesMustStayInSync(t *testing.T) {
-	t.Run("saveAliases followed by saveIntegration keeps them in sync", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		// Save aliases with one model
-		if err := saveAliases("claude", map[string]string{"primary": "model-a"}); err != nil {
-			t.Fatal(err)
-		}
-
-		// Save integration with same model (this is the pattern we use)
-		if err := saveIntegration("claude", []string{"model-a"}); err != nil {
-			t.Fatal(err)
-		}
-
-		loaded, _ := loadIntegration("claude")
-		if loaded.Aliases["primary"] != loaded.Models[0] {
-			t.Errorf("aliases.primary (%q) != models[0] (%q)", loaded.Aliases["primary"], loaded.Models[0])
-		}
-	})
-
-	t.Run("out of sync config is detectable", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		// Simulate out-of-sync state (like manual edit or bug)
-		if err := saveIntegration("claude", []string{"old-model"}); err != nil {
-			t.Fatal(err)
-		}
-		if err := saveAliases("claude", map[string]string{"primary": "new-model"}); err != nil {
-			t.Fatal(err)
-		}
-
-		loaded, _ := loadIntegration("claude")
-
-		// They should be different (this is the bug state)
-		if loaded.Models[0] == loaded.Aliases["primary"] {
-			t.Error("expected out-of-sync state for this test")
-		}
-
-		// The fix: when updating aliases, also update models
-		if err := saveIntegration("claude", []string{loaded.Aliases["primary"]}); err != nil {
-			t.Fatal(err)
-		}
-
-		loaded, _ = loadIntegration("claude")
-		if loaded.Models[0] != loaded.Aliases["primary"] {
-			t.Errorf("after fix: models[0] (%q) should equal aliases.primary (%q)",
-				loaded.Models[0], loaded.Aliases["primary"])
-		}
-	})
-
-	t.Run("updating primary alias updates models too", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		// Initial state
-		if err := saveIntegration("claude", []string{"initial-model"}); err != nil {
-			t.Fatal(err)
-		}
-		if err := saveAliases("claude", map[string]string{"primary": "initial-model"}); err != nil {
-			t.Fatal(err)
-		}
-
-		// Update aliases AND models together
-		newAliases := map[string]string{"primary": "updated-model"}
-		if err := saveAliases("claude", newAliases); err != nil {
-			t.Fatal(err)
-		}
-		if err := saveIntegration("claude", []string{newAliases["primary"]}); err != nil {
-			t.Fatal(err)
-		}
-
-		loaded, _ := loadIntegration("claude")
-		if loaded.Models[0] != "updated-model" {
-			t.Errorf("models[0] should be updated-model, got %q", loaded.Models[0])
-		}
-		if loaded.Aliases["primary"] != "updated-model" {
-			t.Errorf("aliases.primary should be updated-model, got %q", loaded.Aliases["primary"])
-		}
-	})
-}
--- a/cmd/config/config_test.go
+++ b/cmd/config/config_test.go
@@ -1,595 +0,0 @@
-package config
-
-import (
-	"os"
-	"path/filepath"
-	"strings"
-	"testing"
-)
-
-// setTestHome sets both HOME (Unix) and USERPROFILE (Windows) for cross-platform tests
-func setTestHome(t *testing.T, dir string) {
-	t.Setenv("HOME", dir)
-	t.Setenv("USERPROFILE", dir)
-}
-
-// editorPaths is a test helper that safely calls Paths if the runner implements Editor
-func editorPaths(r Runner) []string {
-	if editor, ok := r.(Editor); ok {
-		return editor.Paths()
-	}
-	return nil
-}
-
-func TestIntegrationConfig(t *testing.T) {
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	t.Run("save and load round-trip", func(t *testing.T) {
-		models := []string{"llama3.2", "mistral", "qwen2.5"}
-		if err := saveIntegration("claude", models); err != nil {
-			t.Fatal(err)
-		}
-
-		config, err := loadIntegration("claude")
-		if err != nil {
-			t.Fatal(err)
-		}
-
-		if len(config.Models) != len(models) {
-			t.Errorf("expected %d models, got %d", len(models), len(config.Models))
-		}
-		for i, m := range models {
-			if config.Models[i] != m {
-				t.Errorf("model %d: expected %s, got %s", i, m, config.Models[i])
-			}
-		}
-	})
-
-	t.Run("save and load aliases", func(t *testing.T) {
-		models := []string{"llama3.2"}
-		if err := saveIntegration("claude", models); err != nil {
-			t.Fatal(err)
-		}
-		aliases := map[string]string{
-			"primary": "llama3.2:70b",
-			"fast":    "llama3.2:8b",
-		}
-		if err := saveAliases("claude", aliases); err != nil {
-			t.Fatal(err)
-		}
-
-		config, err := loadIntegration("claude")
-		if err != nil {
-			t.Fatal(err)
-		}
-		if config.Aliases == nil {
-			t.Fatal("expected aliases to be saved")
-		}
-		for k, v := range aliases {
-			if config.Aliases[k] != v {
-				t.Errorf("alias %s: expected %s, got %s", k, v, config.Aliases[k])
-			}
-		}
-	})
-
-	t.Run("saveIntegration preserves aliases", func(t *testing.T) {
-		if err := saveIntegration("claude", []string{"model-a"}); err != nil {
-			t.Fatal(err)
-		}
-		if err := saveAliases("claude", map[string]string{"primary": "model-a", "fast": "model-small"}); err != nil {
-			t.Fatal(err)
-		}
-
-		if err := saveIntegration("claude", []string{"model-b"}); err != nil {
-			t.Fatal(err)
-		}
-		config, err := loadIntegration("claude")
-		if err != nil {
-			t.Fatal(err)
-		}
-		if config.Aliases["primary"] != "model-a" {
-			t.Errorf("expected aliases to be preserved, got %v", config.Aliases)
-		}
-	})
-
-	t.Run("defaultModel returns first model", func(t *testing.T) {
-		saveIntegration("codex", []string{"model-a", "model-b"})
-
-		config, _ := loadIntegration("codex")
-		defaultModel := ""
-		if len(config.Models) > 0 {
-			defaultModel = config.Models[0]
-		}
-		if defaultModel != "model-a" {
-			t.Errorf("expected model-a, got %s", defaultModel)
-		}
-	})
-
-	t.Run("defaultModel returns empty for no models", func(t *testing.T) {
-		config := &integration{Models: []string{}}
-		defaultModel := ""
-		if len(config.Models) > 0 {
-			defaultModel = config.Models[0]
-		}
-		if defaultModel != "" {
-			t.Errorf("expected empty string, got %s", defaultModel)
-		}
-	})
-
-	t.Run("app name is case-insensitive", func(t *testing.T) {
-		saveIntegration("Claude", []string{"model-x"})
-
-		config, err := loadIntegration("claude")
-		if err != nil {
-			t.Fatal(err)
-		}
-		defaultModel := ""
-		if len(config.Models) > 0 {
-			defaultModel = config.Models[0]
-		}
-		if defaultModel != "model-x" {
-			t.Errorf("expected model-x, got %s", defaultModel)
-		}
-	})
-
-	t.Run("multiple integrations in single file", func(t *testing.T) {
-		saveIntegration("app1", []string{"model-1"})
-		saveIntegration("app2", []string{"model-2"})
-
-		config1, _ := loadIntegration("app1")
-		config2, _ := loadIntegration("app2")
-
-		defaultModel1 := ""
-		if len(config1.Models) > 0 {
-			defaultModel1 = config1.Models[0]
-		}
-		defaultModel2 := ""
-		if len(config2.Models) > 0 {
-			defaultModel2 = config2.Models[0]
-		}
-		if defaultModel1 != "model-1" {
-			t.Errorf("expected model-1, got %s", defaultModel1)
-		}
-		if defaultModel2 != "model-2" {
-			t.Errorf("expected model-2, got %s", defaultModel2)
-		}
-	})
-}
-
-func TestListIntegrations(t *testing.T) {
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	t.Run("returns empty when no integrations", func(t *testing.T) {
-		configs, err := listIntegrations()
-		if err != nil {
-			t.Fatal(err)
-		}
-		if len(configs) != 0 {
-			t.Errorf("expected 0 integrations, got %d", len(configs))
-		}
-	})
-
-	t.Run("returns all saved integrations", func(t *testing.T) {
-		saveIntegration("claude", []string{"model-1"})
-		saveIntegration("droid", []string{"model-2"})
-
-		configs, err := listIntegrations()
-		if err != nil {
-			t.Fatal(err)
-		}
-		if len(configs) != 2 {
-			t.Errorf("expected 2 integrations, got %d", len(configs))
-		}
-	})
-}
-
-func TestEditorPaths(t *testing.T) {
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	t.Run("returns empty for claude (no Editor)", func(t *testing.T) {
-		r := integrations["claude"]
-		paths := editorPaths(r)
-		if len(paths) != 0 {
-			t.Errorf("expected no paths for claude, got %v", paths)
-		}
-	})
-
-	t.Run("returns empty for codex (no Editor)", func(t *testing.T) {
-		r := integrations["codex"]
-		paths := editorPaths(r)
-		if len(paths) != 0 {
-			t.Errorf("expected no paths for codex, got %v", paths)
-		}
-	})
-
-	t.Run("returns empty for droid when no config exists", func(t *testing.T) {
-		r := integrations["droid"]
-		paths := editorPaths(r)
-		if len(paths) != 0 {
-			t.Errorf("expected no paths, got %v", paths)
-		}
-	})
-
-	t.Run("returns path for droid when config exists", func(t *testing.T) {
-		settingsDir, _ := os.UserHomeDir()
-		settingsDir = filepath.Join(settingsDir, ".factory")
-		os.MkdirAll(settingsDir, 0o755)
-		os.WriteFile(filepath.Join(settingsDir, "settings.json"), []byte(`{}`), 0o644)
-
-		r := integrations["droid"]
-		paths := editorPaths(r)
-		if len(paths) != 1 {
-			t.Errorf("expected 1 path, got %d", len(paths))
-		}
-	})
-
-	t.Run("returns paths for opencode when configs exist", func(t *testing.T) {
-		home, _ := os.UserHomeDir()
-		configDir := filepath.Join(home, ".config", "opencode")
-		stateDir := filepath.Join(home, ".local", "state", "opencode")
-		os.MkdirAll(configDir, 0o755)
-		os.MkdirAll(stateDir, 0o755)
-		os.WriteFile(filepath.Join(configDir, "opencode.json"), []byte(`{}`), 0o644)
-		os.WriteFile(filepath.Join(stateDir, "model.json"), []byte(`{}`), 0o644)
-
-		r := integrations["opencode"]
-		paths := editorPaths(r)
-		if len(paths) != 2 {
-			t.Errorf("expected 2 paths, got %d: %v", len(paths), paths)
-		}
-	})
-}
-
-func TestLoadIntegration_CorruptedJSON(t *testing.T) {
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	dir := filepath.Join(tmpDir, ".ollama")
-	os.MkdirAll(dir, 0o755)
-	os.WriteFile(filepath.Join(dir, "config.json"), []byte(`{corrupted json`), 0o644)
-
-	_, err := loadIntegration("test")
-	if err == nil {
-		t.Error("expected error for nonexistent integration in corrupted file")
-	}
-}
-
-func TestSaveIntegration_NilModels(t *testing.T) {
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	if err := saveIntegration("test", nil); err != nil {
-		t.Fatalf("saveIntegration with nil models failed: %v", err)
-	}
-
-	config, err := loadIntegration("test")
-	if err != nil {
-		t.Fatalf("loadIntegration failed: %v", err)
-	}
-
-	if config.Models == nil {
-		// nil is acceptable
-	} else if len(config.Models) != 0 {
-		t.Errorf("expected empty or nil models, got %v", config.Models)
-	}
-}
-
-func TestSaveIntegration_EmptyAppName(t *testing.T) {
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	err := saveIntegration("", []string{"model"})
-	if err == nil {
-		t.Error("expected error for empty app name, got nil")
-	}
-	if err != nil && !strings.Contains(err.Error(), "app name cannot be empty") {
-		t.Errorf("expected 'app name cannot be empty' error, got: %v", err)
-	}
-}
-
-func TestLoadIntegration_NonexistentIntegration(t *testing.T) {
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	_, err := loadIntegration("nonexistent")
-	if err == nil {
-		t.Error("expected error for nonexistent integration, got nil")
-	}
-	if !os.IsNotExist(err) {
-		t.Logf("error type is os.ErrNotExist as expected: %v", err)
-	}
-}
-
-func TestConfigPath(t *testing.T) {
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	path, err := configPath()
-	if err != nil {
-		t.Fatal(err)
-	}
-
-	expected := filepath.Join(tmpDir, ".ollama", "config.json")
-	if path != expected {
-		t.Errorf("expected %s, got %s", expected, path)
-	}
-}
-
-func TestLoad(t *testing.T) {
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	t.Run("returns empty config when file does not exist", func(t *testing.T) {
-		cfg, err := load()
-		if err != nil {
-			t.Fatal(err)
-		}
-		if cfg == nil {
-			t.Fatal("expected non-nil config")
-		}
-		if cfg.Integrations == nil {
-			t.Error("expected non-nil Integrations map")
-		}
-		if len(cfg.Integrations) != 0 {
-			t.Errorf("expected empty Integrations, got %d", len(cfg.Integrations))
-		}
-	})
-
-	t.Run("loads existing config", func(t *testing.T) {
-		path, _ := configPath()
-		os.MkdirAll(filepath.Dir(path), 0o755)
-		os.WriteFile(path, []byte(`{"integrations":{"test":{"models":["model-a"]}}}`), 0o644)
-
-		cfg, err := load()
-		if err != nil {
-			t.Fatal(err)
-		}
-		if cfg.Integrations["test"] == nil {
-			t.Fatal("expected test integration")
-		}
-		if len(cfg.Integrations["test"].Models) != 1 {
-			t.Errorf("expected 1 model, got %d", len(cfg.Integrations["test"].Models))
-		}
-	})
-
-	t.Run("returns error for corrupted JSON", func(t *testing.T) {
-		path, _ := configPath()
-		os.MkdirAll(filepath.Dir(path), 0o755)
-		os.WriteFile(path, []byte(`{corrupted`), 0o644)
-
-		_, err := load()
-		if err == nil {
-			t.Error("expected error for corrupted JSON")
-		}
-	})
-}
-
-func TestMigrateConfig(t *testing.T) {
-	t.Run("migrates legacy file to new location", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		legacyDir := filepath.Join(tmpDir, ".ollama", "config")
-		os.MkdirAll(legacyDir, 0o755)
-		data := []byte(`{"integrations":{"claude":{"models":["llama3.2"]}}}`)
-		os.WriteFile(filepath.Join(legacyDir, "config.json"), data, 0o644)
-
-		migrated, err := migrateConfig()
-		if err != nil {
-			t.Fatal(err)
-		}
-		if !migrated {
-			t.Fatal("expected migration to occur")
-		}
-
-		newPath, _ := configPath()
-		got, err := os.ReadFile(newPath)
-		if err != nil {
-			t.Fatalf("new config not found: %v", err)
-		}
-		if string(got) != string(data) {
-			t.Errorf("content mismatch: got %s", got)
-		}
-
-		if _, err := os.Stat(filepath.Join(legacyDir, "config.json")); !os.IsNotExist(err) {
-			t.Error("legacy file should have been removed")
-		}
-
-		if _, err := os.Stat(legacyDir); !os.IsNotExist(err) {
-			t.Error("legacy directory should have been removed")
-		}
-	})
-
-	t.Run("no-op when no legacy file exists", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		migrated, err := migrateConfig()
-		if err != nil {
-			t.Fatal(err)
-		}
-		if migrated {
-			t.Error("expected no migration")
-		}
-	})
-
-	t.Run("skips corrupt legacy file", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		legacyDir := filepath.Join(tmpDir, ".ollama", "config")
-		os.MkdirAll(legacyDir, 0o755)
-		os.WriteFile(filepath.Join(legacyDir, "config.json"), []byte(`{corrupt`), 0o644)
-
-		migrated, err := migrateConfig()
-		if err != nil {
-			t.Fatal(err)
-		}
-		if migrated {
-			t.Error("should not migrate corrupt file")
-		}
-
-		if _, err := os.Stat(filepath.Join(legacyDir, "config.json")); os.IsNotExist(err) {
-			t.Error("corrupt legacy file should not have been deleted")
-		}
-	})
-
-	t.Run("new path takes precedence over legacy", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		legacyDir := filepath.Join(tmpDir, ".ollama", "config")
-		os.MkdirAll(legacyDir, 0o755)
-		os.WriteFile(filepath.Join(legacyDir, "config.json"), []byte(`{"integrations":{"old":{"models":["old-model"]}}}`), 0o644)
-
-		newDir := filepath.Join(tmpDir, ".ollama")
-		os.WriteFile(filepath.Join(newDir, "config.json"), []byte(`{"integrations":{"new":{"models":["new-model"]}}}`), 0o644)
-
-		cfg, err := load()
-		if err != nil {
-			t.Fatal(err)
-		}
-		if _, ok := cfg.Integrations["new"]; !ok {
-			t.Error("expected new-path integration to be loaded")
-		}
-		if _, ok := cfg.Integrations["old"]; ok {
-			t.Error("legacy integration should not have been loaded")
-		}
-	})
-
-	t.Run("idempotent when called twice", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		legacyDir := filepath.Join(tmpDir, ".ollama", "config")
-		os.MkdirAll(legacyDir, 0o755)
-		os.WriteFile(filepath.Join(legacyDir, "config.json"), []byte(`{"integrations":{}}`), 0o644)
-
-		if _, err := migrateConfig(); err != nil {
-			t.Fatal(err)
-		}
-
-		migrated, err := migrateConfig()
-		if err != nil {
-			t.Fatal(err)
-		}
-		if migrated {
-			t.Error("second migration should be a no-op")
-		}
-	})
-
-	t.Run("legacy directory preserved if not empty", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		legacyDir := filepath.Join(tmpDir, ".ollama", "config")
-		os.MkdirAll(legacyDir, 0o755)
-		os.WriteFile(filepath.Join(legacyDir, "config.json"), []byte(`{"integrations":{}}`), 0o644)
-		os.WriteFile(filepath.Join(legacyDir, "other-file.txt"), []byte("keep me"), 0o644)
-
-		if _, err := migrateConfig(); err != nil {
-			t.Fatal(err)
-		}
-
-		if _, err := os.Stat(legacyDir); os.IsNotExist(err) {
-			t.Error("directory with other files should not have been removed")
-		}
-		if _, err := os.Stat(filepath.Join(legacyDir, "other-file.txt")); os.IsNotExist(err) {
-			t.Error("other files in legacy directory should be untouched")
-		}
-	})
-
-	t.Run("save writes to new path after migration", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		legacyDir := filepath.Join(tmpDir, ".ollama", "config")
-		os.MkdirAll(legacyDir, 0o755)
-		os.WriteFile(filepath.Join(legacyDir, "config.json"), []byte(`{"integrations":{"claude":{"models":["llama3.2"]}}}`), 0o644)
-
-		// load triggers migration, then save should write to new path
-		if err := saveIntegration("codex", []string{"qwen2.5"}); err != nil {
-			t.Fatal(err)
-		}
-
-		newPath := filepath.Join(tmpDir, ".ollama", "config.json")
-		if _, err := os.Stat(newPath); os.IsNotExist(err) {
-			t.Error("save should write to new path")
-		}
-
-		// old path should not be recreated
-		if _, err := os.Stat(filepath.Join(legacyDir, "config.json")); !os.IsNotExist(err) {
-			t.Error("save should not recreate legacy path")
-		}
-	})
-
-	t.Run("load triggers migration transparently", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		legacyDir := filepath.Join(tmpDir, ".ollama", "config")
-		os.MkdirAll(legacyDir, 0o755)
-		os.WriteFile(filepath.Join(legacyDir, "config.json"), []byte(`{"integrations":{"claude":{"models":["llama3.2"]}}}`), 0o644)
-
-		cfg, err := load()
-		if err != nil {
-			t.Fatal(err)
-		}
-		if cfg.Integrations["claude"] == nil || cfg.Integrations["claude"].Models[0] != "llama3.2" {
-			t.Error("migration via load() did not preserve data")
-		}
-	})
-}
-
-func TestSave(t *testing.T) {
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	t.Run("creates config file", func(t *testing.T) {
-		cfg := &config{
-			Integrations: map[string]*integration{
-				"test": {Models: []string{"model-a", "model-b"}},
-			},
-		}
-
-		if err := save(cfg); err != nil {
-			t.Fatal(err)
-		}
-
-		path, _ := configPath()
-		if _, err := os.Stat(path); os.IsNotExist(err) {
-			t.Error("config file was not created")
-		}
-	})
-
-	t.Run("round-trip preserves data", func(t *testing.T) {
-		cfg := &config{
-			Integrations: map[string]*integration{
-				"claude": {Models: []string{"llama3.2", "mistral"}},
-				"codex":  {Models: []string{"qwen2.5"}},
-			},
-		}
-
-		if err := save(cfg); err != nil {
-			t.Fatal(err)
-		}
-
-		loaded, err := load()
-		if err != nil {
-			t.Fatal(err)
-		}
-
-		if len(loaded.Integrations) != 2 {
-			t.Errorf("expected 2 integrations, got %d", len(loaded.Integrations))
-		}
-		if loaded.Integrations["claude"] == nil {
-			t.Error("missing claude integration")
-		}
-		if len(loaded.Integrations["claude"].Models) != 2 {
-			t.Errorf("expected 2 models for claude, got %d", len(loaded.Integrations["claude"].Models))
-		}
-	})
-}
--- a/cmd/config/droid.go
+++ b/cmd/config/droid.go
@@ -1,196 +0,0 @@
-package config
-
-import (
-	"context"
-	"encoding/json"
-	"fmt"
-	"os"
-	"os/exec"
-	"path/filepath"
-	"slices"
-
-	"github.com/ollama/ollama/api"
-	"github.com/ollama/ollama/envconfig"
-)
-
-// Droid implements Runner and Editor for Droid integration
-type Droid struct{}
-
-// droidSettings represents the Droid settings.json file (only fields we use)
-type droidSettings struct {
-	CustomModels           []modelEntry    `json:"customModels"`
-	SessionDefaultSettings sessionSettings `json:"sessionDefaultSettings"`
-}
-
-type sessionSettings struct {
-	Model           string `json:"model"`
-	ReasoningEffort string `json:"reasoningEffort"`
-}
-
-type modelEntry struct {
-	Model           string `json:"model"`
-	DisplayName     string `json:"displayName"`
-	BaseURL         string `json:"baseUrl"`
-	APIKey          string `json:"apiKey"`
-	Provider        string `json:"provider"`
-	MaxOutputTokens int    `json:"maxOutputTokens"`
-	SupportsImages  bool   `json:"supportsImages"`
-	ID              string `json:"id"`
-	Index           int    `json:"index"`
-}
-
-func (d *Droid) String() string { return "Droid" }
-
-func (d *Droid) Run(model string, args []string) error {
-	if _, err := exec.LookPath("droid"); err != nil {
-		return fmt.Errorf("droid is not installed, install from https://docs.factory.ai/cli/getting-started/quickstart")
-	}
-
-	// Call Edit() to ensure config is up-to-date before launch
-	models := []string{model}
-	if config, err := loadIntegration("droid"); err == nil && len(config.Models) > 0 {
-		models = config.Models
-	}
-	if err := d.Edit(models); err != nil {
-		return fmt.Errorf("setup failed: %w", err)
-	}
-
-	cmd := exec.Command("droid", args...)
-	cmd.Stdin = os.Stdin
-	cmd.Stdout = os.Stdout
-	cmd.Stderr = os.Stderr
-	return cmd.Run()
-}
-
-func (d *Droid) Paths() []string {
-	home, err := os.UserHomeDir()
-	if err != nil {
-		return nil
-	}
-	p := filepath.Join(home, ".factory", "settings.json")
-	if _, err := os.Stat(p); err == nil {
-		return []string{p}
-	}
-	return nil
-}
-
-func (d *Droid) Edit(models []string) error {
-	if len(models) == 0 {
-		return nil
-	}
-
-	home, err := os.UserHomeDir()
-	if err != nil {
-		return err
-	}
-
-	settingsPath := filepath.Join(home, ".factory", "settings.json")
-	if err := os.MkdirAll(filepath.Dir(settingsPath), 0o755); err != nil {
-		return err
-	}
-
-	// Read file once, unmarshal twice:
-	// map preserves unknown fields for writing back (including extra fields in model entries)
-	settingsMap := make(map[string]any)
-	var settings droidSettings
-	if data, err := os.ReadFile(settingsPath); err == nil {
-		if err := json.Unmarshal(data, &settingsMap); err != nil {
-			return fmt.Errorf("failed to parse settings file: %w, at: %s", err, settingsPath)
-		}
-		json.Unmarshal(data, &settings) // ignore error, zero values are fine
-	}
-
-	// Keep only non-Ollama models from the raw map (preserves extra fields)
-	// Rebuild Ollama models
-	var nonOllamaModels []any
-	if rawModels, ok := settingsMap["customModels"].([]any); ok {
-		for _, raw := range rawModels {
-			if m, ok := raw.(map[string]any); ok {
-				if m["apiKey"] != "ollama" {
-					nonOllamaModels = append(nonOllamaModels, raw)
-				}
-			}
-		}
-	}
-
-	// Build new Ollama model entries with sequential indices (0, 1, 2, ...)
-	client, _ := api.ClientFromEnvironment()
-
-	var newModels []any
-	var defaultModelID string
-	for i, model := range models {
-		maxOutput := 64000
-		if isCloudModel(context.Background(), client, model) {
-			if l, ok := lookupCloudModelLimit(model); ok {
-				maxOutput = l.Output
-			}
-		}
-		modelID := fmt.Sprintf("custom:%s-%d", model, i)
-		newModels = append(newModels, modelEntry{
-			Model:           model,
-			DisplayName:     model,
-			BaseURL:         envconfig.Host().String() + "/v1",
-			APIKey:          "ollama",
-			Provider:        "generic-chat-completion-api",
-			MaxOutputTokens: maxOutput,
-			SupportsImages:  false,
-			ID:              modelID,
-			Index:           i,
-		})
-		if i == 0 {
-			defaultModelID = modelID
-		}
-	}
-
-	settingsMap["customModels"] = append(newModels, nonOllamaModels...)
-
-	// Update session default settings (preserve unknown fields in the nested object)
-	sessionSettings, ok := settingsMap["sessionDefaultSettings"].(map[string]any)
-	if !ok {
-		sessionSettings = make(map[string]any)
-	}
-	sessionSettings["model"] = defaultModelID
-
-	if !isValidReasoningEffort(settings.SessionDefaultSettings.ReasoningEffort) {
-		sessionSettings["reasoningEffort"] = "none"
-	}
-
-	settingsMap["sessionDefaultSettings"] = sessionSettings
-
-	data, err := json.MarshalIndent(settingsMap, "", "  ")
-	if err != nil {
-		return err
-	}
-	return writeWithBackup(settingsPath, data)
-}
-
-func (d *Droid) Models() []string {
-	home, err := os.UserHomeDir()
-	if err != nil {
-		return nil
-	}
-
-	data, err := os.ReadFile(filepath.Join(home, ".factory", "settings.json"))
-	if err != nil {
-		return nil
-	}
-
-	var settings droidSettings
-	if err := json.Unmarshal(data, &settings); err != nil {
-		return nil
-	}
-
-	var result []string
-	for _, m := range settings.CustomModels {
-		if m.APIKey == "ollama" {
-			result = append(result, m.Model)
-		}
-	}
-	return result
-}
-
-var validReasoningEfforts = []string{"high", "medium", "low", "none"}
-
-func isValidReasoningEffort(effort string) bool {
-	return slices.Contains(validReasoningEfforts, effort)
-}
--- a/cmd/config/droid_test.go
+++ b/cmd/config/droid_test.go
--- a/cmd/config/files.go
+++ b/cmd/config/files.go
@@ -1,99 +0,0 @@
-package config
-
-import (
-	"bytes"
-	"encoding/json"
-	"fmt"
-	"os"
-	"path/filepath"
-	"time"
-)
-
-func readJSONFile(path string) (map[string]any, error) {
-	data, err := os.ReadFile(path)
-	if err != nil {
-		return nil, err
-	}
-	var result map[string]any
-	if err := json.Unmarshal(data, &result); err != nil {
-		return nil, err
-	}
-	return result, nil
-}
-
-func copyFile(src, dst string) error {
-	info, err := os.Stat(src)
-	if err != nil {
-		return err
-	}
-	data, err := os.ReadFile(src)
-	if err != nil {
-		return err
-	}
-	return os.WriteFile(dst, data, info.Mode().Perm())
-}
-
-func backupDir() string {
-	return filepath.Join(os.TempDir(), "ollama-backups")
-}
-
-func backupToTmp(srcPath string) (string, error) {
-	dir := backupDir()
-	if err := os.MkdirAll(dir, 0o755); err != nil {
-		return "", err
-	}
-
-	backupPath := filepath.Join(dir, fmt.Sprintf("%s.%d", filepath.Base(srcPath), time.Now().Unix()))
-	if err := copyFile(srcPath, backupPath); err != nil {
-		return "", err
-	}
-	return backupPath, nil
-}
-
-// writeWithBackup writes data to path via temp file + rename, backing up any existing file first
-func writeWithBackup(path string, data []byte) error {
-	var backupPath string
-	// backup must be created before any writes to the target file
-	if existingContent, err := os.ReadFile(path); err == nil {
-		if !bytes.Equal(existingContent, data) {
-			backupPath, err = backupToTmp(path)
-			if err != nil {
-				return fmt.Errorf("backup failed: %w", err)
-			}
-		}
-	} else if !os.IsNotExist(err) {
-		return fmt.Errorf("read existing file: %w", err)
-	}
-
-	dir := filepath.Dir(path)
-	tmp, err := os.CreateTemp(dir, ".tmp-*")
-	if err != nil {
-		return fmt.Errorf("create temp failed: %w", err)
-	}
-	tmpPath := tmp.Name()
-
-	if _, err := tmp.Write(data); err != nil {
-		_ = tmp.Close()
-		_ = os.Remove(tmpPath)
-		return fmt.Errorf("write failed: %w", err)
-	}
-	if err := tmp.Sync(); err != nil {
-		_ = tmp.Close()
-		_ = os.Remove(tmpPath)
-		return fmt.Errorf("sync failed: %w", err)
-	}
-	if err := tmp.Close(); err != nil {
-		_ = os.Remove(tmpPath)
-		return fmt.Errorf("close failed: %w", err)
-	}
-
-	if err := os.Rename(tmpPath, path); err != nil {
-		_ = os.Remove(tmpPath)
-		if backupPath != "" {
-			_ = copyFile(backupPath, path)
-		}
-		return fmt.Errorf("rename failed: %w", err)
-	}
-
-	return nil
-}
--- a/cmd/config/files_test.go
+++ b/cmd/config/files_test.go
@@ -1,502 +0,0 @@
-package config
-
-import (
-	"encoding/json"
-	"fmt"
-	"os"
-	"path/filepath"
-	"runtime"
-	"testing"
-)
-
-func mustMarshal(t *testing.T, v any) []byte {
-	t.Helper()
-	data, err := json.MarshalIndent(v, "", "  ")
-	if err != nil {
-		t.Fatal(err)
-	}
-	return data
-}
-
-func TestWriteWithBackup(t *testing.T) {
-	tmpDir := t.TempDir()
-
-	t.Run("creates file", func(t *testing.T) {
-		path := filepath.Join(tmpDir, "new.json")
-		data := mustMarshal(t, map[string]string{"key": "value"})
-
-		if err := writeWithBackup(path, data); err != nil {
-			t.Fatal(err)
-		}
-
-		content, err := os.ReadFile(path)
-		if err != nil {
-			t.Fatal(err)
-		}
-
-		var result map[string]string
-		if err := json.Unmarshal(content, &result); err != nil {
-			t.Fatal(err)
-		}
-		if result["key"] != "value" {
-			t.Errorf("expected value, got %s", result["key"])
-		}
-	})
-
-	t.Run("creates backup in /tmp/ollama-backups", func(t *testing.T) {
-		path := filepath.Join(tmpDir, "backup.json")
-
-		os.WriteFile(path, []byte(`{"original": true}`), 0o644)
-
-		data := mustMarshal(t, map[string]bool{"updated": true})
-		if err := writeWithBackup(path, data); err != nil {
-			t.Fatal(err)
-		}
-
-		entries, err := os.ReadDir(backupDir())
-		if err != nil {
-			t.Fatal("backup directory not created")
-		}
-
-		var foundBackup bool
-		for _, entry := range entries {
-			if filepath.Ext(entry.Name()) != ".json" {
-				name := entry.Name()
-				if len(name) > len("backup.json.") && name[:len("backup.json.")] == "backup.json." {
-					backupPath := filepath.Join(backupDir(), name)
-					backup, err := os.ReadFile(backupPath)
-					if err == nil {
-						var backupData map[string]bool
-						json.Unmarshal(backup, &backupData)
-						if backupData["original"] {
-							foundBackup = true
-							os.Remove(backupPath)
-							break
-						}
-					}
-				}
-			}
-		}
-
-		if !foundBackup {
-			t.Error("backup file not created in /tmp/ollama-backups")
-		}
-
-		current, _ := os.ReadFile(path)
-		var currentData map[string]bool
-		json.Unmarshal(current, &currentData)
-		if !currentData["updated"] {
-			t.Error("file doesn't contain updated data")
-		}
-	})
-
-	t.Run("no backup for new file", func(t *testing.T) {
-		path := filepath.Join(tmpDir, "nobak.json")
-
-		data := mustMarshal(t, map[string]string{"new": "file"})
-		if err := writeWithBackup(path, data); err != nil {
-			t.Fatal(err)
-		}
-
-		entries, _ := os.ReadDir(backupDir())
-		for _, entry := range entries {
-			if len(entry.Name()) > len("nobak.json.") && entry.Name()[:len("nobak.json.")] == "nobak.json." {
-				t.Error("backup should not exist for new file")
-			}
-		}
-	})
-
-	t.Run("no backup when content unchanged", func(t *testing.T) {
-		path := filepath.Join(tmpDir, "unchanged.json")
-
-		data := mustMarshal(t, map[string]string{"key": "value"})
-
-		if err := writeWithBackup(path, data); err != nil {
-			t.Fatal(err)
-		}
-
-		entries1, _ := os.ReadDir(backupDir())
-		countBefore := 0
-		for _, e := range entries1 {
-			if len(e.Name()) > len("unchanged.json.") && e.Name()[:len("unchanged.json.")] == "unchanged.json." {
-				countBefore++
-			}
-		}
-
-		if err := writeWithBackup(path, data); err != nil {
-			t.Fatal(err)
-		}
-
-		entries2, _ := os.ReadDir(backupDir())
-		countAfter := 0
-		for _, e := range entries2 {
-			if len(e.Name()) > len("unchanged.json.") && e.Name()[:len("unchanged.json.")] == "unchanged.json." {
-				countAfter++
-			}
-		}
-
-		if countAfter != countBefore {
-			t.Errorf("backup was created when content unchanged (before=%d, after=%d)", countBefore, countAfter)
-		}
-	})
-
-	t.Run("backup filename contains unix timestamp", func(t *testing.T) {
-		path := filepath.Join(tmpDir, "timestamped.json")
-
-		os.WriteFile(path, []byte(`{"v": 1}`), 0o644)
-		data := mustMarshal(t, map[string]int{"v": 2})
-		if err := writeWithBackup(path, data); err != nil {
-			t.Fatal(err)
-		}
-
-		entries, _ := os.ReadDir(backupDir())
-		var found bool
-		for _, entry := range entries {
-			name := entry.Name()
-			if len(name) > len("timestamped.json.") && name[:len("timestamped.json.")] == "timestamped.json." {
-				timestamp := name[len("timestamped.json."):]
-				for _, c := range timestamp {
-					if c < '0' || c > '9' {
-						t.Errorf("backup filename timestamp contains non-numeric character: %s", name)
-					}
-				}
-				found = true
-				os.Remove(filepath.Join(backupDir(), name))
-				break
-			}
-		}
-		if !found {
-			t.Error("backup file with timestamp not found")
-		}
-	})
-}
-
-// Edge case tests for files.go
-
-// TestWriteWithBackup_FailsIfBackupFails documents critical behavior: if backup fails, we must not proceed.
-// User could lose their config with no way to recover.
-func TestWriteWithBackup_FailsIfBackupFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("permission tests unreliable on Windows")
-	}
-
-	tmpDir := t.TempDir()
-	path := filepath.Join(tmpDir, "config.json")
-
-	// Create original file
-	originalContent := []byte(`{"original": true}`)
-	os.WriteFile(path, originalContent, 0o644)
-
-	// Make backup directory read-only to force backup failure
-	backupDir := backupDir()
-	os.MkdirAll(backupDir, 0o755)
-	os.Chmod(backupDir, 0o444) // Read-only
-	defer os.Chmod(backupDir, 0o755)
-
-	newContent := []byte(`{"updated": true}`)
-	err := writeWithBackup(path, newContent)
-
-	// Should fail because backup couldn't be created
-	if err == nil {
-		t.Error("expected error when backup fails, got nil")
-	}
-
-	// Original file should be preserved
-	current, _ := os.ReadFile(path)
-	if string(current) != string(originalContent) {
-		t.Errorf("original file was modified despite backup failure: got %s", string(current))
-	}
-}
-
-// TestWriteWithBackup_PermissionDenied verifies clear error when target file has wrong permissions.
-// Common issue when config owned by root or wrong perms.
-func TestWriteWithBackup_PermissionDenied(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("permission tests unreliable on Windows")
-	}
-
-	tmpDir := t.TempDir()
-
-	// Create a read-only directory
-	readOnlyDir := filepath.Join(tmpDir, "readonly")
-	os.MkdirAll(readOnlyDir, 0o755)
-	os.Chmod(readOnlyDir, 0o444)
-	defer os.Chmod(readOnlyDir, 0o755)
-
-	path := filepath.Join(readOnlyDir, "config.json")
-	err := writeWithBackup(path, []byte(`{"test": true}`))
-
-	if err == nil {
-		t.Error("expected permission error, got nil")
-	}
-}
-
-// TestWriteWithBackup_DirectoryDoesNotExist verifies behavior when target directory doesn't exist.
-// writeWithBackup doesn't create directories - caller is responsible.
-func TestWriteWithBackup_DirectoryDoesNotExist(t *testing.T) {
-	tmpDir := t.TempDir()
-	path := filepath.Join(tmpDir, "nonexistent", "subdir", "config.json")
-
-	err := writeWithBackup(path, []byte(`{"test": true}`))
-
-	// Should fail because directory doesn't exist
-	if err == nil {
-		t.Error("expected error for nonexistent directory, got nil")
-	}
-}
-
-// TestWriteWithBackup_SymlinkTarget documents behavior when target is a symlink.
-// Documents what happens if user symlinks their config file.
-func TestWriteWithBackup_SymlinkTarget(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("symlink tests may require admin on Windows")
-	}
-
-	tmpDir := t.TempDir()
-	realFile := filepath.Join(tmpDir, "real.json")
-	symlink := filepath.Join(tmpDir, "link.json")
-
-	// Create real file and symlink
-	os.WriteFile(realFile, []byte(`{"v": 1}`), 0o644)
-	os.Symlink(realFile, symlink)
-
-	// Write through symlink
-	err := writeWithBackup(symlink, []byte(`{"v": 2}`))
-	if err != nil {
-		t.Fatalf("writeWithBackup through symlink failed: %v", err)
-	}
-
-	// The real file should be updated (symlink followed for temp file creation)
-	content, _ := os.ReadFile(symlink)
-	if string(content) != `{"v": 2}` {
-		t.Errorf("symlink target not updated correctly: got %s", string(content))
-	}
-}
-
-// TestBackupToTmp_SpecialCharsInFilename verifies backup works with special characters.
-// User may have config files with unusual names.
-func TestBackupToTmp_SpecialCharsInFilename(t *testing.T) {
-	tmpDir := t.TempDir()
-
-	// File with spaces and special chars
-	path := filepath.Join(tmpDir, "my config (backup).json")
-	os.WriteFile(path, []byte(`{"test": true}`), 0o644)
-
-	backupPath, err := backupToTmp(path)
-	if err != nil {
-		t.Fatalf("backupToTmp with special chars failed: %v", err)
-	}
-
-	// Verify backup exists and has correct content
-	content, err := os.ReadFile(backupPath)
-	if err != nil {
-		t.Fatalf("could not read backup: %v", err)
-	}
-	if string(content) != `{"test": true}` {
-		t.Errorf("backup content mismatch: got %s", string(content))
-	}
-
-	os.Remove(backupPath)
-}
-
-// TestCopyFile_PreservesPermissions verifies that copyFile preserves file permissions.
-func TestCopyFile_PreservesPermissions(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("permission preservation tests unreliable on Windows")
-	}
-
-	tmpDir := t.TempDir()
-	src := filepath.Join(tmpDir, "src.json")
-	dst := filepath.Join(tmpDir, "dst.json")
-
-	// Create source with specific permissions
-	os.WriteFile(src, []byte(`{"test": true}`), 0o600)
-
-	err := copyFile(src, dst)
-	if err != nil {
-		t.Fatalf("copyFile failed: %v", err)
-	}
-
-	srcInfo, _ := os.Stat(src)
-	dstInfo, _ := os.Stat(dst)
-
-	if srcInfo.Mode().Perm() != dstInfo.Mode().Perm() {
-		t.Errorf("permissions not preserved: src=%v, dst=%v", srcInfo.Mode().Perm(), dstInfo.Mode().Perm())
-	}
-}
-
-// TestCopyFile_SourceNotFound verifies clear error when source doesn't exist.
-func TestCopyFile_SourceNotFound(t *testing.T) {
-	tmpDir := t.TempDir()
-	src := filepath.Join(tmpDir, "nonexistent.json")
-	dst := filepath.Join(tmpDir, "dst.json")
-
-	err := copyFile(src, dst)
-	if err == nil {
-		t.Error("expected error for nonexistent source, got nil")
-	}
-}
-
-// TestWriteWithBackup_TargetIsDirectory verifies error when path points to a directory.
-func TestWriteWithBackup_TargetIsDirectory(t *testing.T) {
-	tmpDir := t.TempDir()
-	dirPath := filepath.Join(tmpDir, "actualdir")
-	os.MkdirAll(dirPath, 0o755)
-
-	err := writeWithBackup(dirPath, []byte(`{"test": true}`))
-	if err == nil {
-		t.Error("expected error when target is a directory, got nil")
-	}
-}
-
-// TestWriteWithBackup_EmptyData verifies writing zero bytes works correctly.
-func TestWriteWithBackup_EmptyData(t *testing.T) {
-	tmpDir := t.TempDir()
-	path := filepath.Join(tmpDir, "empty.json")
-
-	err := writeWithBackup(path, []byte{})
-	if err != nil {
-		t.Fatalf("writeWithBackup with empty data failed: %v", err)
-	}
-
-	content, err := os.ReadFile(path)
-	if err != nil {
-		t.Fatalf("could not read file: %v", err)
-	}
-	if len(content) != 0 {
-		t.Errorf("expected empty file, got %d bytes", len(content))
-	}
-}
-
-// TestWriteWithBackup_FileUnreadableButDirWritable verifies behavior when existing file
-// cannot be read (for backup comparison) but directory is writable.
-func TestWriteWithBackup_FileUnreadableButDirWritable(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("permission tests unreliable on Windows")
-	}
-
-	tmpDir := t.TempDir()
-	path := filepath.Join(tmpDir, "unreadable.json")
-
-	// Create file and make it unreadable
-	os.WriteFile(path, []byte(`{"original": true}`), 0o644)
-	os.Chmod(path, 0o000)
-	defer os.Chmod(path, 0o644)
-
-	// Should fail because we can't read the file to compare/backup
-	err := writeWithBackup(path, []byte(`{"updated": true}`))
-	if err == nil {
-		t.Error("expected error when file is unreadable, got nil")
-	}
-}
-
-// TestWriteWithBackup_RapidSuccessiveWrites verifies backup works with multiple writes
-// within the same second (timestamp collision scenario).
-func TestWriteWithBackup_RapidSuccessiveWrites(t *testing.T) {
-	tmpDir := t.TempDir()
-	path := filepath.Join(tmpDir, "rapid.json")
-
-	// Create initial file
-	os.WriteFile(path, []byte(`{"v": 0}`), 0o644)
-
-	// Rapid successive writes
-	for i := 1; i <= 3; i++ {
-		data := []byte(fmt.Sprintf(`{"v": %d}`, i))
-		if err := writeWithBackup(path, data); err != nil {
-			t.Fatalf("write %d failed: %v", i, err)
-		}
-	}
-
-	// Verify final content
-	content, _ := os.ReadFile(path)
-	if string(content) != `{"v": 3}` {
-		t.Errorf("expected final content {\"v\": 3}, got %s", string(content))
-	}
-
-	// Verify at least one backup exists
-	entries, _ := os.ReadDir(backupDir())
-	var backupCount int
-	for _, e := range entries {
-		if len(e.Name()) > len("rapid.json.") && e.Name()[:len("rapid.json.")] == "rapid.json." {
-			backupCount++
-		}
-	}
-	if backupCount == 0 {
-		t.Error("expected at least one backup file from rapid writes")
-	}
-}
-
-// TestWriteWithBackup_BackupDirIsFile verifies error when backup directory path is a file.
-func TestWriteWithBackup_BackupDirIsFile(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("test modifies system temp directory")
-	}
-
-	// Create a file at the backup directory path
-	backupPath := backupDir()
-	// Clean up any existing directory first
-	os.RemoveAll(backupPath)
-	// Create a file instead of directory
-	os.WriteFile(backupPath, []byte("not a directory"), 0o644)
-	defer func() {
-		os.Remove(backupPath)
-		os.MkdirAll(backupPath, 0o755)
-	}()
-
-	tmpDir := t.TempDir()
-	path := filepath.Join(tmpDir, "test.json")
-	os.WriteFile(path, []byte(`{"original": true}`), 0o644)
-
-	err := writeWithBackup(path, []byte(`{"updated": true}`))
-	if err == nil {
-		t.Error("expected error when backup dir is a file, got nil")
-	}
-}
-
-// TestWriteWithBackup_NoOrphanTempFiles verifies temp files are cleaned up on failure.
-func TestWriteWithBackup_NoOrphanTempFiles(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("permission tests unreliable on Windows")
-	}
-
-	tmpDir := t.TempDir()
-
-	// Count existing temp files
-	countTempFiles := func() int {
-		entries, _ := os.ReadDir(tmpDir)
-		count := 0
-		for _, e := range entries {
-			if len(e.Name()) > 4 && e.Name()[:4] == ".tmp" {
-				count++
-			}
-		}
-		return count
-	}
-
-	before := countTempFiles()
-
-	// Create a file, then make directory read-only to cause rename failure
-	path := filepath.Join(tmpDir, "orphan.json")
-	os.WriteFile(path, []byte(`{"v": 1}`), 0o644)
-
-	// Make a subdirectory and try to write there after making parent read-only
-	subDir := filepath.Join(tmpDir, "subdir")
-	os.MkdirAll(subDir, 0o755)
-	subPath := filepath.Join(subDir, "config.json")
-	os.WriteFile(subPath, []byte(`{"v": 1}`), 0o644)
-
-	// Make subdir read-only after creating temp file would succeed but rename would fail
-	// This is tricky to test - the temp file is created in the same dir, so if we can't
-	// rename, we also couldn't create. Let's just verify normal failure cleanup works.
-
-	// Force a failure by making the target a directory
-	badPath := filepath.Join(tmpDir, "isdir")
-	os.MkdirAll(badPath, 0o755)
-
-	_ = writeWithBackup(badPath, []byte(`{"test": true}`))
-
-	after := countTempFiles()
-	if after > before {
-		t.Errorf("orphan temp files left behind: before=%d, after=%d", before, after)
-	}
-}
--- a/cmd/config/integrations.go
+++ b/cmd/config/integrations.go
--- a/cmd/config/integrations_test.go
+++ b/cmd/config/integrations_test.go
--- a/cmd/config/openclaw.go
+++ b/cmd/config/openclaw.go
@@ -1,252 +0,0 @@
-package config
-
-import (
-	"bytes"
-	"encoding/json"
-	"fmt"
-	"io"
-	"os"
-	"os/exec"
-	"path/filepath"
-	"strings"
-
-	"github.com/ollama/ollama/envconfig"
-)
-
-type Openclaw struct{}
-
-func (c *Openclaw) String() string { return "OpenClaw" }
-
-func (c *Openclaw) Run(model string, args []string) error {
-	bin := "openclaw"
-	if _, err := exec.LookPath(bin); err != nil {
-		bin = "clawdbot"
-		if _, err := exec.LookPath(bin); err != nil {
-			return fmt.Errorf("openclaw is not installed, install from https://docs.openclaw.ai")
-		}
-	}
-
-	models := []string{model}
-	if config, err := loadIntegration("openclaw"); err == nil && len(config.Models) > 0 {
-		models = config.Models
-	} else if config, err := loadIntegration("clawdbot"); err == nil && len(config.Models) > 0 {
-		models = config.Models
-	}
-	if err := c.Edit(models); err != nil {
-		return fmt.Errorf("setup failed: %w", err)
-	}
-
-	if !c.onboarded() {
-		// Onboarding not completed: run it (model already set via Edit)
-		// Use "ollama" as gateway token for simple local access
-		cmd := exec.Command(bin, "onboard",
-			"--auth-choice", "skip",
-			"--gateway-token", "ollama",
-		)
-		cmd.Stdin = os.Stdin
-		cmd.Stdout = os.Stdout
-		cmd.Stderr = os.Stderr
-		return cmd.Run()
-	}
-
-	// Onboarding completed: run gateway
-	cmd := exec.Command(bin, append([]string{"gateway"}, args...)...)
-	cmd.Stdin = os.Stdin
-
-	// Capture output to detect "already running" message
-	var outputBuf bytes.Buffer
-	cmd.Stdout = io.MultiWriter(os.Stdout, &outputBuf)
-	cmd.Stderr = io.MultiWriter(os.Stderr, &outputBuf)
-
-	err := cmd.Run()
-	if err != nil && strings.Contains(outputBuf.String(), "Gateway already running") {
-		fmt.Fprintf(os.Stderr, "%sOpenClaw has been configured with Ollama. Gateway is already running.%s\n", ansiGreen, ansiReset)
-		return nil
-	}
-	return err
-}
-
-// onboarded checks if OpenClaw onboarding wizard was completed
-// by looking for the wizard.lastRunAt marker in the config
-func (c *Openclaw) onboarded() bool {
-	home, err := os.UserHomeDir()
-	if err != nil {
-		return false
-	}
-
-	configPath := filepath.Join(home, ".openclaw", "openclaw.json")
-	legacyPath := filepath.Join(home, ".clawdbot", "clawdbot.json")
-
-	config := make(map[string]any)
-	if data, err := os.ReadFile(configPath); err == nil {
-		_ = json.Unmarshal(data, &config)
-	} else if data, err := os.ReadFile(legacyPath); err == nil {
-		_ = json.Unmarshal(data, &config)
-	} else {
-		return false
-	}
-
-	// Check for wizard.lastRunAt marker (set when onboarding completes)
-	wizard, _ := config["wizard"].(map[string]any)
-	if wizard == nil {
-		return false
-	}
-	lastRunAt, _ := wizard["lastRunAt"].(string)
-	return lastRunAt != ""
-}
-
-func (c *Openclaw) Paths() []string {
-	home, _ := os.UserHomeDir()
-	p := filepath.Join(home, ".openclaw", "openclaw.json")
-	if _, err := os.Stat(p); err == nil {
-		return []string{p}
-	}
-	legacy := filepath.Join(home, ".clawdbot", "clawdbot.json")
-	if _, err := os.Stat(legacy); err == nil {
-		return []string{legacy}
-	}
-	return nil
-}
-
-func (c *Openclaw) Edit(models []string) error {
-	if len(models) == 0 {
-		return nil
-	}
-
-	home, err := os.UserHomeDir()
-	if err != nil {
-		return err
-	}
-
-	configPath := filepath.Join(home, ".openclaw", "openclaw.json")
-	legacyPath := filepath.Join(home, ".clawdbot", "clawdbot.json")
-	if err := os.MkdirAll(filepath.Dir(configPath), 0o755); err != nil {
-		return err
-	}
-
-	// Read into map[string]any to preserve unknown fields
-	config := make(map[string]any)
-	if data, err := os.ReadFile(configPath); err == nil {
-		_ = json.Unmarshal(data, &config)
-	} else if data, err := os.ReadFile(legacyPath); err == nil {
-		_ = json.Unmarshal(data, &config)
-	}
-
-	// Navigate/create: models.providers.ollama (preserving other providers)
-	modelsSection, _ := config["models"].(map[string]any)
-	if modelsSection == nil {
-		modelsSection = make(map[string]any)
-	}
-	providers, _ := modelsSection["providers"].(map[string]any)
-	if providers == nil {
-		providers = make(map[string]any)
-	}
-	ollama, _ := providers["ollama"].(map[string]any)
-	if ollama == nil {
-		ollama = make(map[string]any)
-	}
-
-	ollama["baseUrl"] = envconfig.Host().String() + "/v1"
-	// needed to register provider
-	ollama["apiKey"] = "ollama-local"
-	// TODO(parthsareen): potentially move to responses
-	ollama["api"] = "openai-completions"
-
-	// Build map of existing models to preserve user customizations
-	existingModels, _ := ollama["models"].([]any)
-	existingByID := make(map[string]map[string]any)
-	for _, m := range existingModels {
-		if entry, ok := m.(map[string]any); ok {
-			if id, ok := entry["id"].(string); ok {
-				existingByID[id] = entry
-			}
-		}
-	}
-
-	var newModels []any
-	for _, model := range models {
-		entry := map[string]any{
-			"id":        model,
-			"name":      model,
-			"reasoning": false,
-			"input":     []any{"text"},
-			"cost": map[string]any{
-				"input":      0,
-				"output":     0,
-				"cacheRead":  0,
-				"cacheWrite": 0,
-			},
-			// TODO(parthsareen): get these values from API
-			"contextWindow": 131072,
-			"maxTokens":     16384,
-		}
-		// Merge existing fields (user customizations)
-		if existing, ok := existingByID[model]; ok {
-			for k, v := range existing {
-				if _, isNew := entry[k]; !isNew {
-					entry[k] = v
-				}
-			}
-		}
-		newModels = append(newModels, entry)
-	}
-	ollama["models"] = newModels
-
-	providers["ollama"] = ollama
-	modelsSection["providers"] = providers
-	config["models"] = modelsSection
-
-	// Update agents.defaults.model.primary (preserving other agent settings)
-	agents, _ := config["agents"].(map[string]any)
-	if agents == nil {
-		agents = make(map[string]any)
-	}
-	defaults, _ := agents["defaults"].(map[string]any)
-	if defaults == nil {
-		defaults = make(map[string]any)
-	}
-	modelConfig, _ := defaults["model"].(map[string]any)
-	if modelConfig == nil {
-		modelConfig = make(map[string]any)
-	}
-	modelConfig["primary"] = "ollama/" + models[0]
-	defaults["model"] = modelConfig
-	agents["defaults"] = defaults
-	config["agents"] = agents
-
-	data, err := json.MarshalIndent(config, "", "  ")
-	if err != nil {
-		return err
-	}
-	return writeWithBackup(configPath, data)
-}
-
-func (c *Openclaw) Models() []string {
-	home, err := os.UserHomeDir()
-	if err != nil {
-		return nil
-	}
-
-	config, err := readJSONFile(filepath.Join(home, ".openclaw", "openclaw.json"))
-	if err != nil {
-		config, err = readJSONFile(filepath.Join(home, ".clawdbot", "clawdbot.json"))
-		if err != nil {
-			return nil
-		}
-	}
-
-	modelsSection, _ := config["models"].(map[string]any)
-	providers, _ := modelsSection["providers"].(map[string]any)
-	ollama, _ := providers["ollama"].(map[string]any)
-	modelList, _ := ollama["models"].([]any)
-
-	var result []string
-	for _, m := range modelList {
-		if entry, ok := m.(map[string]any); ok {
-			if id, ok := entry["id"].(string); ok {
-				result = append(result, id)
-			}
-		}
-	}
-	return result
-}
--- a/cmd/config/openclaw_test.go
+++ b/cmd/config/openclaw_test.go
@@ -1,878 +0,0 @@
-package config
-
-import (
-	"encoding/json"
-	"fmt"
-	"os"
-	"path/filepath"
-	"testing"
-)
-
-func TestOpenclawIntegration(t *testing.T) {
-	c := &Openclaw{}
-
-	t.Run("String", func(t *testing.T) {
-		if got := c.String(); got != "OpenClaw" {
-			t.Errorf("String() = %q, want %q", got, "OpenClaw")
-		}
-	})
-
-	t.Run("implements Runner", func(t *testing.T) {
-		var _ Runner = c
-	})
-
-	t.Run("implements Editor", func(t *testing.T) {
-		var _ Editor = c
-	})
-}
-
-func TestOpenclawEdit(t *testing.T) {
-	c := &Openclaw{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	configDir := filepath.Join(tmpDir, ".openclaw")
-	configPath := filepath.Join(configDir, "openclaw.json")
-
-	cleanup := func() { os.RemoveAll(configDir) }
-
-	t.Run("fresh install", func(t *testing.T) {
-		cleanup()
-		if err := c.Edit([]string{"llama3.2"}); err != nil {
-			t.Fatal(err)
-		}
-		assertOpenclawModelExists(t, configPath, "llama3.2")
-		assertOpenclawPrimaryModel(t, configPath, "ollama/llama3.2")
-	})
-
-	t.Run("multiple models - first is primary", func(t *testing.T) {
-		cleanup()
-		if err := c.Edit([]string{"llama3.2", "mistral"}); err != nil {
-			t.Fatal(err)
-		}
-		assertOpenclawModelExists(t, configPath, "llama3.2")
-		assertOpenclawModelExists(t, configPath, "mistral")
-		assertOpenclawPrimaryModel(t, configPath, "ollama/llama3.2")
-	})
-
-	t.Run("preserve other providers", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(configPath, []byte(`{"models":{"providers":{"anthropic":{"apiKey":"xxx"}}}}`), 0o644)
-		if err := c.Edit([]string{"llama3.2"}); err != nil {
-			t.Fatal(err)
-		}
-		data, _ := os.ReadFile(configPath)
-		var cfg map[string]any
-		json.Unmarshal(data, &cfg)
-		models := cfg["models"].(map[string]any)
-		providers := models["providers"].(map[string]any)
-		if providers["anthropic"] == nil {
-			t.Error("anthropic provider was removed")
-		}
-	})
-
-	t.Run("preserve top-level keys", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(configPath, []byte(`{"theme":"dark","mcp":{"servers":{}}}`), 0o644)
-		if err := c.Edit([]string{"llama3.2"}); err != nil {
-			t.Fatal(err)
-		}
-		data, _ := os.ReadFile(configPath)
-		var cfg map[string]any
-		json.Unmarshal(data, &cfg)
-		if cfg["theme"] != "dark" {
-			t.Error("theme was removed")
-		}
-		if cfg["mcp"] == nil {
-			t.Error("mcp was removed")
-		}
-	})
-
-	t.Run("preserve user customizations on models", func(t *testing.T) {
-		cleanup()
-		c.Edit([]string{"llama3.2"})
-
-		// User adds custom field
-		data, _ := os.ReadFile(configPath)
-		var cfg map[string]any
-		json.Unmarshal(data, &cfg)
-		models := cfg["models"].(map[string]any)
-		providers := models["providers"].(map[string]any)
-		ollama := providers["ollama"].(map[string]any)
-		modelList := ollama["models"].([]any)
-		entry := modelList[0].(map[string]any)
-		entry["customField"] = "user-value"
-		configData, _ := json.MarshalIndent(cfg, "", "  ")
-		os.WriteFile(configPath, configData, 0o644)
-
-		// Re-run Edit
-		c.Edit([]string{"llama3.2"})
-
-		data, _ = os.ReadFile(configPath)
-		json.Unmarshal(data, &cfg)
-		models = cfg["models"].(map[string]any)
-		providers = models["providers"].(map[string]any)
-		ollama = providers["ollama"].(map[string]any)
-		modelList = ollama["models"].([]any)
-		entry = modelList[0].(map[string]any)
-		if entry["customField"] != "user-value" {
-			t.Error("custom field was lost")
-		}
-	})
-
-	t.Run("edit replaces models list", func(t *testing.T) {
-		cleanup()
-		c.Edit([]string{"llama3.2", "mistral"})
-		c.Edit([]string{"llama3.2"})
-
-		assertOpenclawModelExists(t, configPath, "llama3.2")
-		assertOpenclawModelNotExists(t, configPath, "mistral")
-	})
-
-	t.Run("empty models is no-op", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-		original := `{"existing":"data"}`
-		os.WriteFile(configPath, []byte(original), 0o644)
-
-		c.Edit([]string{})
-
-		data, _ := os.ReadFile(configPath)
-		if string(data) != original {
-			t.Error("empty models should not modify file")
-		}
-	})
-
-	t.Run("corrupted JSON treated as empty", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(configPath, []byte(`{corrupted`), 0o644)
-
-		if err := c.Edit([]string{"llama3.2"}); err != nil {
-			t.Fatal(err)
-		}
-
-		data, _ := os.ReadFile(configPath)
-		var cfg map[string]any
-		if err := json.Unmarshal(data, &cfg); err != nil {
-			t.Error("result should be valid JSON")
-		}
-	})
-
-	t.Run("wrong type models section", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(configPath, []byte(`{"models":"not a map"}`), 0o644)
-
-		if err := c.Edit([]string{"llama3.2"}); err != nil {
-			t.Fatal(err)
-		}
-		assertOpenclawModelExists(t, configPath, "llama3.2")
-	})
-}
-
-func TestOpenclawModels(t *testing.T) {
-	c := &Openclaw{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	t.Run("no config returns nil", func(t *testing.T) {
-		if models := c.Models(); len(models) > 0 {
-			t.Errorf("expected nil/empty, got %v", models)
-		}
-	})
-
-	t.Run("returns all ollama models", func(t *testing.T) {
-		configDir := filepath.Join(tmpDir, ".openclaw")
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(filepath.Join(configDir, "openclaw.json"), []byte(`{
-			"models":{"providers":{"ollama":{"models":[
-				{"id":"llama3.2"},
-				{"id":"mistral"}
-			]}}}
-		}`), 0o644)
-
-		models := c.Models()
-		if len(models) != 2 {
-			t.Errorf("expected 2 models, got %v", models)
-		}
-	})
-}
-
-// Helper functions
-func assertOpenclawModelExists(t *testing.T, path, model string) {
-	t.Helper()
-	data, _ := os.ReadFile(path)
-	var cfg map[string]any
-	json.Unmarshal(data, &cfg)
-	models := cfg["models"].(map[string]any)
-	providers := models["providers"].(map[string]any)
-	ollama := providers["ollama"].(map[string]any)
-	modelList := ollama["models"].([]any)
-	for _, m := range modelList {
-		if entry, ok := m.(map[string]any); ok {
-			if entry["id"] == model {
-				return
-			}
-		}
-	}
-	t.Errorf("model %s not found", model)
-}
-
-func assertOpenclawModelNotExists(t *testing.T, path, model string) {
-	t.Helper()
-	data, _ := os.ReadFile(path)
-	var cfg map[string]any
-	json.Unmarshal(data, &cfg)
-	models, _ := cfg["models"].(map[string]any)
-	providers, _ := models["providers"].(map[string]any)
-	ollama, _ := providers["ollama"].(map[string]any)
-	modelList, _ := ollama["models"].([]any)
-	for _, m := range modelList {
-		if entry, ok := m.(map[string]any); ok {
-			if entry["id"] == model {
-				t.Errorf("model %s should not exist", model)
-			}
-		}
-	}
-}
-
-func assertOpenclawPrimaryModel(t *testing.T, path, expected string) {
-	t.Helper()
-	data, _ := os.ReadFile(path)
-	var cfg map[string]any
-	json.Unmarshal(data, &cfg)
-	agents := cfg["agents"].(map[string]any)
-	defaults := agents["defaults"].(map[string]any)
-	model := defaults["model"].(map[string]any)
-	if model["primary"] != expected {
-		t.Errorf("primary model = %v, want %v", model["primary"], expected)
-	}
-}
-
-func TestOpenclawPaths(t *testing.T) {
-	c := &Openclaw{}
-
-	t.Run("returns path when config exists", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-		configDir := filepath.Join(tmpDir, ".openclaw")
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(filepath.Join(configDir, "openclaw.json"), []byte(`{}`), 0o644)
-
-		paths := c.Paths()
-		if len(paths) != 1 {
-			t.Errorf("expected 1 path, got %d", len(paths))
-		}
-	})
-
-	t.Run("returns nil when config missing", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-		if paths := c.Paths(); paths != nil {
-			t.Errorf("expected nil, got %v", paths)
-		}
-	})
-}
-
-func TestOpenclawModelsEdgeCases(t *testing.T) {
-	c := &Openclaw{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-	configDir := filepath.Join(tmpDir, ".openclaw")
-	configPath := filepath.Join(configDir, "openclaw.json")
-	cleanup := func() { os.RemoveAll(configDir) }
-
-	t.Run("corrupted JSON returns nil", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(configPath, []byte(`{corrupted`), 0o644)
-		if models := c.Models(); models != nil {
-			t.Errorf("expected nil, got %v", models)
-		}
-	})
-
-	t.Run("wrong type at models level", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(configPath, []byte(`{"models":"string"}`), 0o644)
-		if models := c.Models(); models != nil {
-			t.Errorf("expected nil, got %v", models)
-		}
-	})
-
-	t.Run("wrong type at providers level", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(configPath, []byte(`{"models":{"providers":"string"}}`), 0o644)
-		if models := c.Models(); models != nil {
-			t.Errorf("expected nil, got %v", models)
-		}
-	})
-
-	t.Run("wrong type at ollama level", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(configPath, []byte(`{"models":{"providers":{"ollama":"string"}}}`), 0o644)
-		if models := c.Models(); models != nil {
-			t.Errorf("expected nil, got %v", models)
-		}
-	})
-
-	t.Run("model entry missing id", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(configPath, []byte(`{"models":{"providers":{"ollama":{"models":[{"name":"test"}]}}}}`), 0o644)
-		if len(c.Models()) != 0 {
-			t.Error("expected empty for missing id")
-		}
-	})
-
-	t.Run("model id is not string", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(configPath, []byte(`{"models":{"providers":{"ollama":{"models":[{"id":123}]}}}}`), 0o644)
-		if len(c.Models()) != 0 {
-			t.Error("expected empty for non-string id")
-		}
-	})
-}
-
-func TestOpenclawEditSchemaFields(t *testing.T) {
-	c := &Openclaw{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-	configPath := filepath.Join(tmpDir, ".openclaw", "openclaw.json")
-
-	if err := c.Edit([]string{"llama3.2"}); err != nil {
-		t.Fatal(err)
-	}
-
-	data, _ := os.ReadFile(configPath)
-	var cfg map[string]any
-	json.Unmarshal(data, &cfg)
-	models := cfg["models"].(map[string]any)
-	providers := models["providers"].(map[string]any)
-	ollama := providers["ollama"].(map[string]any)
-	modelList := ollama["models"].([]any)
-	entry := modelList[0].(map[string]any)
-
-	// Verify required schema fields
-	if entry["reasoning"] != false {
-		t.Error("reasoning should be false")
-	}
-	if entry["input"] == nil {
-		t.Error("input should be set")
-	}
-	if entry["contextWindow"] == nil {
-		t.Error("contextWindow should be set")
-	}
-	if entry["maxTokens"] == nil {
-		t.Error("maxTokens should be set")
-	}
-	cost := entry["cost"].(map[string]any)
-	if cost["cacheRead"] == nil {
-		t.Error("cost.cacheRead should be set")
-	}
-	if cost["cacheWrite"] == nil {
-		t.Error("cost.cacheWrite should be set")
-	}
-}
-
-func TestOpenclawEditModelNames(t *testing.T) {
-	c := &Openclaw{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-	configPath := filepath.Join(tmpDir, ".openclaw", "openclaw.json")
-	cleanup := func() { os.RemoveAll(filepath.Join(tmpDir, ".openclaw")) }
-
-	t.Run("model with colon tag", func(t *testing.T) {
-		cleanup()
-		if err := c.Edit([]string{"llama3.2:70b"}); err != nil {
-			t.Fatal(err)
-		}
-		assertOpenclawModelExists(t, configPath, "llama3.2:70b")
-		assertOpenclawPrimaryModel(t, configPath, "ollama/llama3.2:70b")
-	})
-
-	t.Run("model with slash", func(t *testing.T) {
-		cleanup()
-		if err := c.Edit([]string{"library/model:tag"}); err != nil {
-			t.Fatal(err)
-		}
-		assertOpenclawModelExists(t, configPath, "library/model:tag")
-		assertOpenclawPrimaryModel(t, configPath, "ollama/library/model:tag")
-	})
-
-	t.Run("model with hyphen", func(t *testing.T) {
-		cleanup()
-		if err := c.Edit([]string{"test-model"}); err != nil {
-			t.Fatal(err)
-		}
-		assertOpenclawModelExists(t, configPath, "test-model")
-	})
-}
-
-func TestOpenclawEditAgentsPreservation(t *testing.T) {
-	c := &Openclaw{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-	configDir := filepath.Join(tmpDir, ".openclaw")
-	configPath := filepath.Join(configDir, "openclaw.json")
-	cleanup := func() { os.RemoveAll(configDir) }
-
-	t.Run("preserve other agent defaults", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(configPath, []byte(`{"agents":{"defaults":{"model":{"primary":"old"},"temperature":0.7}}}`), 0o644)
-
-		c.Edit([]string{"llama3.2"})
-
-		data, _ := os.ReadFile(configPath)
-		var cfg map[string]any
-		json.Unmarshal(data, &cfg)
-		agents := cfg["agents"].(map[string]any)
-		defaults := agents["defaults"].(map[string]any)
-		if defaults["temperature"] != 0.7 {
-			t.Error("temperature setting was lost")
-		}
-	})
-
-	t.Run("preserve other agents besides defaults", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(configPath, []byte(`{"agents":{"defaults":{},"custom-agent":{"foo":"bar"}}}`), 0o644)
-
-		c.Edit([]string{"llama3.2"})
-
-		data, _ := os.ReadFile(configPath)
-		var cfg map[string]any
-		json.Unmarshal(data, &cfg)
-		agents := cfg["agents"].(map[string]any)
-		if agents["custom-agent"] == nil {
-			t.Error("custom-agent was lost")
-		}
-	})
-}
-
-const testOpenclawFixture = `{
-  "theme": "dark",
-  "mcp": {"servers": {"custom": {"enabled": true}}},
-  "models": {
-    "providers": {
-      "anthropic": {"apiKey": "xxx"},
-      "ollama": {
-        "baseUrl": "http://127.0.0.1:11434/v1",
-        "models": [{"id": "old-model", "customField": "preserved"}]
-      }
-    }
-  },
-  "agents": {
-    "defaults": {"model": {"primary": "old"}, "temperature": 0.7},
-    "custom-agent": {"foo": "bar"}
-  }
-}`
-
-func TestOpenclawEdit_RoundTrip(t *testing.T) {
-	c := &Openclaw{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-	configDir := filepath.Join(tmpDir, ".openclaw")
-	configPath := filepath.Join(configDir, "openclaw.json")
-
-	os.MkdirAll(configDir, 0o755)
-	os.WriteFile(configPath, []byte(testOpenclawFixture), 0o644)
-
-	if err := c.Edit([]string{"llama3.2", "mistral"}); err != nil {
-		t.Fatal(err)
-	}
-
-	data, _ := os.ReadFile(configPath)
-	var cfg map[string]any
-	json.Unmarshal(data, &cfg)
-
-	// Verify top-level preserved
-	if cfg["theme"] != "dark" {
-		t.Error("theme not preserved")
-	}
-	mcp := cfg["mcp"].(map[string]any)
-	servers := mcp["servers"].(map[string]any)
-	if servers["custom"] == nil {
-		t.Error("mcp.servers.custom not preserved")
-	}
-
-	// Verify other providers preserved
-	models := cfg["models"].(map[string]any)
-	providers := models["providers"].(map[string]any)
-	if providers["anthropic"] == nil {
-		t.Error("anthropic provider not preserved")
-	}
-
-	// Verify agents preserved
-	agents := cfg["agents"].(map[string]any)
-	if agents["custom-agent"] == nil {
-		t.Error("custom-agent not preserved")
-	}
-	defaults := agents["defaults"].(map[string]any)
-	if defaults["temperature"] != 0.7 {
-		t.Error("temperature not preserved")
-	}
-}
-
-func TestOpenclawEdit_Idempotent(t *testing.T) {
-	c := &Openclaw{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-	configDir := filepath.Join(tmpDir, ".openclaw")
-	configPath := filepath.Join(configDir, "openclaw.json")
-
-	os.MkdirAll(configDir, 0o755)
-	os.WriteFile(configPath, []byte(testOpenclawFixture), 0o644)
-
-	c.Edit([]string{"llama3.2", "mistral"})
-	firstData, _ := os.ReadFile(configPath)
-
-	c.Edit([]string{"llama3.2", "mistral"})
-	secondData, _ := os.ReadFile(configPath)
-
-	if string(firstData) != string(secondData) {
-		t.Error("repeated edits with same models produced different results")
-	}
-}
-
-func TestOpenclawEdit_MultipleConsecutiveEdits(t *testing.T) {
-	c := &Openclaw{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-	configDir := filepath.Join(tmpDir, ".openclaw")
-	configPath := filepath.Join(configDir, "openclaw.json")
-
-	os.MkdirAll(configDir, 0o755)
-	os.WriteFile(configPath, []byte(testOpenclawFixture), 0o644)
-
-	for i := range 10 {
-		models := []string{"model-a", "model-b"}
-		if i%2 == 0 {
-			models = []string{"model-x", "model-y", "model-z"}
-		}
-		if err := c.Edit(models); err != nil {
-			t.Fatalf("edit %d failed: %v", i, err)
-		}
-	}
-
-	data, _ := os.ReadFile(configPath)
-	var cfg map[string]any
-	if err := json.Unmarshal(data, &cfg); err != nil {
-		t.Fatalf("file is not valid JSON after multiple edits: %v", err)
-	}
-
-	if cfg["theme"] != "dark" {
-		t.Error("theme lost after multiple edits")
-	}
-}
-
-func TestOpenclawEdit_BackupCreated(t *testing.T) {
-	c := &Openclaw{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-	configDir := filepath.Join(tmpDir, ".openclaw")
-	configPath := filepath.Join(configDir, "openclaw.json")
-	backupDir := filepath.Join(os.TempDir(), "ollama-backups")
-
-	os.MkdirAll(configDir, 0o755)
-	uniqueMarker := fmt.Sprintf("test-marker-%d", os.Getpid())
-	original := fmt.Sprintf(`{"theme": "%s"}`, uniqueMarker)
-	os.WriteFile(configPath, []byte(original), 0o644)
-
-	if err := c.Edit([]string{"model-a"}); err != nil {
-		t.Fatal(err)
-	}
-
-	backups, _ := filepath.Glob(filepath.Join(backupDir, "openclaw.json.*"))
-	foundBackup := false
-	for _, backup := range backups {
-		data, _ := os.ReadFile(backup)
-		if string(data) == original {
-			foundBackup = true
-			break
-		}
-	}
-
-	if !foundBackup {
-		t.Error("backup with original content not found")
-	}
-}
-
-func TestOpenclawClawdbotAlias(t *testing.T) {
-	for _, alias := range []string{"clawdbot", "moltbot"} {
-		t.Run(alias+" alias resolves to Openclaw runner", func(t *testing.T) {
-			r, ok := integrations[alias]
-			if !ok {
-				t.Fatalf("%s not found in integrations", alias)
-			}
-			if _, ok := r.(*Openclaw); !ok {
-				t.Errorf("%s integration is %T, want *Openclaw", alias, r)
-			}
-		})
-
-		t.Run(alias+" is hidden from selector", func(t *testing.T) {
-			if !integrationAliases[alias] {
-				t.Errorf("%s should be in integrationAliases", alias)
-			}
-		})
-	}
-}
-
-func TestOpenclawLegacyPaths(t *testing.T) {
-	c := &Openclaw{}
-
-	t.Run("falls back to legacy clawdbot path", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-		legacyDir := filepath.Join(tmpDir, ".clawdbot")
-		os.MkdirAll(legacyDir, 0o755)
-		os.WriteFile(filepath.Join(legacyDir, "clawdbot.json"), []byte(`{}`), 0o644)
-
-		paths := c.Paths()
-		if len(paths) != 1 {
-			t.Fatalf("expected 1 path, got %d", len(paths))
-		}
-		if paths[0] != filepath.Join(legacyDir, "clawdbot.json") {
-			t.Errorf("expected legacy path, got %s", paths[0])
-		}
-	})
-
-	t.Run("prefers new path over legacy", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-		newDir := filepath.Join(tmpDir, ".openclaw")
-		legacyDir := filepath.Join(tmpDir, ".clawdbot")
-		os.MkdirAll(newDir, 0o755)
-		os.MkdirAll(legacyDir, 0o755)
-		os.WriteFile(filepath.Join(newDir, "openclaw.json"), []byte(`{}`), 0o644)
-		os.WriteFile(filepath.Join(legacyDir, "clawdbot.json"), []byte(`{}`), 0o644)
-
-		paths := c.Paths()
-		if len(paths) != 1 {
-			t.Fatalf("expected 1 path, got %d", len(paths))
-		}
-		if paths[0] != filepath.Join(newDir, "openclaw.json") {
-			t.Errorf("expected new path, got %s", paths[0])
-		}
-	})
-
-	t.Run("Models reads from legacy path", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-		legacyDir := filepath.Join(tmpDir, ".clawdbot")
-		os.MkdirAll(legacyDir, 0o755)
-		os.WriteFile(filepath.Join(legacyDir, "clawdbot.json"), []byte(`{
-			"models":{"providers":{"ollama":{"models":[{"id":"llama3.2"}]}}}
-		}`), 0o644)
-
-		models := c.Models()
-		if len(models) != 1 || models[0] != "llama3.2" {
-			t.Errorf("expected [llama3.2], got %v", models)
-		}
-	})
-
-	t.Run("Models prefers new path over legacy", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-		newDir := filepath.Join(tmpDir, ".openclaw")
-		legacyDir := filepath.Join(tmpDir, ".clawdbot")
-		os.MkdirAll(newDir, 0o755)
-		os.MkdirAll(legacyDir, 0o755)
-		os.WriteFile(filepath.Join(newDir, "openclaw.json"), []byte(`{
-			"models":{"providers":{"ollama":{"models":[{"id":"new-model"}]}}}
-		}`), 0o644)
-		os.WriteFile(filepath.Join(legacyDir, "clawdbot.json"), []byte(`{
-			"models":{"providers":{"ollama":{"models":[{"id":"legacy-model"}]}}}
-		}`), 0o644)
-
-		models := c.Models()
-		if len(models) != 1 || models[0] != "new-model" {
-			t.Errorf("expected [new-model], got %v", models)
-		}
-	})
-
-	t.Run("Edit reads new path over legacy when both exist", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-		newDir := filepath.Join(tmpDir, ".openclaw")
-		legacyDir := filepath.Join(tmpDir, ".clawdbot")
-		os.MkdirAll(newDir, 0o755)
-		os.MkdirAll(legacyDir, 0o755)
-		os.WriteFile(filepath.Join(newDir, "openclaw.json"), []byte(`{"theme":"new"}`), 0o644)
-		os.WriteFile(filepath.Join(legacyDir, "clawdbot.json"), []byte(`{"theme":"legacy"}`), 0o644)
-
-		if err := c.Edit([]string{"llama3.2"}); err != nil {
-			t.Fatal(err)
-		}
-
-		data, _ := os.ReadFile(filepath.Join(newDir, "openclaw.json"))
-		var cfg map[string]any
-		json.Unmarshal(data, &cfg)
-		if cfg["theme"] != "new" {
-			t.Errorf("expected theme from new config, got %v", cfg["theme"])
-		}
-	})
-
-	t.Run("Edit migrates from legacy config", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-		legacyDir := filepath.Join(tmpDir, ".clawdbot")
-		os.MkdirAll(legacyDir, 0o755)
-		os.WriteFile(filepath.Join(legacyDir, "clawdbot.json"), []byte(`{"theme":"dark"}`), 0o644)
-
-		if err := c.Edit([]string{"llama3.2"}); err != nil {
-			t.Fatal(err)
-		}
-
-		// Should write to new path
-		newPath := filepath.Join(tmpDir, ".openclaw", "openclaw.json")
-		data, err := os.ReadFile(newPath)
-		if err != nil {
-			t.Fatal("expected new config file to be created")
-		}
-		var cfg map[string]any
-		json.Unmarshal(data, &cfg)
-		if cfg["theme"] != "dark" {
-			t.Error("legacy theme setting was not migrated")
-		}
-	})
-}
-
-func TestOpenclawEdit_CreatesDirectoryIfMissing(t *testing.T) {
-	c := &Openclaw{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-	configDir := filepath.Join(tmpDir, ".openclaw")
-
-	if _, err := os.Stat(configDir); !os.IsNotExist(err) {
-		t.Fatal("directory should not exist before test")
-	}
-
-	if err := c.Edit([]string{"model-a"}); err != nil {
-		t.Fatal(err)
-	}
-
-	if _, err := os.Stat(configDir); os.IsNotExist(err) {
-		t.Fatal("directory was not created")
-	}
-}
-
-func TestOpenclawOnboarded(t *testing.T) {
-	c := &Openclaw{}
-
-	t.Run("returns false when no config exists", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-		if c.onboarded() {
-			t.Error("expected false when no config exists")
-		}
-	})
-
-	t.Run("returns false when config exists but no wizard section", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-		configDir := filepath.Join(tmpDir, ".openclaw")
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(filepath.Join(configDir, "openclaw.json"), []byte(`{"theme":"dark"}`), 0o644)
-
-		if c.onboarded() {
-			t.Error("expected false when no wizard section")
-		}
-	})
-
-	t.Run("returns false when wizard section exists but no lastRunAt", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-		configDir := filepath.Join(tmpDir, ".openclaw")
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(filepath.Join(configDir, "openclaw.json"), []byte(`{"wizard":{}}`), 0o644)
-
-		if c.onboarded() {
-			t.Error("expected false when wizard.lastRunAt is missing")
-		}
-	})
-
-	t.Run("returns false when wizard.lastRunAt is empty string", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-		configDir := filepath.Join(tmpDir, ".openclaw")
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(filepath.Join(configDir, "openclaw.json"), []byte(`{"wizard":{"lastRunAt":""}}`), 0o644)
-
-		if c.onboarded() {
-			t.Error("expected false when wizard.lastRunAt is empty")
-		}
-	})
-
-	t.Run("returns true when wizard.lastRunAt is set", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-		configDir := filepath.Join(tmpDir, ".openclaw")
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(filepath.Join(configDir, "openclaw.json"), []byte(`{"wizard":{"lastRunAt":"2024-01-01T00:00:00Z"}}`), 0o644)
-
-		if !c.onboarded() {
-			t.Error("expected true when wizard.lastRunAt is set")
-		}
-	})
-
-	t.Run("checks legacy clawdbot path", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-		legacyDir := filepath.Join(tmpDir, ".clawdbot")
-		os.MkdirAll(legacyDir, 0o755)
-		os.WriteFile(filepath.Join(legacyDir, "clawdbot.json"), []byte(`{"wizard":{"lastRunAt":"2024-01-01T00:00:00Z"}}`), 0o644)
-
-		if !c.onboarded() {
-			t.Error("expected true when legacy config has wizard.lastRunAt")
-		}
-	})
-
-	t.Run("prefers new path over legacy", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-		newDir := filepath.Join(tmpDir, ".openclaw")
-		legacyDir := filepath.Join(tmpDir, ".clawdbot")
-		os.MkdirAll(newDir, 0o755)
-		os.MkdirAll(legacyDir, 0o755)
-		// New path has no wizard marker
-		os.WriteFile(filepath.Join(newDir, "openclaw.json"), []byte(`{}`), 0o644)
-		// Legacy has wizard marker
-		os.WriteFile(filepath.Join(legacyDir, "clawdbot.json"), []byte(`{"wizard":{"lastRunAt":"2024-01-01T00:00:00Z"}}`), 0o644)
-
-		if c.onboarded() {
-			t.Error("expected false - should prefer new path which has no wizard marker")
-		}
-	})
-
-	t.Run("handles corrupted JSON gracefully", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-		configDir := filepath.Join(tmpDir, ".openclaw")
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(filepath.Join(configDir, "openclaw.json"), []byte(`{corrupted`), 0o644)
-
-		if c.onboarded() {
-			t.Error("expected false for corrupted JSON")
-		}
-	})
-
-	t.Run("handles wrong type for wizard section", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-		configDir := filepath.Join(tmpDir, ".openclaw")
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(filepath.Join(configDir, "openclaw.json"), []byte(`{"wizard":"not a map"}`), 0o644)
-
-		if c.onboarded() {
-			t.Error("expected false when wizard is wrong type")
-		}
-	})
-}
--- a/cmd/config/opencode.go
+++ b/cmd/config/opencode.go
@@ -1,287 +0,0 @@
-package config
-
-import (
-	"context"
-	"encoding/json"
-	"fmt"
-	"maps"
-	"os"
-	"os/exec"
-	"path/filepath"
-	"slices"
-	"strings"
-
-	"github.com/ollama/ollama/api"
-	"github.com/ollama/ollama/envconfig"
-)
-
-// OpenCode implements Runner and Editor for OpenCode integration
-type OpenCode struct{}
-
-// cloudModelLimit holds context and output token limits for a cloud model.
-type cloudModelLimit struct {
-	Context int
-	Output  int
-}
-
-// cloudModelLimits maps cloud model base names to their token limits.
-// TODO(parthsareen): grab context/output limits from model info instead of hardcoding
-var cloudModelLimits = map[string]cloudModelLimit{
-	"cogito-2.1:671b":     {Context: 163_840, Output: 65_536},
-	"deepseek-v3.1:671b":  {Context: 163_840, Output: 163_840},
-	"deepseek-v3.2":       {Context: 163_840, Output: 65_536},
-	"glm-4.6":             {Context: 202_752, Output: 131_072},
-	"glm-4.7":             {Context: 202_752, Output: 131_072},
-	"gpt-oss:120b":        {Context: 131_072, Output: 131_072},
-	"gpt-oss:20b":         {Context: 131_072, Output: 131_072},
-	"kimi-k2:1t":          {Context: 262_144, Output: 262_144},
-	"kimi-k2.5":           {Context: 262_144, Output: 262_144},
-	"kimi-k2-thinking":    {Context: 262_144, Output: 262_144},
-	"nemotron-3-nano:30b": {Context: 1_048_576, Output: 131_072},
-	"qwen3-coder:480b":    {Context: 262_144, Output: 65_536},
-	"qwen3-coder-next":    {Context: 262_144, Output: 32_768},
-	"qwen3-next:80b":      {Context: 262_144, Output: 32_768},
-}
-
-// lookupCloudModelLimit returns the token limits for a cloud model.
-// It tries the exact name first, then strips the ":cloud" suffix.
-func lookupCloudModelLimit(name string) (cloudModelLimit, bool) {
-	if l, ok := cloudModelLimits[name]; ok {
-		return l, true
-	}
-	base := strings.TrimSuffix(name, ":cloud")
-	if base != name {
-		if l, ok := cloudModelLimits[base]; ok {
-			return l, true
-		}
-	}
-	return cloudModelLimit{}, false
-}
-
-func (o *OpenCode) String() string { return "OpenCode" }
-
-func (o *OpenCode) Run(model string, args []string) error {
-	if _, err := exec.LookPath("opencode"); err != nil {
-		return fmt.Errorf("opencode is not installed, install from https://opencode.ai")
-	}
-
-	// Call Edit() to ensure config is up-to-date before launch
-	models := []string{model}
-	if config, err := loadIntegration("opencode"); err == nil && len(config.Models) > 0 {
-		models = config.Models
-	}
-	if err := o.Edit(models); err != nil {
-		return fmt.Errorf("setup failed: %w", err)
-	}
-
-	cmd := exec.Command("opencode", args...)
-	cmd.Stdin = os.Stdin
-	cmd.Stdout = os.Stdout
-	cmd.Stderr = os.Stderr
-	return cmd.Run()
-}
-
-func (o *OpenCode) Paths() []string {
-	home, err := os.UserHomeDir()
-	if err != nil {
-		return nil
-	}
-
-	var paths []string
-	p := filepath.Join(home, ".config", "opencode", "opencode.json")
-	if _, err := os.Stat(p); err == nil {
-		paths = append(paths, p)
-	}
-	sp := filepath.Join(home, ".local", "state", "opencode", "model.json")
-	if _, err := os.Stat(sp); err == nil {
-		paths = append(paths, sp)
-	}
-	return paths
-}
-
-func (o *OpenCode) Edit(modelList []string) error {
-	if len(modelList) == 0 {
-		return nil
-	}
-
-	home, err := os.UserHomeDir()
-	if err != nil {
-		return err
-	}
-
-	configPath := filepath.Join(home, ".config", "opencode", "opencode.json")
-	if err := os.MkdirAll(filepath.Dir(configPath), 0o755); err != nil {
-		return err
-	}
-
-	config := make(map[string]any)
-	if data, err := os.ReadFile(configPath); err == nil {
-		_ = json.Unmarshal(data, &config) // Ignore parse errors; treat missing/corrupt files as empty
-	}
-
-	config["$schema"] = "https://opencode.ai/config.json"
-
-	provider, ok := config["provider"].(map[string]any)
-	if !ok {
-		provider = make(map[string]any)
-	}
-
-	ollama, ok := provider["ollama"].(map[string]any)
-	if !ok {
-		ollama = map[string]any{
-			"npm":  "@ai-sdk/openai-compatible",
-			"name": "Ollama (local)",
-			"options": map[string]any{
-				"baseURL": envconfig.Host().String() + "/v1",
-			},
-		}
-	}
-
-	models, ok := ollama["models"].(map[string]any)
-	if !ok {
-		models = make(map[string]any)
-	}
-
-	selectedSet := make(map[string]bool)
-	for _, m := range modelList {
-		selectedSet[m] = true
-	}
-
-	for name, cfg := range models {
-		if cfgMap, ok := cfg.(map[string]any); ok {
-			if isOllamaModel(cfgMap) && !selectedSet[name] {
-				delete(models, name)
-			}
-		}
-	}
-
-	client, _ := api.ClientFromEnvironment()
-
-	for _, model := range modelList {
-		if existing, ok := models[model].(map[string]any); ok {
-			// migrate existing models without _launch marker
-			if isOllamaModel(existing) {
-				existing["_launch"] = true
-				if name, ok := existing["name"].(string); ok {
-					existing["name"] = strings.TrimSuffix(name, " [Ollama]")
-				}
-			}
-			if isCloudModel(context.Background(), client, model) {
-				if l, ok := lookupCloudModelLimit(model); ok {
-					existing["limit"] = map[string]any{
-						"context": l.Context,
-						"output":  l.Output,
-					}
-				}
-			}
-			continue
-		}
-		entry := map[string]any{
-			"name":    model,
-			"_launch": true,
-		}
-		if isCloudModel(context.Background(), client, model) {
-			if l, ok := lookupCloudModelLimit(model); ok {
-				entry["limit"] = map[string]any{
-					"context": l.Context,
-					"output":  l.Output,
-				}
-			}
-		}
-		models[model] = entry
-	}
-
-	ollama["models"] = models
-	provider["ollama"] = ollama
-	config["provider"] = provider
-
-	configData, err := json.MarshalIndent(config, "", "  ")
-	if err != nil {
-		return err
-	}
-	if err := writeWithBackup(configPath, configData); err != nil {
-		return err
-	}
-
-	statePath := filepath.Join(home, ".local", "state", "opencode", "model.json")
-	if err := os.MkdirAll(filepath.Dir(statePath), 0o755); err != nil {
-		return err
-	}
-
-	state := map[string]any{
-		"recent":   []any{},
-		"favorite": []any{},
-		"variant":  map[string]any{},
-	}
-	if data, err := os.ReadFile(statePath); err == nil {
-		_ = json.Unmarshal(data, &state) // Ignore parse errors; use defaults
-	}
-
-	recent, _ := state["recent"].([]any)
-
-	modelSet := make(map[string]bool)
-	for _, m := range modelList {
-		modelSet[m] = true
-	}
-
-	// Filter out existing Ollama models we're about to re-add
-	newRecent := slices.DeleteFunc(slices.Clone(recent), func(entry any) bool {
-		e, ok := entry.(map[string]any)
-		if !ok || e["providerID"] != "ollama" {
-			return false
-		}
-		modelID, _ := e["modelID"].(string)
-		return modelSet[modelID]
-	})
-
-	// Prepend models in reverse order so first model ends up first
-	for _, model := range slices.Backward(modelList) {
-		newRecent = slices.Insert(newRecent, 0, any(map[string]any{
-			"providerID": "ollama",
-			"modelID":    model,
-		}))
-	}
-
-	const maxRecentModels = 10
-	newRecent = newRecent[:min(len(newRecent), maxRecentModels)]
-
-	state["recent"] = newRecent
-
-	stateData, err := json.MarshalIndent(state, "", "  ")
-	if err != nil {
-		return err
-	}
-	return writeWithBackup(statePath, stateData)
-}
-
-func (o *OpenCode) Models() []string {
-	home, err := os.UserHomeDir()
-	if err != nil {
-		return nil
-	}
-	config, err := readJSONFile(filepath.Join(home, ".config", "opencode", "opencode.json"))
-	if err != nil {
-		return nil
-	}
-	provider, _ := config["provider"].(map[string]any)
-	ollama, _ := provider["ollama"].(map[string]any)
-	models, _ := ollama["models"].(map[string]any)
-	if len(models) == 0 {
-		return nil
-	}
-	keys := slices.Collect(maps.Keys(models))
-	slices.Sort(keys)
-	return keys
-}
-
-// isOllamaModel reports whether a model config entry is managed by us
-func isOllamaModel(cfg map[string]any) bool {
-	if v, ok := cfg["_launch"].(bool); ok && v {
-		return true
-	}
-	// previously used [Ollama] as a suffix for the model managed by ollama launch
-	if name, ok := cfg["name"].(string); ok {
-		return strings.HasSuffix(name, "[Ollama]")
-	}
-	return false
-}
--- a/cmd/config/opencode_test.go
+++ b/cmd/config/opencode_test.go
@@ -1,668 +0,0 @@
-package config
-
-import (
-	"encoding/json"
-	"fmt"
-	"os"
-	"path/filepath"
-	"testing"
-)
-
-func TestOpenCodeIntegration(t *testing.T) {
-	o := &OpenCode{}
-
-	t.Run("String", func(t *testing.T) {
-		if got := o.String(); got != "OpenCode" {
-			t.Errorf("String() = %q, want %q", got, "OpenCode")
-		}
-	})
-
-	t.Run("implements Runner", func(t *testing.T) {
-		var _ Runner = o
-	})
-
-	t.Run("implements Editor", func(t *testing.T) {
-		var _ Editor = o
-	})
-}
-
-func TestOpenCodeEdit(t *testing.T) {
-	o := &OpenCode{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	configDir := filepath.Join(tmpDir, ".config", "opencode")
-	configPath := filepath.Join(configDir, "opencode.json")
-	stateDir := filepath.Join(tmpDir, ".local", "state", "opencode")
-	statePath := filepath.Join(stateDir, "model.json")
-
-	cleanup := func() {
-		os.RemoveAll(configDir)
-		os.RemoveAll(stateDir)
-	}
-
-	t.Run("fresh install", func(t *testing.T) {
-		cleanup()
-		if err := o.Edit([]string{"llama3.2"}); err != nil {
-			t.Fatal(err)
-		}
-		assertOpenCodeModelExists(t, configPath, "llama3.2")
-		assertOpenCodeRecentModel(t, statePath, 0, "ollama", "llama3.2")
-	})
-
-	t.Run("preserve other providers", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(configPath, []byte(`{"provider":{"anthropic":{"apiKey":"xxx"}}}`), 0o644)
-		if err := o.Edit([]string{"llama3.2"}); err != nil {
-			t.Fatal(err)
-		}
-		data, _ := os.ReadFile(configPath)
-		var cfg map[string]any
-		json.Unmarshal(data, &cfg)
-		provider := cfg["provider"].(map[string]any)
-		if provider["anthropic"] == nil {
-			t.Error("anthropic provider was removed")
-		}
-		assertOpenCodeModelExists(t, configPath, "llama3.2")
-	})
-
-	t.Run("preserve other models", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(configPath, []byte(`{"provider":{"ollama":{"models":{"mistral":{"name":"Mistral"}}}}}`), 0o644)
-		if err := o.Edit([]string{"llama3.2"}); err != nil {
-			t.Fatal(err)
-		}
-		assertOpenCodeModelExists(t, configPath, "mistral")
-		assertOpenCodeModelExists(t, configPath, "llama3.2")
-	})
-
-	t.Run("update existing model", func(t *testing.T) {
-		cleanup()
-		o.Edit([]string{"llama3.2"})
-		o.Edit([]string{"llama3.2"})
-		assertOpenCodeModelExists(t, configPath, "llama3.2")
-	})
-
-	t.Run("preserve top-level keys", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(configPath, []byte(`{"theme":"dark","keybindings":{}}`), 0o644)
-		if err := o.Edit([]string{"llama3.2"}); err != nil {
-			t.Fatal(err)
-		}
-		data, _ := os.ReadFile(configPath)
-		var cfg map[string]any
-		json.Unmarshal(data, &cfg)
-		if cfg["theme"] != "dark" {
-			t.Error("theme was removed")
-		}
-		if cfg["keybindings"] == nil {
-			t.Error("keybindings was removed")
-		}
-	})
-
-	t.Run("model state - insert at index 0", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(stateDir, 0o755)
-		os.WriteFile(statePath, []byte(`{"recent":[{"providerID":"anthropic","modelID":"claude"}],"favorite":[],"variant":{}}`), 0o644)
-		if err := o.Edit([]string{"llama3.2"}); err != nil {
-			t.Fatal(err)
-		}
-		assertOpenCodeRecentModel(t, statePath, 0, "ollama", "llama3.2")
-		assertOpenCodeRecentModel(t, statePath, 1, "anthropic", "claude")
-	})
-
-	t.Run("model state - preserve favorites and variants", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(stateDir, 0o755)
-		os.WriteFile(statePath, []byte(`{"recent":[],"favorite":[{"providerID":"x","modelID":"y"}],"variant":{"a":"b"}}`), 0o644)
-		if err := o.Edit([]string{"llama3.2"}); err != nil {
-			t.Fatal(err)
-		}
-		data, _ := os.ReadFile(statePath)
-		var state map[string]any
-		json.Unmarshal(data, &state)
-		if len(state["favorite"].([]any)) != 1 {
-			t.Error("favorite was modified")
-		}
-		if state["variant"].(map[string]any)["a"] != "b" {
-			t.Error("variant was modified")
-		}
-	})
-
-	t.Run("model state - deduplicate on re-add", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(stateDir, 0o755)
-		os.WriteFile(statePath, []byte(`{"recent":[{"providerID":"ollama","modelID":"llama3.2"},{"providerID":"anthropic","modelID":"claude"}],"favorite":[],"variant":{}}`), 0o644)
-		if err := o.Edit([]string{"llama3.2"}); err != nil {
-			t.Fatal(err)
-		}
-		data, _ := os.ReadFile(statePath)
-		var state map[string]any
-		json.Unmarshal(data, &state)
-		recent := state["recent"].([]any)
-		if len(recent) != 2 {
-			t.Errorf("expected 2 recent entries, got %d", len(recent))
-		}
-		assertOpenCodeRecentModel(t, statePath, 0, "ollama", "llama3.2")
-	})
-
-	t.Run("remove model", func(t *testing.T) {
-		cleanup()
-		// First add two models
-		o.Edit([]string{"llama3.2", "mistral"})
-		assertOpenCodeModelExists(t, configPath, "llama3.2")
-		assertOpenCodeModelExists(t, configPath, "mistral")
-
-		// Then remove one by only selecting the other
-		o.Edit([]string{"llama3.2"})
-		assertOpenCodeModelExists(t, configPath, "llama3.2")
-		assertOpenCodeModelNotExists(t, configPath, "mistral")
-	})
-
-	t.Run("preserve user customizations on managed models", func(t *testing.T) {
-		cleanup()
-		if err := o.Edit([]string{"llama3.2"}); err != nil {
-			t.Fatal(err)
-		}
-
-		// Add custom fields to the model entry (simulating user edits)
-		data, _ := os.ReadFile(configPath)
-		var cfg map[string]any
-		json.Unmarshal(data, &cfg)
-		provider := cfg["provider"].(map[string]any)
-		ollama := provider["ollama"].(map[string]any)
-		models := ollama["models"].(map[string]any)
-		entry := models["llama3.2"].(map[string]any)
-		entry["_myPref"] = "custom-value"
-		entry["_myNum"] = 42
-		configData, _ := json.MarshalIndent(cfg, "", "  ")
-		os.WriteFile(configPath, configData, 0o644)
-
-		// Re-run Edit — should preserve custom fields
-		if err := o.Edit([]string{"llama3.2"}); err != nil {
-			t.Fatal(err)
-		}
-
-		data, _ = os.ReadFile(configPath)
-		json.Unmarshal(data, &cfg)
-		provider = cfg["provider"].(map[string]any)
-		ollama = provider["ollama"].(map[string]any)
-		models = ollama["models"].(map[string]any)
-		entry = models["llama3.2"].(map[string]any)
-
-		if entry["_myPref"] != "custom-value" {
-			t.Errorf("_myPref was lost: got %v", entry["_myPref"])
-		}
-		if entry["_myNum"] != float64(42) {
-			t.Errorf("_myNum was lost: got %v", entry["_myNum"])
-		}
-		if v, ok := entry["_launch"].(bool); !ok || !v {
-			t.Errorf("_launch marker missing or false: got %v", entry["_launch"])
-		}
-	})
-
-	t.Run("migrate legacy [Ollama] suffix entries", func(t *testing.T) {
-		cleanup()
-		// Write a config with a legacy entry (has [Ollama] suffix but no _launch marker)
-		os.MkdirAll(configDir, 0o755)
-		os.WriteFile(configPath, []byte(`{"provider":{"ollama":{"models":{"llama3.2":{"name":"llama3.2 [Ollama]"}}}}}`), 0o644)
-
-		if err := o.Edit([]string{"llama3.2"}); err != nil {
-			t.Fatal(err)
-		}
-
-		data, _ := os.ReadFile(configPath)
-		var cfg map[string]any
-		json.Unmarshal(data, &cfg)
-		provider := cfg["provider"].(map[string]any)
-		ollama := provider["ollama"].(map[string]any)
-		models := ollama["models"].(map[string]any)
-		entry := models["llama3.2"].(map[string]any)
-
-		// _launch marker should be added
-		if v, ok := entry["_launch"].(bool); !ok || !v {
-			t.Errorf("_launch marker not added during migration: got %v", entry["_launch"])
-		}
-		// [Ollama] suffix should be stripped
-		if name, ok := entry["name"].(string); !ok || name != "llama3.2" {
-			t.Errorf("name suffix not stripped: got %q", entry["name"])
-		}
-	})
-
-	t.Run("remove model preserves non-ollama models", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-		// Add a non-Ollama model manually
-		os.WriteFile(configPath, []byte(`{"provider":{"ollama":{"models":{"external":{"name":"External Model"}}}}}`), 0o644)
-
-		o.Edit([]string{"llama3.2"})
-		assertOpenCodeModelExists(t, configPath, "llama3.2")
-		assertOpenCodeModelExists(t, configPath, "external") // Should be preserved
-	})
-}
-
-func assertOpenCodeModelExists(t *testing.T, path, model string) {
-	t.Helper()
-	data, err := os.ReadFile(path)
-	if err != nil {
-		t.Fatal(err)
-	}
-	var cfg map[string]any
-	if err := json.Unmarshal(data, &cfg); err != nil {
-		t.Fatal(err)
-	}
-	provider, ok := cfg["provider"].(map[string]any)
-	if !ok {
-		t.Fatal("provider not found")
-	}
-	ollama, ok := provider["ollama"].(map[string]any)
-	if !ok {
-		t.Fatal("ollama provider not found")
-	}
-	models, ok := ollama["models"].(map[string]any)
-	if !ok {
-		t.Fatal("models not found")
-	}
-	if models[model] == nil {
-		t.Errorf("model %s not found", model)
-	}
-}
-
-func assertOpenCodeModelNotExists(t *testing.T, path, model string) {
-	t.Helper()
-	data, err := os.ReadFile(path)
-	if err != nil {
-		t.Fatal(err)
-	}
-	var cfg map[string]any
-	if err := json.Unmarshal(data, &cfg); err != nil {
-		t.Fatal(err)
-	}
-	provider, ok := cfg["provider"].(map[string]any)
-	if !ok {
-		return // No provider means no model
-	}
-	ollama, ok := provider["ollama"].(map[string]any)
-	if !ok {
-		return // No ollama means no model
-	}
-	models, ok := ollama["models"].(map[string]any)
-	if !ok {
-		return // No models means no model
-	}
-	if models[model] != nil {
-		t.Errorf("model %s should not exist but was found", model)
-	}
-}
-
-func assertOpenCodeRecentModel(t *testing.T, path string, index int, providerID, modelID string) {
-	t.Helper()
-	data, err := os.ReadFile(path)
-	if err != nil {
-		t.Fatal(err)
-	}
-	var state map[string]any
-	if err := json.Unmarshal(data, &state); err != nil {
-		t.Fatal(err)
-	}
-	recent, ok := state["recent"].([]any)
-	if !ok {
-		t.Fatal("recent not found")
-	}
-	if index >= len(recent) {
-		t.Fatalf("index %d out of range (len=%d)", index, len(recent))
-	}
-	entry, ok := recent[index].(map[string]any)
-	if !ok {
-		t.Fatal("entry is not a map")
-	}
-	if entry["providerID"] != providerID {
-		t.Errorf("expected providerID %s, got %s", providerID, entry["providerID"])
-	}
-	if entry["modelID"] != modelID {
-		t.Errorf("expected modelID %s, got %s", modelID, entry["modelID"])
-	}
-}
-
-// Edge case tests for opencode.go
-
-func TestOpenCodeEdit_CorruptedConfigJSON(t *testing.T) {
-	o := &OpenCode{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	configDir := filepath.Join(tmpDir, ".config", "opencode")
-	configPath := filepath.Join(configDir, "opencode.json")
-
-	os.MkdirAll(configDir, 0o755)
-	os.WriteFile(configPath, []byte(`{corrupted json content`), 0o644)
-
-	// Should not panic - corrupted JSON should be treated as empty
-	err := o.Edit([]string{"llama3.2"})
-	if err != nil {
-		t.Fatalf("Edit failed with corrupted config: %v", err)
-	}
-
-	// Verify valid JSON was created
-	data, _ := os.ReadFile(configPath)
-	var cfg map[string]any
-	if err := json.Unmarshal(data, &cfg); err != nil {
-		t.Errorf("resulting config is not valid JSON: %v", err)
-	}
-}
-
-func TestOpenCodeEdit_CorruptedStateJSON(t *testing.T) {
-	o := &OpenCode{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	stateDir := filepath.Join(tmpDir, ".local", "state", "opencode")
-	statePath := filepath.Join(stateDir, "model.json")
-
-	os.MkdirAll(stateDir, 0o755)
-	os.WriteFile(statePath, []byte(`{corrupted state`), 0o644)
-
-	err := o.Edit([]string{"llama3.2"})
-	if err != nil {
-		t.Fatalf("Edit failed with corrupted state: %v", err)
-	}
-
-	// Verify valid state was created
-	data, _ := os.ReadFile(statePath)
-	var state map[string]any
-	if err := json.Unmarshal(data, &state); err != nil {
-		t.Errorf("resulting state is not valid JSON: %v", err)
-	}
-}
-
-func TestOpenCodeEdit_WrongTypeProvider(t *testing.T) {
-	o := &OpenCode{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	configDir := filepath.Join(tmpDir, ".config", "opencode")
-	configPath := filepath.Join(configDir, "opencode.json")
-
-	os.MkdirAll(configDir, 0o755)
-	os.WriteFile(configPath, []byte(`{"provider": "not a map"}`), 0o644)
-
-	err := o.Edit([]string{"llama3.2"})
-	if err != nil {
-		t.Fatalf("Edit with wrong type provider failed: %v", err)
-	}
-
-	// Verify provider is now correct type
-	data, _ := os.ReadFile(configPath)
-	var cfg map[string]any
-	json.Unmarshal(data, &cfg)
-
-	provider, ok := cfg["provider"].(map[string]any)
-	if !ok {
-		t.Fatalf("provider should be map after setup, got %T", cfg["provider"])
-	}
-	if provider["ollama"] == nil {
-		t.Error("ollama provider should be created")
-	}
-}
-
-func TestOpenCodeEdit_WrongTypeRecent(t *testing.T) {
-	o := &OpenCode{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	stateDir := filepath.Join(tmpDir, ".local", "state", "opencode")
-	statePath := filepath.Join(stateDir, "model.json")
-
-	os.MkdirAll(stateDir, 0o755)
-	os.WriteFile(statePath, []byte(`{"recent": "not an array", "favorite": [], "variant": {}}`), 0o644)
-
-	err := o.Edit([]string{"llama3.2"})
-	if err != nil {
-		t.Fatalf("Edit with wrong type recent failed: %v", err)
-	}
-
-	// The function should handle this gracefully
-	data, _ := os.ReadFile(statePath)
-	var state map[string]any
-	json.Unmarshal(data, &state)
-
-	// recent should be properly set after setup
-	recent, ok := state["recent"].([]any)
-	if !ok {
-		t.Logf("Note: recent type after setup is %T (documenting behavior)", state["recent"])
-	} else if len(recent) == 0 {
-		t.Logf("Note: recent is empty (documenting behavior)")
-	}
-}
-
-func TestOpenCodeEdit_EmptyModels(t *testing.T) {
-	o := &OpenCode{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	configDir := filepath.Join(tmpDir, ".config", "opencode")
-	configPath := filepath.Join(configDir, "opencode.json")
-
-	os.MkdirAll(configDir, 0o755)
-	originalContent := `{"provider":{"ollama":{"models":{"existing":{}}}}}`
-	os.WriteFile(configPath, []byte(originalContent), 0o644)
-
-	// Empty models should be no-op
-	err := o.Edit([]string{})
-	if err != nil {
-		t.Fatalf("Edit with empty models failed: %v", err)
-	}
-
-	// Original content should be preserved (file not modified)
-	data, _ := os.ReadFile(configPath)
-	if string(data) != originalContent {
-		t.Errorf("empty models should not modify file, but content changed")
-	}
-}
-
-func TestOpenCodeEdit_SpecialCharsInModelName(t *testing.T) {
-	o := &OpenCode{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	// Model name with special characters (though unusual)
-	specialModel := `model-with-"quotes"`
-
-	err := o.Edit([]string{specialModel})
-	if err != nil {
-		t.Fatalf("Edit with special chars failed: %v", err)
-	}
-
-	// Verify it was stored correctly
-	configDir := filepath.Join(tmpDir, ".config", "opencode")
-	configPath := filepath.Join(configDir, "opencode.json")
-	data, _ := os.ReadFile(configPath)
-
-	var cfg map[string]any
-	if err := json.Unmarshal(data, &cfg); err != nil {
-		t.Fatalf("resulting config is invalid JSON: %v", err)
-	}
-
-	// Model should be accessible
-	provider, _ := cfg["provider"].(map[string]any)
-	ollama, _ := provider["ollama"].(map[string]any)
-	models, _ := ollama["models"].(map[string]any)
-
-	if models[specialModel] == nil {
-		t.Errorf("model with special chars not found in config")
-	}
-}
-
-func readOpenCodeModel(t *testing.T, configPath, model string) map[string]any {
-	t.Helper()
-	data, err := os.ReadFile(configPath)
-	if err != nil {
-		t.Fatal(err)
-	}
-	var cfg map[string]any
-	json.Unmarshal(data, &cfg)
-	provider := cfg["provider"].(map[string]any)
-	ollama := provider["ollama"].(map[string]any)
-	models := ollama["models"].(map[string]any)
-	entry, ok := models[model].(map[string]any)
-	if !ok {
-		t.Fatalf("model %s not found in config", model)
-	}
-	return entry
-}
-
-func TestOpenCodeEdit_LocalModelNoLimit(t *testing.T) {
-	o := &OpenCode{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	configPath := filepath.Join(tmpDir, ".config", "opencode", "opencode.json")
-
-	if err := o.Edit([]string{"llama3.2"}); err != nil {
-		t.Fatal(err)
-	}
-
-	entry := readOpenCodeModel(t, configPath, "llama3.2")
-	if entry["limit"] != nil {
-		t.Errorf("local model should not have limit set, got %v", entry["limit"])
-	}
-}
-
-func TestOpenCodeEdit_PreservesUserLimit(t *testing.T) {
-	o := &OpenCode{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	configDir := filepath.Join(tmpDir, ".config", "opencode")
-	configPath := filepath.Join(configDir, "opencode.json")
-
-	// Set up a model with a user-configured limit
-	os.MkdirAll(configDir, 0o755)
-	os.WriteFile(configPath, []byte(`{
-		"provider": {
-			"ollama": {
-				"models": {
-					"llama3.2": {
-						"name": "llama3.2",
-						"_launch": true,
-						"limit": {"context": 8192, "output": 4096}
-					}
-				}
-			}
-		}
-	}`), 0o644)
-
-	// Re-edit should preserve the user's limit (not delete it)
-	if err := o.Edit([]string{"llama3.2"}); err != nil {
-		t.Fatal(err)
-	}
-
-	entry := readOpenCodeModel(t, configPath, "llama3.2")
-	limit, ok := entry["limit"].(map[string]any)
-	if !ok {
-		t.Fatal("user-configured limit was removed")
-	}
-	if limit["context"] != float64(8192) {
-		t.Errorf("context limit changed: got %v, want 8192", limit["context"])
-	}
-	if limit["output"] != float64(4096) {
-		t.Errorf("output limit changed: got %v, want 4096", limit["output"])
-	}
-}
-
-func TestOpenCodeEdit_CloudModelLimitStructure(t *testing.T) {
-	// Verify that when a cloud model entry has limits set (as Edit would do),
-	// the structure matches what opencode expects and re-edit preserves them.
-	o := &OpenCode{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	configDir := filepath.Join(tmpDir, ".config", "opencode")
-	configPath := filepath.Join(configDir, "opencode.json")
-
-	expected := cloudModelLimits["glm-4.7"]
-
-	// Simulate a cloud model that already has the limit set by a previous Edit
-	os.MkdirAll(configDir, 0o755)
-	os.WriteFile(configPath, []byte(fmt.Sprintf(`{
-		"provider": {
-			"ollama": {
-				"models": {
-					"glm-4.7:cloud": {
-						"name": "glm-4.7:cloud",
-						"_launch": true,
-						"limit": {"context": %d, "output": %d}
-					}
-				}
-			}
-		}
-	}`, expected.Context, expected.Output)), 0o644)
-
-	// Re-edit should preserve the cloud model limit
-	if err := o.Edit([]string{"glm-4.7:cloud"}); err != nil {
-		t.Fatal(err)
-	}
-
-	entry := readOpenCodeModel(t, configPath, "glm-4.7:cloud")
-	limit, ok := entry["limit"].(map[string]any)
-	if !ok {
-		t.Fatal("cloud model limit was removed on re-edit")
-	}
-	if limit["context"] != float64(expected.Context) {
-		t.Errorf("context = %v, want %d", limit["context"], expected.Context)
-	}
-	if limit["output"] != float64(expected.Output) {
-		t.Errorf("output = %v, want %d", limit["output"], expected.Output)
-	}
-}
-
-func TestLookupCloudModelLimit(t *testing.T) {
-	tests := []struct {
-		name        string
-		wantOK      bool
-		wantContext int
-		wantOutput  int
-	}{
-		{"glm-4.7", true, 202_752, 131_072},
-		{"glm-4.7:cloud", true, 202_752, 131_072},
-		{"kimi-k2.5", true, 262_144, 262_144},
-		{"kimi-k2.5:cloud", true, 262_144, 262_144},
-		{"deepseek-v3.2", true, 163_840, 65_536},
-		{"deepseek-v3.2:cloud", true, 163_840, 65_536},
-		{"qwen3-coder:480b", true, 262_144, 65_536},
-		{"qwen3-coder-next:cloud", true, 262_144, 32_768},
-		{"llama3.2", false, 0, 0},
-		{"unknown-model:cloud", false, 0, 0},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			l, ok := lookupCloudModelLimit(tt.name)
-			if ok != tt.wantOK {
-				t.Errorf("lookupCloudModelLimit(%q) ok = %v, want %v", tt.name, ok, tt.wantOK)
-			}
-			if ok {
-				if l.Context != tt.wantContext {
-					t.Errorf("context = %d, want %d", l.Context, tt.wantContext)
-				}
-				if l.Output != tt.wantOutput {
-					t.Errorf("output = %d, want %d", l.Output, tt.wantOutput)
-				}
-			}
-		})
-	}
-}
-
-func TestOpenCodeModels_NoConfig(t *testing.T) {
-	o := &OpenCode{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	models := o.Models()
-	if len(models) > 0 {
-		t.Errorf("expected nil/empty for missing config, got %v", models)
-	}
-}
--- a/cmd/config/pi.go
+++ b/cmd/config/pi.go
@@ -1,237 +0,0 @@
-package config
-
-import (
-	"context"
-	"encoding/json"
-	"fmt"
-	"net/http"
-	"os"
-	"os/exec"
-	"path/filepath"
-	"slices"
-	"strings"
-
-	"github.com/ollama/ollama/api"
-	"github.com/ollama/ollama/envconfig"
-	"github.com/ollama/ollama/types/model"
-)
-
-// Pi implements Runner and Editor for Pi (Pi Coding Agent) integration
-type Pi struct{}
-
-func (p *Pi) String() string { return "Pi" }
-
-func (p *Pi) Run(model string, args []string) error {
-	if _, err := exec.LookPath("pi"); err != nil {
-		return fmt.Errorf("pi is not installed, install with: npm install -g @mariozechner/pi-coding-agent")
-	}
-
-	// Call Edit() to ensure config is up-to-date before launch
-	models := []string{model}
-	if config, err := loadIntegration("pi"); err == nil && len(config.Models) > 0 {
-		models = config.Models
-	}
-	if err := p.Edit(models); err != nil {
-		return fmt.Errorf("setup failed: %w", err)
-	}
-
-	cmd := exec.Command("pi", args...)
-	cmd.Stdin = os.Stdin
-	cmd.Stdout = os.Stdout
-	cmd.Stderr = os.Stderr
-	return cmd.Run()
-}
-
-func (p *Pi) Paths() []string {
-	home, err := os.UserHomeDir()
-	if err != nil {
-		return nil
-	}
-
-	var paths []string
-	modelsPath := filepath.Join(home, ".pi", "agent", "models.json")
-	if _, err := os.Stat(modelsPath); err == nil {
-		paths = append(paths, modelsPath)
-	}
-	settingsPath := filepath.Join(home, ".pi", "agent", "settings.json")
-	if _, err := os.Stat(settingsPath); err == nil {
-		paths = append(paths, settingsPath)
-	}
-	return paths
-}
-
-func (p *Pi) Edit(models []string) error {
-	if len(models) == 0 {
-		return nil
-	}
-
-	home, err := os.UserHomeDir()
-	if err != nil {
-		return err
-	}
-
-	configPath := filepath.Join(home, ".pi", "agent", "models.json")
-	if err := os.MkdirAll(filepath.Dir(configPath), 0o755); err != nil {
-		return err
-	}
-
-	config := make(map[string]any)
-	if data, err := os.ReadFile(configPath); err == nil {
-		_ = json.Unmarshal(data, &config)
-	}
-
-	providers, ok := config["providers"].(map[string]any)
-	if !ok {
-		providers = make(map[string]any)
-	}
-
-	ollama, ok := providers["ollama"].(map[string]any)
-	if !ok {
-		ollama = map[string]any{
-			"baseUrl": envconfig.Host().String() + "/v1",
-			"api":     "openai-completions",
-			"apiKey":  "ollama",
-		}
-	}
-
-	existingModels, ok := ollama["models"].([]any)
-	if !ok {
-		existingModels = make([]any, 0)
-	}
-
-	// Build set of selected models to track which need to be added
-	selectedSet := make(map[string]bool, len(models))
-	for _, m := range models {
-		selectedSet[m] = true
-	}
-
-	// Build new models list:
-	// 1. Keep user-managed models (no _launch marker) - untouched
-	// 2. Keep ollama-managed models (_launch marker) that are still selected
-	// 3. Add new ollama-managed models
-	var newModels []any
-	for _, m := range existingModels {
-		if modelObj, ok := m.(map[string]any); ok {
-			if id, ok := modelObj["id"].(string); ok {
-				// User-managed model (no _launch marker) - always preserve
-				if !isPiOllamaModel(modelObj) {
-					newModels = append(newModels, m)
-				} else if selectedSet[id] {
-					// Ollama-managed and still selected - keep it
-					newModels = append(newModels, m)
-					selectedSet[id] = false
-				}
-			}
-		}
-	}
-
-	// Add newly selected models that weren't already in the list
-	client := api.NewClient(envconfig.Host(), http.DefaultClient)
-	ctx := context.Background()
-	for _, model := range models {
-		if selectedSet[model] {
-			newModels = append(newModels, createConfig(ctx, client, model))
-		}
-	}
-
-	ollama["models"] = newModels
-	providers["ollama"] = ollama
-	config["providers"] = providers
-
-	configData, err := json.MarshalIndent(config, "", "  ")
-	if err != nil {
-		return err
-	}
-	if err := writeWithBackup(configPath, configData); err != nil {
-		return err
-	}
-
-	// Update settings.json with default provider and model
-	settingsPath := filepath.Join(home, ".pi", "agent", "settings.json")
-	settings := make(map[string]any)
-	if data, err := os.ReadFile(settingsPath); err == nil {
-		_ = json.Unmarshal(data, &settings)
-	}
-
-	settings["defaultProvider"] = "ollama"
-	settings["defaultModel"] = models[0]
-
-	settingsData, err := json.MarshalIndent(settings, "", "  ")
-	if err != nil {
-		return err
-	}
-	return writeWithBackup(settingsPath, settingsData)
-}
-
-func (p *Pi) Models() []string {
-	home, err := os.UserHomeDir()
-	if err != nil {
-		return nil
-	}
-
-	configPath := filepath.Join(home, ".pi", "agent", "models.json")
-	config, err := readJSONFile(configPath)
-	if err != nil {
-		return nil
-	}
-
-	providers, _ := config["providers"].(map[string]any)
-	ollama, _ := providers["ollama"].(map[string]any)
-	models, _ := ollama["models"].([]any)
-
-	var result []string
-	for _, m := range models {
-		if modelObj, ok := m.(map[string]any); ok {
-			if id, ok := modelObj["id"].(string); ok {
-				result = append(result, id)
-			}
-		}
-	}
-	slices.Sort(result)
-	return result
-}
-
-// isPiOllamaModel reports whether a model config entry is managed by ollama launch
-func isPiOllamaModel(cfg map[string]any) bool {
-	if v, ok := cfg["_launch"].(bool); ok && v {
-		return true
-	}
-	return false
-}
-
-// createConfig builds Pi model config with capability detection
-func createConfig(ctx context.Context, client *api.Client, modelID string) map[string]any {
-	cfg := map[string]any{
-		"id":      modelID,
-		"_launch": true,
-	}
-
-	resp, err := client.Show(ctx, &api.ShowRequest{Model: modelID})
-	if err != nil {
-		return cfg
-	}
-
-	// Set input types based on vision capability
-	if slices.Contains(resp.Capabilities, model.CapabilityVision) {
-		cfg["input"] = []string{"text", "image"}
-	} else {
-		cfg["input"] = []string{"text"}
-	}
-
-	// Set reasoning based on thinking capability
-	if slices.Contains(resp.Capabilities, model.CapabilityThinking) {
-		cfg["reasoning"] = true
-	}
-
-	// Extract context window from ModelInfo
-	for key, val := range resp.ModelInfo {
-		if strings.HasSuffix(key, ".context_length") {
-			if ctxLen, ok := val.(float64); ok && ctxLen > 0 {
-				cfg["contextWindow"] = int(ctxLen)
-			}
-			break
-		}
-	}
-
-	return cfg
-}
--- a/cmd/config/pi_test.go
+++ b/cmd/config/pi_test.go
@@ -1,830 +0,0 @@
-package config
-
-import (
-	"context"
-	"encoding/json"
-	"fmt"
-	"net/http"
-	"net/http/httptest"
-	"net/url"
-	"os"
-	"path/filepath"
-	"testing"
-
-	"github.com/ollama/ollama/api"
-	"github.com/ollama/ollama/types/model"
-)
-
-func TestPiIntegration(t *testing.T) {
-	pi := &Pi{}
-
-	t.Run("String", func(t *testing.T) {
-		if got := pi.String(); got != "Pi" {
-			t.Errorf("String() = %q, want %q", got, "Pi")
-		}
-	})
-
-	t.Run("implements Runner", func(t *testing.T) {
-		var _ Runner = pi
-	})
-
-	t.Run("implements Editor", func(t *testing.T) {
-		var _ Editor = pi
-	})
-}
-
-func TestPiPaths(t *testing.T) {
-	pi := &Pi{}
-
-	t.Run("returns empty when no config exists", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		paths := pi.Paths()
-		if len(paths) != 0 {
-			t.Errorf("Paths() = %v, want empty", paths)
-		}
-	})
-
-	t.Run("returns path when config exists", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		configDir := filepath.Join(tmpDir, ".pi", "agent")
-		if err := os.MkdirAll(configDir, 0o755); err != nil {
-			t.Fatal(err)
-		}
-		configPath := filepath.Join(configDir, "models.json")
-		if err := os.WriteFile(configPath, []byte("{}"), 0o644); err != nil {
-			t.Fatal(err)
-		}
-
-		paths := pi.Paths()
-		if len(paths) != 1 || paths[0] != configPath {
-			t.Errorf("Paths() = %v, want [%s]", paths, configPath)
-		}
-	})
-}
-
-func TestPiEdit(t *testing.T) {
-	// Mock Ollama server for createConfig calls during Edit
-	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		if r.URL.Path == "/api/show" {
-			fmt.Fprintf(w, `{"capabilities":[],"model_info":{}}`)
-			return
-		}
-		w.WriteHeader(http.StatusNotFound)
-	}))
-	defer srv.Close()
-	t.Setenv("OLLAMA_HOST", srv.URL)
-
-	pi := &Pi{}
-	tmpDir := t.TempDir()
-	setTestHome(t, tmpDir)
-
-	configDir := filepath.Join(tmpDir, ".pi", "agent")
-	configPath := filepath.Join(configDir, "models.json")
-
-	cleanup := func() {
-		os.RemoveAll(configDir)
-	}
-
-	readConfig := func() map[string]any {
-		data, _ := os.ReadFile(configPath)
-		var cfg map[string]any
-		json.Unmarshal(data, &cfg)
-		return cfg
-	}
-
-	t.Run("returns nil for empty models", func(t *testing.T) {
-		if err := pi.Edit([]string{}); err != nil {
-			t.Errorf("Edit([]) error = %v, want nil", err)
-		}
-	})
-
-	t.Run("creates config with models", func(t *testing.T) {
-		cleanup()
-
-		models := []string{"llama3.2", "qwen3:8b"}
-		if err := pi.Edit(models); err != nil {
-			t.Fatalf("Edit() error = %v", err)
-		}
-
-		cfg := readConfig()
-
-		providers, ok := cfg["providers"].(map[string]any)
-		if !ok {
-			t.Error("Config missing providers")
-		}
-
-		ollama, ok := providers["ollama"].(map[string]any)
-		if !ok {
-			t.Error("Providers missing ollama")
-		}
-
-		modelsArray, ok := ollama["models"].([]any)
-		if !ok || len(modelsArray) != 2 {
-			t.Errorf("Expected 2 models, got %v", modelsArray)
-		}
-
-		if ollama["baseUrl"] == nil {
-			t.Error("Missing baseUrl")
-		}
-		if ollama["api"] != "openai-completions" {
-			t.Errorf("Expected api=openai-completions, got %v", ollama["api"])
-		}
-		if ollama["apiKey"] != "ollama" {
-			t.Errorf("Expected apiKey=ollama, got %v", ollama["apiKey"])
-		}
-	})
-
-	t.Run("updates existing config preserving ollama provider settings", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-
-		existingConfig := `{
-			"providers": {
-				"ollama": {
-					"baseUrl": "http://custom:8080/v1",
-					"api": "custom-api",
-					"apiKey": "custom-key",
-					"models": [
-						{"id": "old-model", "_launch": true}
-					]
-				}
-			}
-		}`
-		if err := os.WriteFile(configPath, []byte(existingConfig), 0o644); err != nil {
-			t.Fatal(err)
-		}
-
-		models := []string{"new-model"}
-		if err := pi.Edit(models); err != nil {
-			t.Fatalf("Edit() error = %v", err)
-		}
-
-		cfg := readConfig()
-		providers := cfg["providers"].(map[string]any)
-		ollama := providers["ollama"].(map[string]any)
-
-		if ollama["baseUrl"] != "http://custom:8080/v1" {
-			t.Errorf("Custom baseUrl not preserved, got %v", ollama["baseUrl"])
-		}
-		if ollama["api"] != "custom-api" {
-			t.Errorf("Custom api not preserved, got %v", ollama["api"])
-		}
-		if ollama["apiKey"] != "custom-key" {
-			t.Errorf("Custom apiKey not preserved, got %v", ollama["apiKey"])
-		}
-
-		modelsArray := ollama["models"].([]any)
-		if len(modelsArray) != 1 {
-			t.Errorf("Expected 1 model after update, got %d", len(modelsArray))
-		} else {
-			modelEntry := modelsArray[0].(map[string]any)
-			if modelEntry["id"] != "new-model" {
-				t.Errorf("Expected new-model, got %v", modelEntry["id"])
-			}
-			// Verify _launch marker is present
-			if modelEntry["_launch"] != true {
-				t.Errorf("Expected _launch marker to be true")
-			}
-		}
-	})
-
-	t.Run("replaces old models with new ones", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-
-		// Old models must have _launch marker to be managed by us
-		existingConfig := `{
-			"providers": {
-				"ollama": {
-					"baseUrl": "http://localhost:11434/v1",
-					"api": "openai-completions",
-					"apiKey": "ollama",
-					"models": [
-						{"id": "old-model-1", "_launch": true},
-						{"id": "old-model-2", "_launch": true}
-					]
-				}
-			}
-		}`
-		if err := os.WriteFile(configPath, []byte(existingConfig), 0o644); err != nil {
-			t.Fatal(err)
-		}
-
-		newModels := []string{"new-model-1", "new-model-2"}
-		if err := pi.Edit(newModels); err != nil {
-			t.Fatalf("Edit() error = %v", err)
-		}
-
-		cfg := readConfig()
-		providers := cfg["providers"].(map[string]any)
-		ollama := providers["ollama"].(map[string]any)
-		modelsArray := ollama["models"].([]any)
-
-		if len(modelsArray) != 2 {
-			t.Errorf("Expected 2 models, got %d", len(modelsArray))
-		}
-
-		modelIDs := make(map[string]bool)
-		for _, m := range modelsArray {
-			modelObj := m.(map[string]any)
-			id := modelObj["id"].(string)
-			modelIDs[id] = true
-		}
-
-		if !modelIDs["new-model-1"] || !modelIDs["new-model-2"] {
-			t.Errorf("Expected new models, got %v", modelIDs)
-		}
-		if modelIDs["old-model-1"] || modelIDs["old-model-2"] {
-			t.Errorf("Old models should have been removed, got %v", modelIDs)
-		}
-	})
-
-	t.Run("handles partial overlap in model list", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-
-		// Models must have _launch marker to be managed
-		existingConfig := `{
-			"providers": {
-				"ollama": {
-					"baseUrl": "http://localhost:11434/v1",
-					"api": "openai-completions",
-					"apiKey": "ollama",
-					"models": [
-						{"id": "keep-model", "_launch": true},
-						{"id": "remove-model", "_launch": true}
-					]
-				}
-			}
-		}`
-		if err := os.WriteFile(configPath, []byte(existingConfig), 0o644); err != nil {
-			t.Fatal(err)
-		}
-
-		newModels := []string{"keep-model", "add-model"}
-		if err := pi.Edit(newModels); err != nil {
-			t.Fatalf("Edit() error = %v", err)
-		}
-
-		cfg := readConfig()
-		providers := cfg["providers"].(map[string]any)
-		ollama := providers["ollama"].(map[string]any)
-		modelsArray := ollama["models"].([]any)
-
-		if len(modelsArray) != 2 {
-			t.Errorf("Expected 2 models, got %d", len(modelsArray))
-		}
-
-		modelIDs := make(map[string]bool)
-		for _, m := range modelsArray {
-			modelObj := m.(map[string]any)
-			id := modelObj["id"].(string)
-			modelIDs[id] = true
-		}
-
-		if !modelIDs["keep-model"] || !modelIDs["add-model"] {
-			t.Errorf("Expected keep-model and add-model, got %v", modelIDs)
-		}
-		if modelIDs["remove-model"] {
-			t.Errorf("remove-model should have been removed")
-		}
-	})
-
-	t.Run("handles corrupt config gracefully", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-
-		if err := os.WriteFile(configPath, []byte("{invalid json}"), 0o644); err != nil {
-			t.Fatal(err)
-		}
-
-		models := []string{"test-model"}
-		if err := pi.Edit(models); err != nil {
-			t.Fatalf("Edit() should not fail with corrupt config, got %v", err)
-		}
-
-		data, err := os.ReadFile(configPath)
-		if err != nil {
-			t.Fatalf("Failed to read config: %v", err)
-		}
-
-		var cfg map[string]any
-		if err := json.Unmarshal(data, &cfg); err != nil {
-			t.Fatalf("Config should be valid after Edit, got parse error: %v", err)
-		}
-
-		providers := cfg["providers"].(map[string]any)
-		ollama := providers["ollama"].(map[string]any)
-		modelsArray := ollama["models"].([]any)
-
-		if len(modelsArray) != 1 {
-			t.Errorf("Expected 1 model, got %d", len(modelsArray))
-		}
-	})
-
-	// CRITICAL SAFETY TEST: verifies we don't stomp on user configs
-	t.Run("preserves user-managed models without _launch marker", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-
-		// User has manually configured models in ollama provider (no _launch marker)
-		existingConfig := `{
-			"providers": {
-				"ollama": {
-					"baseUrl": "http://localhost:11434/v1",
-					"api": "openai-completions",
-					"apiKey": "ollama",
-					"models": [
-						{"id": "user-model-1"},
-						{"id": "user-model-2", "customField": "preserved"},
-						{"id": "ollama-managed", "_launch": true}
-					]
-				}
-			}
-		}`
-		if err := os.WriteFile(configPath, []byte(existingConfig), 0o644); err != nil {
-			t.Fatal(err)
-		}
-
-		// Add a new ollama-managed model
-		newModels := []string{"new-ollama-model"}
-		if err := pi.Edit(newModels); err != nil {
-			t.Fatalf("Edit() error = %v", err)
-		}
-
-		cfg := readConfig()
-		providers := cfg["providers"].(map[string]any)
-		ollama := providers["ollama"].(map[string]any)
-		modelsArray := ollama["models"].([]any)
-
-		// Should have: new-ollama-model (managed) + 2 user models (preserved)
-		if len(modelsArray) != 3 {
-			t.Errorf("Expected 3 models (1 new managed + 2 preserved user models), got %d", len(modelsArray))
-		}
-
-		modelIDs := make(map[string]map[string]any)
-		for _, m := range modelsArray {
-			modelObj := m.(map[string]any)
-			id := modelObj["id"].(string)
-			modelIDs[id] = modelObj
-		}
-
-		// Verify new model has _launch marker
-		if m, ok := modelIDs["new-ollama-model"]; !ok {
-			t.Errorf("new-ollama-model should be present")
-		} else if m["_launch"] != true {
-			t.Errorf("new-ollama-model should have _launch marker")
-		}
-
-		// Verify user models are preserved
-		if _, ok := modelIDs["user-model-1"]; !ok {
-			t.Errorf("user-model-1 should be preserved")
-		}
-		if _, ok := modelIDs["user-model-2"]; !ok {
-			t.Errorf("user-model-2 should be preserved")
-		} else if modelIDs["user-model-2"]["customField"] != "preserved" {
-			t.Errorf("user-model-2 customField should be preserved")
-		}
-
-		// Verify old ollama-managed model is removed (not in new list)
-		if _, ok := modelIDs["ollama-managed"]; ok {
-			t.Errorf("ollama-managed should be removed (old ollama model not in new selection)")
-		}
-	})
-
-	t.Run("updates settings.json with default provider and model", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-
-		// Create existing settings with other fields
-		settingsPath := filepath.Join(configDir, "settings.json")
-		existingSettings := `{
-			"theme": "dark",
-			"customSetting": "value",
-			"defaultProvider": "anthropic",
-			"defaultModel": "claude-3"
-		}`
-		if err := os.WriteFile(settingsPath, []byte(existingSettings), 0o644); err != nil {
-			t.Fatal(err)
-		}
-
-		models := []string{"llama3.2"}
-		if err := pi.Edit(models); err != nil {
-			t.Fatalf("Edit() error = %v", err)
-		}
-
-		data, err := os.ReadFile(settingsPath)
-		if err != nil {
-			t.Fatalf("Failed to read settings: %v", err)
-		}
-
-		var settings map[string]any
-		if err := json.Unmarshal(data, &settings); err != nil {
-			t.Fatalf("Failed to parse settings: %v", err)
-		}
-
-		// Verify defaultProvider is set to ollama
-		if settings["defaultProvider"] != "ollama" {
-			t.Errorf("defaultProvider = %v, want ollama", settings["defaultProvider"])
-		}
-
-		// Verify defaultModel is set to first model
-		if settings["defaultModel"] != "llama3.2" {
-			t.Errorf("defaultModel = %v, want llama3.2", settings["defaultModel"])
-		}
-
-		// Verify other fields are preserved
-		if settings["theme"] != "dark" {
-			t.Errorf("theme = %v, want dark (preserved)", settings["theme"])
-		}
-		if settings["customSetting"] != "value" {
-			t.Errorf("customSetting = %v, want value (preserved)", settings["customSetting"])
-		}
-	})
-
-	t.Run("creates settings.json if it does not exist", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-
-		models := []string{"qwen3:8b"}
-		if err := pi.Edit(models); err != nil {
-			t.Fatalf("Edit() error = %v", err)
-		}
-
-		settingsPath := filepath.Join(configDir, "settings.json")
-		data, err := os.ReadFile(settingsPath)
-		if err != nil {
-			t.Fatalf("settings.json should be created: %v", err)
-		}
-
-		var settings map[string]any
-		if err := json.Unmarshal(data, &settings); err != nil {
-			t.Fatalf("Failed to parse settings: %v", err)
-		}
-
-		if settings["defaultProvider"] != "ollama" {
-			t.Errorf("defaultProvider = %v, want ollama", settings["defaultProvider"])
-		}
-		if settings["defaultModel"] != "qwen3:8b" {
-			t.Errorf("defaultModel = %v, want qwen3:8b", settings["defaultModel"])
-		}
-	})
-
-	t.Run("handles corrupt settings.json gracefully", func(t *testing.T) {
-		cleanup()
-		os.MkdirAll(configDir, 0o755)
-
-		// Create corrupt settings
-		settingsPath := filepath.Join(configDir, "settings.json")
-		if err := os.WriteFile(settingsPath, []byte("{invalid"), 0o644); err != nil {
-			t.Fatal(err)
-		}
-
-		models := []string{"test-model"}
-		if err := pi.Edit(models); err != nil {
-			t.Fatalf("Edit() should not fail with corrupt settings, got %v", err)
-		}
-
-		data, err := os.ReadFile(settingsPath)
-		if err != nil {
-			t.Fatalf("Failed to read settings: %v", err)
-		}
-
-		var settings map[string]any
-		if err := json.Unmarshal(data, &settings); err != nil {
-			t.Fatalf("settings.json should be valid after Edit, got parse error: %v", err)
-		}
-
-		if settings["defaultProvider"] != "ollama" {
-			t.Errorf("defaultProvider = %v, want ollama", settings["defaultProvider"])
-		}
-		if settings["defaultModel"] != "test-model" {
-			t.Errorf("defaultModel = %v, want test-model", settings["defaultModel"])
-		}
-	})
-}
-
-func TestPiModels(t *testing.T) {
-	pi := &Pi{}
-
-	t.Run("returns nil when no config exists", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		models := pi.Models()
-		if models != nil {
-			t.Errorf("Models() = %v, want nil", models)
-		}
-	})
-
-	t.Run("returns models from config", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		configDir := filepath.Join(tmpDir, ".pi", "agent")
-		if err := os.MkdirAll(configDir, 0o755); err != nil {
-			t.Fatal(err)
-		}
-		config := `{
-			"providers": {
-				"ollama": {
-					"models": [
-						{"id": "llama3.2"},
-						{"id": "qwen3:8b"}
-					]
-				}
-			}
-		}`
-		configPath := filepath.Join(configDir, "models.json")
-		if err := os.WriteFile(configPath, []byte(config), 0o644); err != nil {
-			t.Fatal(err)
-		}
-
-		models := pi.Models()
-		if len(models) != 2 {
-			t.Errorf("Models() returned %d models, want 2", len(models))
-		}
-		if models[0] != "llama3.2" || models[1] != "qwen3:8b" {
-			t.Errorf("Models() = %v, want [llama3.2 qwen3:8b] (sorted)", models)
-		}
-	})
-
-	t.Run("returns sorted models", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		configDir := filepath.Join(tmpDir, ".pi", "agent")
-		if err := os.MkdirAll(configDir, 0o755); err != nil {
-			t.Fatal(err)
-		}
-		config := `{
-			"providers": {
-				"ollama": {
-					"models": [
-						{"id": "z-model"},
-						{"id": "a-model"},
-						{"id": "m-model"}
-					]
-				}
-			}
-		}`
-		configPath := filepath.Join(configDir, "models.json")
-		if err := os.WriteFile(configPath, []byte(config), 0o644); err != nil {
-			t.Fatal(err)
-		}
-
-		models := pi.Models()
-		if models[0] != "a-model" || models[1] != "m-model" || models[2] != "z-model" {
-			t.Errorf("Models() = %v, want [a-model m-model z-model] (sorted)", models)
-		}
-	})
-
-	t.Run("returns nil when models array is missing", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		configDir := filepath.Join(tmpDir, ".pi", "agent")
-		if err := os.MkdirAll(configDir, 0o755); err != nil {
-			t.Fatal(err)
-		}
-		config := `{
-			"providers": {
-				"ollama": {}
-			}
-		}`
-		configPath := filepath.Join(configDir, "models.json")
-		if err := os.WriteFile(configPath, []byte(config), 0o644); err != nil {
-			t.Fatal(err)
-		}
-
-		models := pi.Models()
-		if models != nil {
-			t.Errorf("Models() = %v, want nil when models array is missing", models)
-		}
-	})
-
-	t.Run("handles corrupt config gracefully", func(t *testing.T) {
-		tmpDir := t.TempDir()
-		setTestHome(t, tmpDir)
-
-		configDir := filepath.Join(tmpDir, ".pi", "agent")
-		if err := os.MkdirAll(configDir, 0o755); err != nil {
-			t.Fatal(err)
-		}
-		configPath := filepath.Join(configDir, "models.json")
-		if err := os.WriteFile(configPath, []byte("{invalid json}"), 0o644); err != nil {
-			t.Fatal(err)
-		}
-
-		models := pi.Models()
-		if models != nil {
-			t.Errorf("Models() = %v, want nil for corrupt config", models)
-		}
-	})
-}
-
-func TestIsPiOllamaModel(t *testing.T) {
-	tests := []struct {
-		name string
-		cfg  map[string]any
-		want bool
-	}{
-		{"with _launch true", map[string]any{"id": "m", "_launch": true}, true},
-		{"with _launch false", map[string]any{"id": "m", "_launch": false}, false},
-		{"without _launch", map[string]any{"id": "m"}, false},
-		{"with _launch non-bool", map[string]any{"id": "m", "_launch": "yes"}, false},
-		{"empty map", map[string]any{}, false},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			if got := isPiOllamaModel(tt.cfg); got != tt.want {
-				t.Errorf("isPiOllamaModel(%v) = %v, want %v", tt.cfg, got, tt.want)
-			}
-		})
-	}
-}
-
-func TestCreateConfig(t *testing.T) {
-	t.Run("sets vision input when model has vision capability", func(t *testing.T) {
-		srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-			if r.URL.Path == "/api/show" {
-				fmt.Fprintf(w, `{"capabilities":["vision"],"model_info":{}}`)
-				return
-			}
-			w.WriteHeader(http.StatusNotFound)
-		}))
-		defer srv.Close()
-
-		u, _ := url.Parse(srv.URL)
-		client := api.NewClient(u, srv.Client())
-
-		cfg := createConfig(context.Background(), client, "llava:7b")
-
-		if cfg["id"] != "llava:7b" {
-			t.Errorf("id = %v, want llava:7b", cfg["id"])
-		}
-		if cfg["_launch"] != true {
-			t.Error("expected _launch = true")
-		}
-		input, ok := cfg["input"].([]string)
-		if !ok || len(input) != 2 || input[0] != "text" || input[1] != "image" {
-			t.Errorf("input = %v, want [text image]", cfg["input"])
-		}
-	})
-
-	t.Run("sets text-only input when model lacks vision", func(t *testing.T) {
-		srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-			if r.URL.Path == "/api/show" {
-				fmt.Fprintf(w, `{"capabilities":["completion"],"model_info":{}}`)
-				return
-			}
-			w.WriteHeader(http.StatusNotFound)
-		}))
-		defer srv.Close()
-
-		u, _ := url.Parse(srv.URL)
-		client := api.NewClient(u, srv.Client())
-
-		cfg := createConfig(context.Background(), client, "llama3.2")
-
-		input, ok := cfg["input"].([]string)
-		if !ok || len(input) != 1 || input[0] != "text" {
-			t.Errorf("input = %v, want [text]", cfg["input"])
-		}
-		if _, ok := cfg["reasoning"]; ok {
-			t.Error("reasoning should not be set for non-thinking model")
-		}
-	})
-
-	t.Run("sets reasoning when model has thinking capability", func(t *testing.T) {
-		srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-			if r.URL.Path == "/api/show" {
-				fmt.Fprintf(w, `{"capabilities":["thinking"],"model_info":{}}`)
-				return
-			}
-			w.WriteHeader(http.StatusNotFound)
-		}))
-		defer srv.Close()
-
-		u, _ := url.Parse(srv.URL)
-		client := api.NewClient(u, srv.Client())
-
-		cfg := createConfig(context.Background(), client, "qwq")
-
-		if cfg["reasoning"] != true {
-			t.Error("expected reasoning = true for thinking model")
-		}
-	})
-
-	t.Run("extracts context window from model info", func(t *testing.T) {
-		srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-			if r.URL.Path == "/api/show" {
-				fmt.Fprintf(w, `{"capabilities":[],"model_info":{"llama.context_length":131072}}`)
-				return
-			}
-			w.WriteHeader(http.StatusNotFound)
-		}))
-		defer srv.Close()
-
-		u, _ := url.Parse(srv.URL)
-		client := api.NewClient(u, srv.Client())
-
-		cfg := createConfig(context.Background(), client, "llama3.2")
-
-		if cfg["contextWindow"] != 131072 {
-			t.Errorf("contextWindow = %v, want 131072", cfg["contextWindow"])
-		}
-	})
-
-	t.Run("handles all capabilities together", func(t *testing.T) {
-		srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-			if r.URL.Path == "/api/show" {
-				fmt.Fprintf(w, `{"capabilities":["vision","thinking"],"model_info":{"qwen3.context_length":32768}}`)
-				return
-			}
-			w.WriteHeader(http.StatusNotFound)
-		}))
-		defer srv.Close()
-
-		u, _ := url.Parse(srv.URL)
-		client := api.NewClient(u, srv.Client())
-
-		cfg := createConfig(context.Background(), client, "qwen3-vision")
-
-		input := cfg["input"].([]string)
-		if len(input) != 2 || input[0] != "text" || input[1] != "image" {
-			t.Errorf("input = %v, want [text image]", input)
-		}
-		if cfg["reasoning"] != true {
-			t.Error("expected reasoning = true")
-		}
-		if cfg["contextWindow"] != 32768 {
-			t.Errorf("contextWindow = %v, want 32768", cfg["contextWindow"])
-		}
-	})
-
-	t.Run("returns minimal config when show fails", func(t *testing.T) {
-		srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-			w.WriteHeader(http.StatusNotFound)
-			fmt.Fprintf(w, `{"error":"model not found"}`)
-		}))
-		defer srv.Close()
-
-		u, _ := url.Parse(srv.URL)
-		client := api.NewClient(u, srv.Client())
-
-		cfg := createConfig(context.Background(), client, "missing-model")
-
-		if cfg["id"] != "missing-model" {
-			t.Errorf("id = %v, want missing-model", cfg["id"])
-		}
-		if cfg["_launch"] != true {
-			t.Error("expected _launch = true")
-		}
-		// Should not have capability fields
-		if _, ok := cfg["input"]; ok {
-			t.Error("input should not be set when show fails")
-		}
-		if _, ok := cfg["reasoning"]; ok {
-			t.Error("reasoning should not be set when show fails")
-		}
-		if _, ok := cfg["contextWindow"]; ok {
-			t.Error("contextWindow should not be set when show fails")
-		}
-	})
-
-	t.Run("skips zero context length", func(t *testing.T) {
-		srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-			if r.URL.Path == "/api/show" {
-				fmt.Fprintf(w, `{"capabilities":[],"model_info":{"llama.context_length":0}}`)
-				return
-			}
-			w.WriteHeader(http.StatusNotFound)
-		}))
-		defer srv.Close()
-
-		u, _ := url.Parse(srv.URL)
-		client := api.NewClient(u, srv.Client())
-
-		cfg := createConfig(context.Background(), client, "test-model")
-
-		if _, ok := cfg["contextWindow"]; ok {
-			t.Error("contextWindow should not be set for zero value")
-		}
-	})
-}
-
-// Ensure Capability constants used in createConfig match expected values
-func TestPiCapabilityConstants(t *testing.T) {
-	if model.CapabilityVision != "vision" {
-		t.Errorf("CapabilityVision = %q, want %q", model.CapabilityVision, "vision")
-	}
-	if model.CapabilityThinking != "thinking" {
-		t.Errorf("CapabilityThinking = %q, want %q", model.CapabilityThinking, "thinking")
-	}
-}
--- a/cmd/config/selector.go
+++ b/cmd/config/selector.go
@@ -1,58 +0,0 @@
-package config
-
-import (
-	"errors"
-	"fmt"
-	"os"
-
-	"golang.org/x/term"
-)
-
-// ANSI escape sequences for terminal formatting.
-const (
-	ansiBold  = "\033[1m"
-	ansiReset = "\033[0m"
-	ansiGray  = "\033[37m"
-	ansiGreen = "\033[32m"
-)
-
-// ErrCancelled is returned when the user cancels a selection.
-var ErrCancelled = errors.New("cancelled")
-
-// errCancelled is kept as an alias for backward compatibility within the package.
-var errCancelled = ErrCancelled
-
-// DefaultConfirmPrompt provides a TUI-based confirmation prompt.
-// When set, confirmPrompt delegates to it instead of using raw terminal I/O.
-var DefaultConfirmPrompt func(prompt string) (bool, error)
-
-func confirmPrompt(prompt string) (bool, error) {
-	if DefaultConfirmPrompt != nil {
-		return DefaultConfirmPrompt(prompt)
-	}
-
-	fd := int(os.Stdin.Fd())
-	oldState, err := term.MakeRaw(fd)
-	if err != nil {
-		return false, err
-	}
-	defer term.Restore(fd, oldState)
-
-	fmt.Fprintf(os.Stderr, "%s (\033[1my\033[0m/n) ", prompt)
-
-	buf := make([]byte, 1)
-	for {
-		if _, err := os.Stdin.Read(buf); err != nil {
-			return false, err
-		}
-
-		switch buf[0] {
-		case 'Y', 'y', 13:
-			fmt.Fprintf(os.Stderr, "yes\r\n")
-			return true, nil
-		case 'N', 'n', 27, 3:
-			fmt.Fprintf(os.Stderr, "no\r\n")
-			return false, nil
-		}
-	}
-}
--- a/cmd/config/selector_test.go
+++ b/cmd/config/selector_test.go
@@ -1,19 +0,0 @@
-package config
-
-import (
-	"testing"
-)
-
-func TestErrCancelled(t *testing.T) {
-	t.Run("NotNil", func(t *testing.T) {
-		if errCancelled == nil {
-			t.Error("errCancelled should not be nil")
-		}
-	})
-
-	t.Run("Message", func(t *testing.T) {
-		if errCancelled.Error() != "cancelled" {
-			t.Errorf("expected 'cancelled', got %q", errCancelled.Error())
-		}
-	})
-}
--- a/cmd/interactive.go
+++ b/cmd/interactive.go
@@ -40,7 +40,6 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 		fmt.Fprintln(os.Stderr, "  /bye            Exit")
 		fmt.Fprintln(os.Stderr, "  /?, /help       Help for a command")
 		fmt.Fprintln(os.Stderr, "  /? shortcuts    Help for keyboard shortcuts")
-
 		fmt.Fprintln(os.Stderr, "")
 		fmt.Fprintln(os.Stderr, "Use \"\"\" to begin a multi-line message.")

@@ -116,7 +115,7 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 		Prompt:         ">>> ",
 		AltPrompt:      "... ",
 		Placeholder:    "Send a message (/? for help)",
-		AltPlaceholder: "Press Enter to send",
+		AltPlaceholder: `Use """ to end multi-line input`,
 	})
 	if err != nil {
 		return err
@@ -159,7 +158,6 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 			sb.WriteString(before)
 			if !ok {
 				fmt.Fprintln(&sb)
-				scanner.Prompt.UseAlt = true
 				continue
 			}

--- a/cmd/start_darwin.go
+++ b/cmd/start_darwin.go
@@ -10,21 +10,19 @@ import (
 	"github.com/ollama/ollama/api"
 )

-var errNotRunning = errors.New("could not connect to ollama server, run 'ollama serve' to start it")
-
 func startApp(ctx context.Context, client *api.Client) error {
 	exe, err := os.Executable()
 	if err != nil {
-		return errNotRunning
+		return err
 	}
 	link, err := os.Readlink(exe)
 	if err != nil {
-		return errNotRunning
+		return err
 	}
 	r := regexp.MustCompile(`^.*/Ollama\s?\d*.app`)
 	m := r.FindStringSubmatch(link)
 	if len(m) != 1 {
-		return errNotRunning
+		return errors.New("could not find ollama app")
 	}
 	if err := exec.Command("/usr/bin/open", "-j", "-a", m[0], "--args", "--fast-startup").Run(); err != nil {
 		return err
--- a/cmd/tui/confirm.go
+++ b/cmd/tui/confirm.go
@@ -1,109 +0,0 @@
-package tui
-
-import (
-	"fmt"
-
-	tea "github.com/charmbracelet/bubbletea"
-	"github.com/charmbracelet/lipgloss"
-)
-
-var (
-	confirmActiveStyle = lipgloss.NewStyle().
-				Bold(true).
-				Background(lipgloss.AdaptiveColor{Light: "254", Dark: "236"})
-
-	confirmInactiveStyle = lipgloss.NewStyle().
-				Foreground(lipgloss.AdaptiveColor{Light: "242", Dark: "246"})
-)
-
-type confirmModel struct {
-	prompt    string
-	yes       bool
-	confirmed bool
-	cancelled bool
-	width     int
-}
-
-func (m confirmModel) Init() tea.Cmd {
-	return nil
-}
-
-func (m confirmModel) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
-	switch msg := msg.(type) {
-	case tea.WindowSizeMsg:
-		wasSet := m.width > 0
-		m.width = msg.Width
-		if wasSet {
-			return m, tea.EnterAltScreen
-		}
-		return m, nil
-
-	case tea.KeyMsg:
-		switch msg.String() {
-		case "ctrl+c", "esc", "n":
-			m.cancelled = true
-			return m, tea.Quit
-		case "y":
-			m.yes = true
-			m.confirmed = true
-			return m, tea.Quit
-		case "enter":
-			m.confirmed = true
-			return m, tea.Quit
-		case "left", "h":
-			m.yes = true
-		case "right", "l":
-			m.yes = false
-		case "tab":
-			m.yes = !m.yes
-		}
-	}
-
-	return m, nil
-}
-
-func (m confirmModel) View() string {
-	if m.confirmed || m.cancelled {
-		return ""
-	}
-
-	var yesBtn, noBtn string
-	if m.yes {
-		yesBtn = confirmActiveStyle.Render(" Yes ")
-		noBtn = confirmInactiveStyle.Render(" No ")
-	} else {
-		yesBtn = confirmInactiveStyle.Render(" Yes ")
-		noBtn = confirmActiveStyle.Render(" No ")
-	}
-
-	s := selectorTitleStyle.Render(m.prompt) + "\n\n"
-	s += "  " + yesBtn + "  " + noBtn + "\n\n"
-	s += selectorHelpStyle.Render("←/→ navigate • enter confirm • esc cancel")
-
-	if m.width > 0 {
-		return lipgloss.NewStyle().MaxWidth(m.width).Render(s)
-	}
-	return s
-}
-
-// RunConfirm shows a bubbletea yes/no confirmation prompt.
-// Returns true if the user confirmed, false if cancelled.
-func RunConfirm(prompt string) (bool, error) {
-	m := confirmModel{
-		prompt: prompt,
-		yes:    true, // default to yes
-	}
-
-	p := tea.NewProgram(m)
-	finalModel, err := p.Run()
-	if err != nil {
-		return false, fmt.Errorf("error running confirm: %w", err)
-	}
-
-	fm := finalModel.(confirmModel)
-	if fm.cancelled {
-		return false, ErrCancelled
-	}
-
-	return fm.yes, nil
-}
--- a/cmd/tui/confirm_test.go
+++ b/cmd/tui/confirm_test.go
@@ -1,208 +0,0 @@
-package tui
-
-import (
-	"strings"
-	"testing"
-
-	tea "github.com/charmbracelet/bubbletea"
-)
-
-func TestConfirmModel_DefaultsToYes(t *testing.T) {
-	m := confirmModel{prompt: "Download test?", yes: true}
-	if !m.yes {
-		t.Error("should default to yes")
-	}
-}
-
-func TestConfirmModel_View_ContainsPrompt(t *testing.T) {
-	m := confirmModel{prompt: "Download qwen3:8b?", yes: true}
-	got := m.View()
-	if !strings.Contains(got, "Download qwen3:8b?") {
-		t.Error("should contain the prompt text")
-	}
-}
-
-func TestConfirmModel_View_ContainsButtons(t *testing.T) {
-	m := confirmModel{prompt: "Download?", yes: true}
-	got := m.View()
-	if !strings.Contains(got, "Yes") {
-		t.Error("should contain Yes button")
-	}
-	if !strings.Contains(got, "No") {
-		t.Error("should contain No button")
-	}
-}
-
-func TestConfirmModel_View_ContainsHelp(t *testing.T) {
-	m := confirmModel{prompt: "Download?", yes: true}
-	got := m.View()
-	if !strings.Contains(got, "enter confirm") {
-		t.Error("should contain help text")
-	}
-}
-
-func TestConfirmModel_View_ClearsAfterConfirm(t *testing.T) {
-	m := confirmModel{prompt: "Download?", confirmed: true}
-	if m.View() != "" {
-		t.Error("View should return empty string after confirmation")
-	}
-}
-
-func TestConfirmModel_View_ClearsAfterCancel(t *testing.T) {
-	m := confirmModel{prompt: "Download?", cancelled: true}
-	if m.View() != "" {
-		t.Error("View should return empty string after cancellation")
-	}
-}
-
-func TestConfirmModel_EnterConfirmsYes(t *testing.T) {
-	m := confirmModel{prompt: "Download?", yes: true}
-	updated, cmd := m.Update(tea.KeyMsg{Type: tea.KeyEnter})
-	fm := updated.(confirmModel)
-	if !fm.confirmed {
-		t.Error("enter should set confirmed=true")
-	}
-	if !fm.yes {
-		t.Error("enter with yes selected should keep yes=true")
-	}
-	if cmd == nil {
-		t.Error("enter should return tea.Quit")
-	}
-}
-
-func TestConfirmModel_EnterConfirmsNo(t *testing.T) {
-	m := confirmModel{prompt: "Download?", yes: false}
-	updated, cmd := m.Update(tea.KeyMsg{Type: tea.KeyEnter})
-	fm := updated.(confirmModel)
-	if !fm.confirmed {
-		t.Error("enter should set confirmed=true")
-	}
-	if fm.yes {
-		t.Error("enter with no selected should keep yes=false")
-	}
-	if cmd == nil {
-		t.Error("enter should return tea.Quit")
-	}
-}
-
-func TestConfirmModel_EscCancels(t *testing.T) {
-	m := confirmModel{prompt: "Download?", yes: true}
-	updated, cmd := m.Update(tea.KeyMsg{Type: tea.KeyEsc})
-	fm := updated.(confirmModel)
-	if !fm.cancelled {
-		t.Error("esc should set cancelled=true")
-	}
-	if cmd == nil {
-		t.Error("esc should return tea.Quit")
-	}
-}
-
-func TestConfirmModel_CtrlCCancels(t *testing.T) {
-	m := confirmModel{prompt: "Download?", yes: true}
-	updated, cmd := m.Update(tea.KeyMsg{Type: tea.KeyCtrlC})
-	fm := updated.(confirmModel)
-	if !fm.cancelled {
-		t.Error("ctrl+c should set cancelled=true")
-	}
-	if cmd == nil {
-		t.Error("ctrl+c should return tea.Quit")
-	}
-}
-
-func TestConfirmModel_NCancels(t *testing.T) {
-	m := confirmModel{prompt: "Download?", yes: true}
-	updated, cmd := m.Update(tea.KeyMsg{Type: tea.KeyRunes, Runes: []rune{'n'}})
-	fm := updated.(confirmModel)
-	if !fm.cancelled {
-		t.Error("'n' should set cancelled=true")
-	}
-	if cmd == nil {
-		t.Error("'n' should return tea.Quit")
-	}
-}
-
-func TestConfirmModel_YConfirmsYes(t *testing.T) {
-	m := confirmModel{prompt: "Download?", yes: false}
-	updated, cmd := m.Update(tea.KeyMsg{Type: tea.KeyRunes, Runes: []rune{'y'}})
-	fm := updated.(confirmModel)
-	if !fm.confirmed {
-		t.Error("'y' should set confirmed=true")
-	}
-	if !fm.yes {
-		t.Error("'y' should set yes=true")
-	}
-	if cmd == nil {
-		t.Error("'y' should return tea.Quit")
-	}
-}
-
-func TestConfirmModel_ArrowKeysNavigate(t *testing.T) {
-	m := confirmModel{prompt: "Download?", yes: true}
-
-	// Right moves to No
-	updated, _ := m.Update(tea.KeyMsg{Type: tea.KeyRunes, Runes: []rune{'l'}})
-	fm := updated.(confirmModel)
-	if fm.yes {
-		t.Error("right/l should move to No")
-	}
-	if fm.confirmed || fm.cancelled {
-		t.Error("navigation should not confirm or cancel")
-	}
-
-	// Left moves back to Yes
-	updated, _ = fm.Update(tea.KeyMsg{Type: tea.KeyRunes, Runes: []rune{'h'}})
-	fm = updated.(confirmModel)
-	if !fm.yes {
-		t.Error("left/h should move to Yes")
-	}
-}
-
-func TestConfirmModel_TabToggles(t *testing.T) {
-	m := confirmModel{prompt: "Download?", yes: true}
-
-	updated, _ := m.Update(tea.KeyMsg{Type: tea.KeyTab})
-	fm := updated.(confirmModel)
-	if fm.yes {
-		t.Error("tab should toggle from Yes to No")
-	}
-
-	updated, _ = fm.Update(tea.KeyMsg{Type: tea.KeyTab})
-	fm = updated.(confirmModel)
-	if !fm.yes {
-		t.Error("tab should toggle from No to Yes")
-	}
-}
-
-func TestConfirmModel_WindowSizeUpdatesWidth(t *testing.T) {
-	m := confirmModel{prompt: "Download?"}
-	updated, _ := m.Update(tea.WindowSizeMsg{Width: 100, Height: 40})
-	fm := updated.(confirmModel)
-	if fm.width != 100 {
-		t.Errorf("expected width 100, got %d", fm.width)
-	}
-}
-
-func TestConfirmModel_ResizeEntersAltScreen(t *testing.T) {
-	m := confirmModel{prompt: "Download?", width: 80}
-	_, cmd := m.Update(tea.WindowSizeMsg{Width: 100, Height: 40})
-	if cmd == nil {
-		t.Error("resize (width already set) should return a command")
-	}
-}
-
-func TestConfirmModel_InitialWindowSizeNoAltScreen(t *testing.T) {
-	m := confirmModel{prompt: "Download?"}
-	_, cmd := m.Update(tea.WindowSizeMsg{Width: 80, Height: 40})
-	if cmd != nil {
-		t.Error("initial WindowSizeMsg should not return a command")
-	}
-}
-
-func TestConfirmModel_ViewMaxWidth(t *testing.T) {
-	m := confirmModel{prompt: "Download?", yes: true, width: 40}
-	got := m.View()
-	// Just ensure it doesn't panic and returns content
-	if got == "" {
-		t.Error("View with width set should still return content")
-	}
-}
--- a/cmd/tui/selector.go
+++ b/cmd/tui/selector.go
@@ -1,654 +0,0 @@
-package tui
-
-import (
-	"errors"
-	"fmt"
-	"strings"
-
-	tea "github.com/charmbracelet/bubbletea"
-	"github.com/charmbracelet/lipgloss"
-)
-
-var (
-	selectorTitleStyle = lipgloss.NewStyle().
-				Bold(true)
-
-	selectorItemStyle = lipgloss.NewStyle().
-				PaddingLeft(4)
-
-	selectorSelectedItemStyle = lipgloss.NewStyle().
-					PaddingLeft(2).
-					Bold(true).
-					Background(lipgloss.AdaptiveColor{Light: "254", Dark: "236"})
-
-	selectorDescStyle = lipgloss.NewStyle().
-				Foreground(lipgloss.AdaptiveColor{Light: "242", Dark: "246"})
-
-	selectorDescLineStyle = selectorDescStyle.
-				PaddingLeft(6)
-
-	selectorFilterStyle = lipgloss.NewStyle().
-				Foreground(lipgloss.AdaptiveColor{Light: "242", Dark: "246"}).
-				Italic(true)
-
-	selectorInputStyle = lipgloss.NewStyle().
-				Foreground(lipgloss.AdaptiveColor{Light: "235", Dark: "252"})
-
-	selectorCheckboxStyle = lipgloss.NewStyle().
-				Foreground(lipgloss.AdaptiveColor{Light: "242", Dark: "246"})
-
-	selectorCheckboxCheckedStyle = lipgloss.NewStyle().
-					Bold(true)
-
-	selectorDefaultTagStyle = lipgloss.NewStyle().
-				Foreground(lipgloss.AdaptiveColor{Light: "242", Dark: "246"}).
-				Italic(true)
-
-	selectorHelpStyle = lipgloss.NewStyle().
-				Foreground(lipgloss.AdaptiveColor{Light: "244", Dark: "244"})
-
-	selectorMoreStyle = lipgloss.NewStyle().
-				PaddingLeft(6).
-				Foreground(lipgloss.AdaptiveColor{Light: "242", Dark: "246"}).
-				Italic(true)
-
-	sectionHeaderStyle = lipgloss.NewStyle().
-				PaddingLeft(2).
-				Bold(true).
-				Foreground(lipgloss.AdaptiveColor{Light: "240", Dark: "249"})
-)
-
-const maxSelectorItems = 10
-
-// ErrCancelled is returned when the user cancels the selection.
-var ErrCancelled = errors.New("cancelled")
-
-type SelectItem struct {
-	Name        string
-	Description string
-	Recommended bool
-}
-
-// selectorModel is the bubbletea model for single selection.
-type selectorModel struct {
-	title        string
-	items        []SelectItem
-	filter       string
-	cursor       int
-	scrollOffset int
-	selected     string
-	cancelled    bool
-	helpText     string
-	width        int
-}
-
-func (m selectorModel) filteredItems() []SelectItem {
-	if m.filter == "" {
-		return m.items
-	}
-	filterLower := strings.ToLower(m.filter)
-	var result []SelectItem
-	for _, item := range m.items {
-		if strings.Contains(strings.ToLower(item.Name), filterLower) {
-			result = append(result, item)
-		}
-	}
-	return result
-}
-
-func (m selectorModel) Init() tea.Cmd {
-	return nil
-}
-
-// otherStart returns the index of the first non-recommended item in the filtered list.
-// When filtering, all items scroll together so this returns 0.
-func (m selectorModel) otherStart() int {
-	if m.filter != "" {
-		return 0
-	}
-	filtered := m.filteredItems()
-	for i, item := range filtered {
-		if !item.Recommended {
-			return i
-		}
-	}
-	return len(filtered)
-}
-
-// updateNavigation handles navigation keys (up/down/pgup/pgdown/filter/backspace).
-// It does NOT handle Enter, Esc, or CtrlC. This is used by both the standalone
-// selector and the TUI modal (which intercepts Enter/Esc for its own logic).
-func (m *selectorModel) updateNavigation(msg tea.KeyMsg) {
-	filtered := m.filteredItems()
-	otherStart := m.otherStart()
-
-	switch msg.Type {
-	case tea.KeyUp:
-		if m.cursor > 0 {
-			m.cursor--
-			m.updateScroll(otherStart)
-		}
-
-	case tea.KeyDown:
-		if m.cursor < len(filtered)-1 {
-			m.cursor++
-			m.updateScroll(otherStart)
-		}
-
-	case tea.KeyPgUp:
-		m.cursor -= maxSelectorItems
-		if m.cursor < 0 {
-			m.cursor = 0
-		}
-		m.updateScroll(otherStart)
-
-	case tea.KeyPgDown:
-		m.cursor += maxSelectorItems
-		if m.cursor >= len(filtered) {
-			m.cursor = len(filtered) - 1
-		}
-		m.updateScroll(otherStart)
-
-	case tea.KeyBackspace:
-		if len(m.filter) > 0 {
-			m.filter = m.filter[:len(m.filter)-1]
-			m.cursor = 0
-			m.scrollOffset = 0
-		}
-
-	case tea.KeyRunes:
-		m.filter += string(msg.Runes)
-		m.cursor = 0
-		m.scrollOffset = 0
-	}
-}
-
-// updateScroll adjusts scrollOffset based on cursor position.
-// When not filtering, scrollOffset is relative to the "More" (non-recommended) section.
-// When filtering, it's relative to the full filtered list.
-func (m *selectorModel) updateScroll(otherStart int) {
-	if m.filter != "" {
-		if m.cursor < m.scrollOffset {
-			m.scrollOffset = m.cursor
-		}
-		if m.cursor >= m.scrollOffset+maxSelectorItems {
-			m.scrollOffset = m.cursor - maxSelectorItems + 1
-		}
-		return
-	}
-
-	// Cursor is in recommended section — reset "More" scroll to top
-	if m.cursor < otherStart {
-		m.scrollOffset = 0
-		return
-	}
-
-	// Cursor is in "More" section — scroll relative to others
-	posInOthers := m.cursor - otherStart
-	maxOthers := maxSelectorItems - otherStart
-	if maxOthers < 3 {
-		maxOthers = 3
-	}
-	if posInOthers < m.scrollOffset {
-		m.scrollOffset = posInOthers
-	}
-	if posInOthers >= m.scrollOffset+maxOthers {
-		m.scrollOffset = posInOthers - maxOthers + 1
-	}
-}
-
-func (m selectorModel) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
-	switch msg := msg.(type) {
-	case tea.WindowSizeMsg:
-		wasSet := m.width > 0
-		m.width = msg.Width
-		if wasSet {
-			return m, tea.EnterAltScreen
-		}
-		return m, nil
-
-	case tea.KeyMsg:
-		switch msg.Type {
-		case tea.KeyCtrlC, tea.KeyEsc:
-			m.cancelled = true
-			return m, tea.Quit
-
-		case tea.KeyEnter:
-			filtered := m.filteredItems()
-			if len(filtered) > 0 && m.cursor < len(filtered) {
-				m.selected = filtered[m.cursor].Name
-			}
-			return m, tea.Quit
-
-		default:
-			m.updateNavigation(msg)
-		}
-	}
-
-	return m, nil
-}
-
-func (m selectorModel) renderItem(s *strings.Builder, item SelectItem, idx int) {
-	if idx == m.cursor {
-		s.WriteString(selectorSelectedItemStyle.Render("▸ " + item.Name))
-	} else {
-		s.WriteString(selectorItemStyle.Render(item.Name))
-	}
-	s.WriteString("\n")
-	if item.Description != "" {
-		s.WriteString(selectorDescLineStyle.Render(item.Description))
-		s.WriteString("\n")
-	}
-}
-
-// renderContent renders the selector content (title, items, help text) without
-// checking the cancelled/selected state. This is used by both View() (standalone mode)
-// and by the TUI modal which embeds a selectorModel.
-func (m selectorModel) renderContent() string {
-	var s strings.Builder
-
-	s.WriteString(selectorTitleStyle.Render(m.title))
-	s.WriteString(" ")
-	if m.filter == "" {
-		s.WriteString(selectorFilterStyle.Render("Type to filter..."))
-	} else {
-		s.WriteString(selectorInputStyle.Render(m.filter))
-	}
-	s.WriteString("\n\n")
-
-	filtered := m.filteredItems()
-
-	if len(filtered) == 0 {
-		s.WriteString(selectorItemStyle.Render(selectorDescStyle.Render("(no matches)")))
-		s.WriteString("\n")
-	} else if m.filter != "" {
-		s.WriteString(sectionHeaderStyle.Render("Top Results"))
-		s.WriteString("\n")
-
-		displayCount := min(len(filtered), maxSelectorItems)
-		for i := range displayCount {
-			idx := m.scrollOffset + i
-			if idx >= len(filtered) {
-				break
-			}
-			m.renderItem(&s, filtered[idx], idx)
-		}
-
-		if remaining := len(filtered) - m.scrollOffset - displayCount; remaining > 0 {
-			s.WriteString(selectorMoreStyle.Render(fmt.Sprintf("... and %d more", remaining)))
-			s.WriteString("\n")
-		}
-	} else {
-		// Split into pinned recommended and scrollable others
-		var recItems, otherItems []int
-		for i, item := range filtered {
-			if item.Recommended {
-				recItems = append(recItems, i)
-			} else {
-				otherItems = append(otherItems, i)
-			}
-		}
-
-		// Always render all recommended items (pinned)
-		if len(recItems) > 0 {
-			s.WriteString(sectionHeaderStyle.Render("Recommended"))
-			s.WriteString("\n")
-			for _, idx := range recItems {
-				m.renderItem(&s, filtered[idx], idx)
-			}
-		}
-
-		if len(otherItems) > 0 {
-			s.WriteString("\n")
-			s.WriteString(sectionHeaderStyle.Render("More"))
-			s.WriteString("\n")
-
-			maxOthers := maxSelectorItems - len(recItems)
-			if maxOthers < 3 {
-				maxOthers = 3
-			}
-			displayCount := min(len(otherItems), maxOthers)
-
-			for i := range displayCount {
-				idx := m.scrollOffset + i
-				if idx >= len(otherItems) {
-					break
-				}
-				m.renderItem(&s, filtered[otherItems[idx]], otherItems[idx])
-			}
-
-			if remaining := len(otherItems) - m.scrollOffset - displayCount; remaining > 0 {
-				s.WriteString(selectorMoreStyle.Render(fmt.Sprintf("... and %d more", remaining)))
-				s.WriteString("\n")
-			}
-		}
-	}
-
-	s.WriteString("\n")
-	help := "↑/↓ navigate • enter select • esc cancel"
-	if m.helpText != "" {
-		help = m.helpText
-	}
-	s.WriteString(selectorHelpStyle.Render(help))
-
-	return s.String()
-}
-
-func (m selectorModel) View() string {
-	if m.cancelled || m.selected != "" {
-		return ""
-	}
-
-	s := m.renderContent()
-	if m.width > 0 {
-		return lipgloss.NewStyle().MaxWidth(m.width).Render(s)
-	}
-	return s
-}
-
-func SelectSingle(title string, items []SelectItem) (string, error) {
-	if len(items) == 0 {
-		return "", fmt.Errorf("no items to select from")
-	}
-
-	m := selectorModel{
-		title: title,
-		items: items,
-	}
-
-	p := tea.NewProgram(m)
-	finalModel, err := p.Run()
-	if err != nil {
-		return "", fmt.Errorf("error running selector: %w", err)
-	}
-
-	fm := finalModel.(selectorModel)
-	if fm.cancelled {
-		return "", ErrCancelled
-	}
-
-	return fm.selected, nil
-}
-
-// multiSelectorModel is the bubbletea model for multi selection.
-type multiSelectorModel struct {
-	title        string
-	items        []SelectItem
-	itemIndex    map[string]int
-	filter       string
-	cursor       int
-	scrollOffset int
-	checked      map[int]bool
-	checkOrder   []int
-	cancelled    bool
-	confirmed    bool
-	width        int
-}
-
-func newMultiSelectorModel(title string, items []SelectItem, preChecked []string) multiSelectorModel {
-	m := multiSelectorModel{
-		title:     title,
-		items:     items,
-		itemIndex: make(map[string]int, len(items)),
-		checked:   make(map[int]bool),
-	}
-
-	for i, item := range items {
-		m.itemIndex[item.Name] = i
-	}
-
-	for _, name := range preChecked {
-		if idx, ok := m.itemIndex[name]; ok {
-			m.checked[idx] = true
-			m.checkOrder = append(m.checkOrder, idx)
-		}
-	}
-
-	return m
-}
-
-func (m multiSelectorModel) filteredItems() []SelectItem {
-	if m.filter == "" {
-		return m.items
-	}
-	filterLower := strings.ToLower(m.filter)
-	var result []SelectItem
-	for _, item := range m.items {
-		if strings.Contains(strings.ToLower(item.Name), filterLower) {
-			result = append(result, item)
-		}
-	}
-	return result
-}
-
-func (m *multiSelectorModel) toggleItem() {
-	filtered := m.filteredItems()
-	if len(filtered) == 0 || m.cursor >= len(filtered) {
-		return
-	}
-
-	item := filtered[m.cursor]
-	origIdx := m.itemIndex[item.Name]
-
-	if m.checked[origIdx] {
-		delete(m.checked, origIdx)
-		for i, idx := range m.checkOrder {
-			if idx == origIdx {
-				m.checkOrder = append(m.checkOrder[:i], m.checkOrder[i+1:]...)
-				break
-			}
-		}
-	} else {
-		m.checked[origIdx] = true
-		m.checkOrder = append(m.checkOrder, origIdx)
-	}
-}
-
-func (m multiSelectorModel) selectedCount() int {
-	return len(m.checkOrder)
-}
-
-func (m multiSelectorModel) Init() tea.Cmd {
-	return nil
-}
-
-func (m multiSelectorModel) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
-	switch msg := msg.(type) {
-	case tea.WindowSizeMsg:
-		wasSet := m.width > 0
-		m.width = msg.Width
-		if wasSet {
-			return m, tea.EnterAltScreen
-		}
-		return m, nil
-
-	case tea.KeyMsg:
-		filtered := m.filteredItems()
-
-		switch msg.Type {
-		case tea.KeyCtrlC, tea.KeyEsc:
-			m.cancelled = true
-			return m, tea.Quit
-
-		case tea.KeyEnter:
-			if len(m.checkOrder) > 0 {
-				m.confirmed = true
-				return m, tea.Quit
-			}
-
-		case tea.KeySpace:
-			m.toggleItem()
-
-		case tea.KeyUp:
-			if m.cursor > 0 {
-				m.cursor--
-				if m.cursor < m.scrollOffset {
-					m.scrollOffset = m.cursor
-				}
-			}
-
-		case tea.KeyDown:
-			if m.cursor < len(filtered)-1 {
-				m.cursor++
-				if m.cursor >= m.scrollOffset+maxSelectorItems {
-					m.scrollOffset = m.cursor - maxSelectorItems + 1
-				}
-			}
-
-		case tea.KeyPgUp:
-			m.cursor -= maxSelectorItems
-			if m.cursor < 0 {
-				m.cursor = 0
-			}
-			m.scrollOffset -= maxSelectorItems
-			if m.scrollOffset < 0 {
-				m.scrollOffset = 0
-			}
-
-		case tea.KeyPgDown:
-			m.cursor += maxSelectorItems
-			if m.cursor >= len(filtered) {
-				m.cursor = len(filtered) - 1
-			}
-			if m.cursor >= m.scrollOffset+maxSelectorItems {
-				m.scrollOffset = m.cursor - maxSelectorItems + 1
-			}
-
-		case tea.KeyBackspace:
-			if len(m.filter) > 0 {
-				m.filter = m.filter[:len(m.filter)-1]
-				m.cursor = 0
-				m.scrollOffset = 0
-			}
-
-		case tea.KeyRunes:
-			m.filter += string(msg.Runes)
-			m.cursor = 0
-			m.scrollOffset = 0
-		}
-	}
-
-	return m, nil
-}
-
-func (m multiSelectorModel) View() string {
-	if m.cancelled || m.confirmed {
-		return ""
-	}
-
-	var s strings.Builder
-
-	s.WriteString(selectorTitleStyle.Render(m.title))
-	s.WriteString(" ")
-	if m.filter == "" {
-		s.WriteString(selectorFilterStyle.Render("Type to filter..."))
-	} else {
-		s.WriteString(selectorInputStyle.Render(m.filter))
-	}
-	s.WriteString("\n\n")
-
-	filtered := m.filteredItems()
-
-	if len(filtered) == 0 {
-		s.WriteString(selectorItemStyle.Render(selectorDescStyle.Render("(no matches)")))
-		s.WriteString("\n")
-	} else {
-		displayCount := min(len(filtered), maxSelectorItems)
-		shownRecHeader := false
-		prevWasRec := false
-
-		for i := range displayCount {
-			idx := m.scrollOffset + i
-			if idx >= len(filtered) {
-				break
-			}
-			item := filtered[idx]
-			origIdx := m.itemIndex[item.Name]
-
-			if m.filter == "" {
-				if item.Recommended && !shownRecHeader {
-					s.WriteString(sectionHeaderStyle.Render("Recommended"))
-					s.WriteString("\n")
-					shownRecHeader = true
-				} else if !item.Recommended && prevWasRec {
-					s.WriteString("\n")
-				}
-				prevWasRec = item.Recommended
-			}
-
-			var checkbox string
-			if m.checked[origIdx] {
-				checkbox = selectorCheckboxCheckedStyle.Render("[x]")
-			} else {
-				checkbox = selectorCheckboxStyle.Render("[ ]")
-			}
-
-			var line string
-			if idx == m.cursor {
-				line = selectorSelectedItemStyle.Render("▸ ") + checkbox + " " + selectorSelectedItemStyle.Render(item.Name)
-			} else {
-				line = "  " + checkbox + " " + item.Name
-			}
-
-			if len(m.checkOrder) > 0 && m.checkOrder[0] == origIdx {
-				line += " " + selectorDefaultTagStyle.Render("(default)")
-			}
-
-			s.WriteString(line)
-			s.WriteString("\n")
-		}
-
-		if remaining := len(filtered) - m.scrollOffset - displayCount; remaining > 0 {
-			s.WriteString(selectorMoreStyle.Render(fmt.Sprintf("... and %d more", remaining)))
-			s.WriteString("\n")
-		}
-	}
-
-	s.WriteString("\n")
-
-	count := m.selectedCount()
-	if count == 0 {
-		s.WriteString(selectorDescStyle.Render("  Select at least one model."))
-	} else {
-		s.WriteString(selectorDescStyle.Render(fmt.Sprintf("  %d selected - press enter to continue", count)))
-	}
-	s.WriteString("\n\n")
-
-	s.WriteString(selectorHelpStyle.Render("↑/↓ navigate • space toggle • enter confirm • esc cancel"))
-
-	result := s.String()
-	if m.width > 0 {
-		return lipgloss.NewStyle().MaxWidth(m.width).Render(result)
-	}
-	return result
-}
-
-func SelectMultiple(title string, items []SelectItem, preChecked []string) ([]string, error) {
-	if len(items) == 0 {
-		return nil, fmt.Errorf("no items to select from")
-	}
-
-	m := newMultiSelectorModel(title, items, preChecked)
-
-	p := tea.NewProgram(m)
-	finalModel, err := p.Run()
-	if err != nil {
-		return nil, fmt.Errorf("error running selector: %w", err)
-	}
-
-	fm := finalModel.(multiSelectorModel)
-	if fm.cancelled {
-		return nil, ErrCancelled
-	}
-
-	if !fm.confirmed {
-		return nil, ErrCancelled
-	}
-
-	var result []string
-	for _, idx := range fm.checkOrder {
-		result = append(result, fm.items[idx].Name)
-	}
-
-	return result, nil
-}
--- a/cmd/tui/selector_test.go
+++ b/cmd/tui/selector_test.go
@@ -1,410 +0,0 @@
-package tui
-
-import (
-	"strings"
-	"testing"
-
-	tea "github.com/charmbracelet/bubbletea"
-)
-
-func items(names ...string) []SelectItem {
-	var out []SelectItem
-	for _, n := range names {
-		out = append(out, SelectItem{Name: n})
-	}
-	return out
-}
-
-func recItems(names ...string) []SelectItem {
-	var out []SelectItem
-	for _, n := range names {
-		out = append(out, SelectItem{Name: n, Recommended: true})
-	}
-	return out
-}
-
-func mixedItems() []SelectItem {
-	return []SelectItem{
-		{Name: "rec-a", Recommended: true},
-		{Name: "rec-b", Recommended: true},
-		{Name: "other-1"},
-		{Name: "other-2"},
-		{Name: "other-3"},
-		{Name: "other-4"},
-		{Name: "other-5"},
-		{Name: "other-6"},
-		{Name: "other-7"},
-		{Name: "other-8"},
-		{Name: "other-9"},
-		{Name: "other-10"},
-	}
-}
-
-func TestFilteredItems(t *testing.T) {
-	tests := []struct {
-		name   string
-		items  []SelectItem
-		filter string
-		want   []string
-	}{
-		{
-			name:   "no filter returns all",
-			items:  items("alpha", "beta", "gamma"),
-			filter: "",
-			want:   []string{"alpha", "beta", "gamma"},
-		},
-		{
-			name:   "filter matches substring",
-			items:  items("llama3.2", "qwen3:8b", "llama2"),
-			filter: "llama",
-			want:   []string{"llama3.2", "llama2"},
-		},
-		{
-			name:   "filter is case insensitive",
-			items:  items("Qwen3:8b", "llama3.2"),
-			filter: "QWEN",
-			want:   []string{"Qwen3:8b"},
-		},
-		{
-			name:   "no matches",
-			items:  items("alpha", "beta"),
-			filter: "zzz",
-			want:   nil,
-		},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			m := selectorModel{items: tt.items, filter: tt.filter}
-			got := m.filteredItems()
-			var gotNames []string
-			for _, item := range got {
-				gotNames = append(gotNames, item.Name)
-			}
-			if len(gotNames) != len(tt.want) {
-				t.Fatalf("got %v, want %v", gotNames, tt.want)
-			}
-			for i := range tt.want {
-				if gotNames[i] != tt.want[i] {
-					t.Errorf("index %d: got %q, want %q", i, gotNames[i], tt.want[i])
-				}
-			}
-		})
-	}
-}
-
-func TestOtherStart(t *testing.T) {
-	tests := []struct {
-		name   string
-		items  []SelectItem
-		filter string
-		want   int
-	}{
-		{
-			name:  "all recommended",
-			items: recItems("a", "b", "c"),
-			want:  3,
-		},
-		{
-			name:  "none recommended",
-			items: items("a", "b"),
-			want:  0,
-		},
-		{
-			name: "mixed",
-			items: []SelectItem{
-				{Name: "rec-a", Recommended: true},
-				{Name: "rec-b", Recommended: true},
-				{Name: "other-1"},
-				{Name: "other-2"},
-			},
-			want: 2,
-		},
-		{
-			name:  "empty",
-			items: nil,
-			want:  0,
-		},
-		{
-			name: "filtering returns 0",
-			items: []SelectItem{
-				{Name: "rec-a", Recommended: true},
-				{Name: "other-1"},
-			},
-			filter: "rec",
-			want:   0,
-		},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			m := selectorModel{items: tt.items, filter: tt.filter}
-			if got := m.otherStart(); got != tt.want {
-				t.Errorf("otherStart() = %d, want %d", got, tt.want)
-			}
-		})
-	}
-}
-
-func TestUpdateScroll(t *testing.T) {
-	tests := []struct {
-		name       string
-		cursor     int
-		offset     int
-		otherStart int
-		filter     string
-		wantOffset int
-	}{
-		{
-			name:       "cursor in recommended resets scroll",
-			cursor:     1,
-			offset:     5,
-			otherStart: 3,
-			wantOffset: 0,
-		},
-		{
-			name:       "cursor at start of others",
-			cursor:     2,
-			offset:     0,
-			otherStart: 2,
-			wantOffset: 0,
-		},
-		{
-			name:       "cursor scrolls down in others",
-			cursor:     12,
-			offset:     0,
-			otherStart: 2,
-			wantOffset: 3, // posInOthers=10, maxOthers=8, 10-8+1=3
-		},
-		{
-			name:       "cursor scrolls up in others",
-			cursor:     4,
-			offset:     5,
-			otherStart: 2,
-			wantOffset: 2, // posInOthers=2 < offset=5
-		},
-		{
-			name:       "filter mode standard scroll down",
-			cursor:     12,
-			offset:     0,
-			filter:     "x",
-			otherStart: 0,
-			wantOffset: 3, // 12 - 10 + 1 = 3
-		},
-		{
-			name:       "filter mode standard scroll up",
-			cursor:     2,
-			offset:     5,
-			filter:     "x",
-			otherStart: 0,
-			wantOffset: 2,
-		},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			m := selectorModel{
-				cursor:       tt.cursor,
-				scrollOffset: tt.offset,
-				filter:       tt.filter,
-			}
-			m.updateScroll(tt.otherStart)
-			if m.scrollOffset != tt.wantOffset {
-				t.Errorf("scrollOffset = %d, want %d", m.scrollOffset, tt.wantOffset)
-			}
-		})
-	}
-}
-
-func TestRenderContent_SectionHeaders(t *testing.T) {
-	m := selectorModel{
-		title: "Pick:",
-		items: []SelectItem{
-			{Name: "rec-a", Recommended: true},
-			{Name: "other-1"},
-		},
-	}
-	content := m.renderContent()
-
-	if !strings.Contains(content, "Recommended") {
-		t.Error("should contain 'Recommended' header")
-	}
-	if !strings.Contains(content, "More") {
-		t.Error("should contain 'More' header")
-	}
-}
-
-func TestRenderContent_FilteredHeader(t *testing.T) {
-	m := selectorModel{
-		title:  "Pick:",
-		items:  items("alpha", "beta", "alphabet"),
-		filter: "alpha",
-	}
-	content := m.renderContent()
-
-	if !strings.Contains(content, "Top Results") {
-		t.Error("filtered view should contain 'Top Results' header")
-	}
-	if strings.Contains(content, "Recommended") {
-		t.Error("filtered view should not contain 'Recommended' header")
-	}
-}
-
-func TestRenderContent_NoMatches(t *testing.T) {
-	m := selectorModel{
-		title:  "Pick:",
-		items:  items("alpha"),
-		filter: "zzz",
-	}
-	content := m.renderContent()
-
-	if !strings.Contains(content, "(no matches)") {
-		t.Error("should show '(no matches)' when filter has no results")
-	}
-}
-
-func TestRenderContent_SelectedItemIndicator(t *testing.T) {
-	m := selectorModel{
-		title:  "Pick:",
-		items:  items("alpha", "beta"),
-		cursor: 0,
-	}
-	content := m.renderContent()
-
-	if !strings.Contains(content, "▸") {
-		t.Error("selected item should have ▸ indicator")
-	}
-}
-
-func TestRenderContent_Description(t *testing.T) {
-	m := selectorModel{
-		title: "Pick:",
-		items: []SelectItem{
-			{Name: "alpha", Description: "the first letter"},
-		},
-	}
-	content := m.renderContent()
-
-	if !strings.Contains(content, "the first letter") {
-		t.Error("should render item description")
-	}
-}
-
-func TestRenderContent_PinnedRecommended(t *testing.T) {
-	m := selectorModel{
-		title: "Pick:",
-		items: mixedItems(),
-		// cursor deep in "More" section
-		cursor:       8,
-		scrollOffset: 3,
-	}
-	content := m.renderContent()
-
-	// Recommended items should always be visible (pinned)
-	if !strings.Contains(content, "rec-a") {
-		t.Error("recommended items should always be rendered (pinned)")
-	}
-	if !strings.Contains(content, "rec-b") {
-		t.Error("recommended items should always be rendered (pinned)")
-	}
-}
-
-func TestRenderContent_MoreOverflowIndicator(t *testing.T) {
-	m := selectorModel{
-		title: "Pick:",
-		items: mixedItems(), // 2 rec + 10 other = 12 total, maxSelectorItems=10
-	}
-	content := m.renderContent()
-
-	if !strings.Contains(content, "... and") {
-		t.Error("should show overflow indicator when more items than visible")
-	}
-}
-
-func TestUpdateNavigation_CursorBounds(t *testing.T) {
-	m := selectorModel{
-		items:  items("a", "b", "c"),
-		cursor: 0,
-	}
-
-	// Up at top stays at 0
-	m.updateNavigation(keyMsg(KeyUp))
-	if m.cursor != 0 {
-		t.Errorf("cursor should stay at 0 when pressing up at top, got %d", m.cursor)
-	}
-
-	// Down moves to 1
-	m.updateNavigation(keyMsg(KeyDown))
-	if m.cursor != 1 {
-		t.Errorf("cursor should be 1 after down, got %d", m.cursor)
-	}
-
-	// Down to end
-	m.updateNavigation(keyMsg(KeyDown))
-	m.updateNavigation(keyMsg(KeyDown))
-	if m.cursor != 2 {
-		t.Errorf("cursor should be 2 at bottom, got %d", m.cursor)
-	}
-}
-
-func TestUpdateNavigation_FilterResetsState(t *testing.T) {
-	m := selectorModel{
-		items:        items("alpha", "beta"),
-		cursor:       1,
-		scrollOffset: 5,
-	}
-
-	m.updateNavigation(runeMsg('x'))
-	if m.filter != "x" {
-		t.Errorf("filter should be 'x', got %q", m.filter)
-	}
-	if m.cursor != 0 {
-		t.Errorf("cursor should reset to 0 on filter, got %d", m.cursor)
-	}
-	if m.scrollOffset != 0 {
-		t.Errorf("scrollOffset should reset to 0 on filter, got %d", m.scrollOffset)
-	}
-}
-
-func TestUpdateNavigation_Backspace(t *testing.T) {
-	m := selectorModel{
-		items:  items("alpha"),
-		filter: "abc",
-		cursor: 1,
-	}
-
-	m.updateNavigation(keyMsg(KeyBackspace))
-	if m.filter != "ab" {
-		t.Errorf("filter should be 'ab' after backspace, got %q", m.filter)
-	}
-	if m.cursor != 0 {
-		t.Errorf("cursor should reset to 0 on backspace, got %d", m.cursor)
-	}
-}
-
-// Key message helpers for testing
-
-type keyType = int
-
-const (
-	KeyUp        keyType = iota
-	KeyDown      keyType = iota
-	KeyBackspace keyType = iota
-)
-
-func keyMsg(k keyType) tea.KeyMsg {
-	switch k {
-	case KeyUp:
-		return tea.KeyMsg{Type: tea.KeyUp}
-	case KeyDown:
-		return tea.KeyMsg{Type: tea.KeyDown}
-	case KeyBackspace:
-		return tea.KeyMsg{Type: tea.KeyBackspace}
-	default:
-		return tea.KeyMsg{}
-	}
-}
-
-func runeMsg(r rune) tea.KeyMsg {
-	return tea.KeyMsg{Type: tea.KeyRunes, Runes: []rune{r}}
-}
--- a/cmd/tui/signin.go
+++ b/cmd/tui/signin.go
@@ -1,128 +0,0 @@
-package tui
-
-import (
-	"fmt"
-	"strings"
-	"time"
-
-	tea "github.com/charmbracelet/bubbletea"
-	"github.com/charmbracelet/lipgloss"
-	"github.com/ollama/ollama/cmd/config"
-)
-
-type signInModel struct {
-	modelName string
-	signInURL string
-	spinner   int
-	width     int
-	userName  string
-	cancelled bool
-}
-
-func (m signInModel) Init() tea.Cmd {
-	return tea.Tick(200*time.Millisecond, func(t time.Time) tea.Msg {
-		return signInTickMsg{}
-	})
-}
-
-func (m signInModel) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
-	switch msg := msg.(type) {
-	case tea.WindowSizeMsg:
-		wasSet := m.width > 0
-		m.width = msg.Width
-		if wasSet {
-			return m, tea.EnterAltScreen
-		}
-		return m, nil
-
-	case tea.KeyMsg:
-		switch msg.Type {
-		case tea.KeyCtrlC, tea.KeyEsc:
-			m.cancelled = true
-			return m, tea.Quit
-		}
-
-	case signInTickMsg:
-		m.spinner++
-		if m.spinner%5 == 0 {
-			return m, tea.Batch(
-				tea.Tick(200*time.Millisecond, func(t time.Time) tea.Msg {
-					return signInTickMsg{}
-				}),
-				checkSignIn,
-			)
-		}
-		return m, tea.Tick(200*time.Millisecond, func(t time.Time) tea.Msg {
-			return signInTickMsg{}
-		})
-
-	case signInCheckMsg:
-		if msg.signedIn {
-			m.userName = msg.userName
-			return m, tea.Quit
-		}
-	}
-
-	return m, nil
-}
-
-func (m signInModel) View() string {
-	if m.userName != "" {
-		return ""
-	}
-	return renderSignIn(m.modelName, m.signInURL, m.spinner, m.width)
-}
-
-func renderSignIn(modelName, signInURL string, spinner, width int) string {
-	spinnerFrames := []string{"⠋", "⠙", "⠹", "⠸", "⠼", "⠴", "⠦", "⠧", "⠇", "⠏"}
-	frame := spinnerFrames[spinner%len(spinnerFrames)]
-
-	urlColor := lipgloss.NewStyle().
-		Foreground(lipgloss.Color("117"))
-	urlWrap := lipgloss.NewStyle().PaddingLeft(2)
-	if width > 4 {
-		urlWrap = urlWrap.Width(width - 4)
-	}
-
-	var s strings.Builder
-
-	fmt.Fprintf(&s, "To use %s, please sign in.\n\n", selectorSelectedItemStyle.Render(modelName))
-
-	// Wrap in OSC 8 hyperlink so the entire URL is clickable even when wrapped.
-	// Padding is outside the hyperlink so spaces don't get underlined.
-	link := fmt.Sprintf("\033]8;;%s\033\\%s\033]8;;\033\\", signInURL, urlColor.Render(signInURL))
-	s.WriteString("Navigate to:\n")
-	s.WriteString(urlWrap.Render(link))
-	s.WriteString("\n\n")
-
-	s.WriteString(lipgloss.NewStyle().Foreground(lipgloss.AdaptiveColor{Light: "242", Dark: "246"}).Render(
-		frame + " Waiting for sign in to complete..."))
-	s.WriteString("\n\n")
-
-	s.WriteString(selectorHelpStyle.Render("esc cancel"))
-
-	return lipgloss.NewStyle().PaddingLeft(2).Render(s.String())
-}
-
-// RunSignIn shows a bubbletea sign-in dialog and polls until the user signs in or cancels.
-func RunSignIn(modelName, signInURL string) (string, error) {
-	config.OpenBrowser(signInURL)
-
-	m := signInModel{
-		modelName: modelName,
-		signInURL: signInURL,
-	}
-
-	p := tea.NewProgram(m)
-	finalModel, err := p.Run()
-	if err != nil {
-		return "", fmt.Errorf("error running sign-in: %w", err)
-	}
-
-	fm := finalModel.(signInModel)
-	if fm.cancelled {
-		return "", ErrCancelled
-	}
-
-	return fm.userName, nil
-}
--- a/cmd/tui/signin_test.go
+++ b/cmd/tui/signin_test.go
@@ -1,175 +0,0 @@
-package tui
-
-import (
-	"strings"
-	"testing"
-
-	tea "github.com/charmbracelet/bubbletea"
-)
-
-func TestRenderSignIn_ContainsModelName(t *testing.T) {
-	got := renderSignIn("glm-4.7:cloud", "https://example.com/signin", 0, 80)
-	if !strings.Contains(got, "glm-4.7:cloud") {
-		t.Error("should contain model name")
-	}
-	if !strings.Contains(got, "please sign in") {
-		t.Error("should contain sign-in prompt")
-	}
-}
-
-func TestRenderSignIn_ContainsURL(t *testing.T) {
-	url := "https://ollama.com/connect?key=abc123"
-	got := renderSignIn("test:cloud", url, 0, 120)
-	if !strings.Contains(got, url) {
-		t.Errorf("should contain URL %q", url)
-	}
-}
-
-func TestRenderSignIn_OSC8Hyperlink(t *testing.T) {
-	url := "https://ollama.com/connect?key=abc123"
-	got := renderSignIn("test:cloud", url, 0, 120)
-
-	// Should contain OSC 8 open sequence with the URL
-	osc8Open := "\033]8;;" + url + "\033\\"
-	if !strings.Contains(got, osc8Open) {
-		t.Error("should contain OSC 8 open sequence with URL")
-	}
-
-	// Should contain OSC 8 close sequence
-	osc8Close := "\033]8;;\033\\"
-	if !strings.Contains(got, osc8Close) {
-		t.Error("should contain OSC 8 close sequence")
-	}
-}
-
-func TestRenderSignIn_ContainsSpinner(t *testing.T) {
-	got := renderSignIn("test:cloud", "https://example.com", 0, 80)
-	if !strings.Contains(got, "Waiting for sign in to complete") {
-		t.Error("should contain waiting message")
-	}
-	if !strings.Contains(got, "⠋") {
-		t.Error("should contain first spinner frame at spinner=0")
-	}
-}
-
-func TestRenderSignIn_SpinnerAdvances(t *testing.T) {
-	got0 := renderSignIn("test:cloud", "https://example.com", 0, 80)
-	got1 := renderSignIn("test:cloud", "https://example.com", 1, 80)
-	if got0 == got1 {
-		t.Error("different spinner values should produce different output")
-	}
-}
-
-func TestRenderSignIn_ContainsEscHelp(t *testing.T) {
-	got := renderSignIn("test:cloud", "https://example.com", 0, 80)
-	if !strings.Contains(got, "esc cancel") {
-		t.Error("should contain esc cancel help text")
-	}
-}
-
-func TestSignInModel_EscCancels(t *testing.T) {
-	m := signInModel{
-		modelName: "test:cloud",
-		signInURL: "https://example.com",
-	}
-
-	updated, cmd := m.Update(tea.KeyMsg{Type: tea.KeyEsc})
-	fm := updated.(signInModel)
-	if !fm.cancelled {
-		t.Error("esc should set cancelled=true")
-	}
-	if cmd == nil {
-		t.Error("esc should return tea.Quit")
-	}
-}
-
-func TestSignInModel_CtrlCCancels(t *testing.T) {
-	m := signInModel{
-		modelName: "test:cloud",
-		signInURL: "https://example.com",
-	}
-
-	updated, cmd := m.Update(tea.KeyMsg{Type: tea.KeyCtrlC})
-	fm := updated.(signInModel)
-	if !fm.cancelled {
-		t.Error("ctrl+c should set cancelled=true")
-	}
-	if cmd == nil {
-		t.Error("ctrl+c should return tea.Quit")
-	}
-}
-
-func TestSignInModel_SignedInQuitsClean(t *testing.T) {
-	m := signInModel{
-		modelName: "test:cloud",
-		signInURL: "https://example.com",
-	}
-
-	updated, cmd := m.Update(signInCheckMsg{signedIn: true, userName: "alice"})
-	fm := updated.(signInModel)
-	if fm.userName != "alice" {
-		t.Errorf("expected userName 'alice', got %q", fm.userName)
-	}
-	if cmd == nil {
-		t.Error("successful sign-in should return tea.Quit")
-	}
-}
-
-func TestSignInModel_SignedInViewClears(t *testing.T) {
-	m := signInModel{
-		modelName: "test:cloud",
-		signInURL: "https://example.com",
-		userName:  "alice",
-	}
-
-	got := m.View()
-	if got != "" {
-		t.Errorf("View should return empty string after sign-in, got %q", got)
-	}
-}
-
-func TestSignInModel_NotSignedInContinues(t *testing.T) {
-	m := signInModel{
-		modelName: "test:cloud",
-		signInURL: "https://example.com",
-	}
-
-	updated, _ := m.Update(signInCheckMsg{signedIn: false})
-	fm := updated.(signInModel)
-	if fm.userName != "" {
-		t.Error("should not set userName when not signed in")
-	}
-	if fm.cancelled {
-		t.Error("should not cancel when check returns not signed in")
-	}
-}
-
-func TestSignInModel_WindowSizeUpdatesWidth(t *testing.T) {
-	m := signInModel{
-		modelName: "test:cloud",
-		signInURL: "https://example.com",
-	}
-
-	updated, _ := m.Update(tea.WindowSizeMsg{Width: 120, Height: 40})
-	fm := updated.(signInModel)
-	if fm.width != 120 {
-		t.Errorf("expected width 120, got %d", fm.width)
-	}
-}
-
-func TestSignInModel_TickAdvancesSpinner(t *testing.T) {
-	m := signInModel{
-		modelName: "test:cloud",
-		signInURL: "https://example.com",
-		spinner:   0,
-	}
-
-	updated, cmd := m.Update(signInTickMsg{})
-	fm := updated.(signInModel)
-	if fm.spinner != 1 {
-		t.Errorf("expected spinner=1, got %d", fm.spinner)
-	}
-	if cmd == nil {
-		t.Error("tick should return a command")
-	}
-}
--- a/cmd/tui/tui.go
+++ b/cmd/tui/tui.go
@@ -1,603 +0,0 @@
-package tui
-
-import (
-	"context"
-	"errors"
-	"fmt"
-	"strings"
-	"time"
-
-	tea "github.com/charmbracelet/bubbletea"
-	"github.com/charmbracelet/lipgloss"
-	"github.com/ollama/ollama/api"
-	"github.com/ollama/ollama/cmd/config"
-	"github.com/ollama/ollama/version"
-)
-
-var (
-	versionStyle = lipgloss.NewStyle().
-			Foreground(lipgloss.AdaptiveColor{Light: "243", Dark: "250"})
-
-	menuItemStyle = lipgloss.NewStyle().
-			PaddingLeft(2)
-
-	menuSelectedItemStyle = lipgloss.NewStyle().
-				Bold(true).
-				Background(lipgloss.AdaptiveColor{Light: "254", Dark: "236"})
-
-	menuDescStyle = selectorDescStyle.
-			PaddingLeft(4)
-
-	greyedStyle = menuItemStyle.
-			Foreground(lipgloss.AdaptiveColor{Light: "242", Dark: "246"})
-
-	greyedSelectedStyle = menuSelectedItemStyle.
-				Foreground(lipgloss.AdaptiveColor{Light: "242", Dark: "246"})
-
-	modelStyle = lipgloss.NewStyle().
-			Foreground(lipgloss.AdaptiveColor{Light: "243", Dark: "250"})
-
-	notInstalledStyle = lipgloss.NewStyle().
-				Foreground(lipgloss.AdaptiveColor{Light: "242", Dark: "246"}).
-				Italic(true)
-)
-
-type menuItem struct {
-	title       string
-	description string
-	integration string // integration name for loading model config, empty if not an integration
-	isRunModel  bool
-	isOthers    bool
-}
-
-var mainMenuItems = []menuItem{
-	{
-		title:       "Run a model",
-		description: "Start an interactive chat with a model",
-		isRunModel:  true,
-	},
-	{
-		title:       "Launch Claude Code",
-		description: "Agentic coding across large codebases",
-		integration: "claude",
-	},
-	{
-		title:       "Launch Codex",
-		description: "OpenAI's open-source coding agent",
-		integration: "codex",
-	},
-	{
-		title:       "Launch OpenClaw",
-		description: "Personal AI with 100+ skills",
-		integration: "openclaw",
-	},
-}
-
-var othersMenuItem = menuItem{
-	title:       "More...",
-	description: "Show additional integrations",
-	isOthers:    true,
-}
-
-// getOtherIntegrations dynamically builds the "Others" list from the integration
-// registry, excluding any integrations already present in the pinned mainMenuItems.
-func getOtherIntegrations() []menuItem {
-	pinned := map[string]bool{
-		"run": true, // not an integration but in the pinned list
-	}
-	for _, item := range mainMenuItems {
-		if item.integration != "" {
-			pinned[item.integration] = true
-		}
-	}
-
-	var others []menuItem
-	for _, info := range config.ListIntegrationInfos() {
-		if pinned[info.Name] {
-			continue
-		}
-		desc := info.Description
-		if desc == "" {
-			desc = "Open " + info.DisplayName + " integration"
-		}
-		others = append(others, menuItem{
-			title:       "Launch " + info.DisplayName,
-			description: desc,
-			integration: info.Name,
-		})
-	}
-	return others
-}
-
-type model struct {
-	items           []menuItem
-	cursor          int
-	quitting        bool
-	selected        bool
-	changeModel     bool
-	showOthers      bool
-	availableModels map[string]bool
-	err             error
-
-	showingModal  bool
-	modalSelector selectorModel
-	modalItems    []SelectItem
-
-	showingSignIn   bool
-	signInURL       string
-	signInModel     string
-	signInSpinner   int
-	signInFromModal bool   // true if sign-in was triggered from modal (not main menu)
-
-	width     int    // terminal width from WindowSizeMsg
-	statusMsg string // temporary status message shown near help text
-}
-
-type signInTickMsg struct{}
-
-type signInCheckMsg struct {
-	signedIn bool
-	userName string
-}
-
-type clearStatusMsg struct{}
-
-func (m *model) modelExists(name string) bool {
-	if m.availableModels == nil || name == "" {
-		return false
-	}
-	if m.availableModels[name] {
-		return true
-	}
-	// Check for prefix match (e.g., "llama2" matches "llama2:latest")
-	for modelName := range m.availableModels {
-		if strings.HasPrefix(modelName, name+":") {
-			return true
-		}
-	}
-	return false
-}
-
-func (m *model) buildModalItems() []SelectItem {
-	modelItems, _ := config.GetModelItems(context.Background())
-	var items []SelectItem
-	for _, item := range modelItems {
-		items = append(items, SelectItem{Name: item.Name, Description: item.Description, Recommended: item.Recommended})
-	}
-	return items
-}
-
-func (m *model) openModelModal() {
-	m.modalItems = m.buildModalItems()
-	m.modalSelector = selectorModel{
-		title:    "Select model:",
-		items:    m.modalItems,
-		helpText: "↑/↓ navigate • enter select • ← back",
-	}
-	m.showingModal = true
-}
-
-func isCloudModel(name string) bool {
-	return strings.HasSuffix(name, ":cloud")
-}
-
-// checkCloudSignIn checks if a cloud model needs sign-in.
-// Returns a command to start sign-in if needed, or nil if already signed in.
-func (m *model) checkCloudSignIn(modelName string, fromModal bool) tea.Cmd {
-	if modelName == "" || !isCloudModel(modelName) {
-		return nil
-	}
-	client, err := api.ClientFromEnvironment()
-	if err != nil {
-		return nil
-	}
-	user, err := client.Whoami(context.Background())
-	if err == nil && user != nil && user.Name != "" {
-		return nil
-	}
-	var aErr api.AuthorizationError
-	if errors.As(err, &aErr) && aErr.SigninURL != "" {
-		return m.startSignIn(modelName, aErr.SigninURL, fromModal)
-	}
-	return nil
-}
-
-// startSignIn initiates the sign-in flow for a cloud model.
-// fromModal indicates if this was triggered from the model picker modal.
-func (m *model) startSignIn(modelName, signInURL string, fromModal bool) tea.Cmd {
-	m.showingModal = false
-	m.showingSignIn = true
-	m.signInURL = signInURL
-	m.signInModel = modelName
-	m.signInSpinner = 0
-	m.signInFromModal = fromModal
-
-	config.OpenBrowser(signInURL)
-
-	return tea.Tick(200*time.Millisecond, func(t time.Time) tea.Msg {
-		return signInTickMsg{}
-	})
-}
-
-func checkSignIn() tea.Msg {
-	client, err := api.ClientFromEnvironment()
-	if err != nil {
-		return signInCheckMsg{signedIn: false}
-	}
-	user, err := client.Whoami(context.Background())
-	if err == nil && user != nil && user.Name != "" {
-		return signInCheckMsg{signedIn: true, userName: user.Name}
-	}
-	return signInCheckMsg{signedIn: false}
-}
-
-func (m *model) loadAvailableModels() {
-	m.availableModels = make(map[string]bool)
-	client, err := api.ClientFromEnvironment()
-	if err != nil {
-		return
-	}
-	models, err := client.List(context.Background())
-	if err != nil {
-		return
-	}
-	for _, mdl := range models.Models {
-		m.availableModels[mdl.Name] = true
-	}
-}
-
-func (m *model) buildItems() {
-	others := getOtherIntegrations()
-	m.items = make([]menuItem, 0, len(mainMenuItems)+1+len(others))
-	m.items = append(m.items, mainMenuItems...)
-
-	if m.showOthers {
-		m.items = append(m.items, others...)
-	} else {
-		m.items = append(m.items, othersMenuItem)
-	}
-}
-
-func isOthersIntegration(name string) bool {
-	for _, item := range getOtherIntegrations() {
-		if item.integration == name {
-			return true
-		}
-	}
-	return false
-}
-
-func initialModel() model {
-	m := model{
-		cursor: 0,
-	}
-	m.loadAvailableModels()
-
-	lastSelection := config.LastSelection()
-	if isOthersIntegration(lastSelection) {
-		m.showOthers = true
-	}
-
-	m.buildItems()
-
-	if lastSelection != "" {
-		for i, item := range m.items {
-			if lastSelection == "run" && item.isRunModel {
-				m.cursor = i
-				break
-			} else if item.integration == lastSelection {
-				m.cursor = i
-				break
-			}
-		}
-	}
-
-	return m
-}
-
-func (m model) Init() tea.Cmd {
-	return nil
-}
-
-func (m model) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
-	if wmsg, ok := msg.(tea.WindowSizeMsg); ok {
-		wasSet := m.width > 0
-		m.width = wmsg.Width
-		if wasSet {
-			return m, tea.EnterAltScreen
-		}
-		return m, nil
-	}
-
-	if _, ok := msg.(clearStatusMsg); ok {
-		m.statusMsg = ""
-		return m, nil
-	}
-
-	if m.showingSignIn {
-		switch msg := msg.(type) {
-		case tea.KeyMsg:
-			switch msg.Type {
-			case tea.KeyCtrlC, tea.KeyEsc:
-				m.showingSignIn = false
-				if m.signInFromModal {
-					m.showingModal = true
-				}
-				return m, nil
-			}
-
-		case signInTickMsg:
-			m.signInSpinner++
-			// Check sign-in status every 5th tick (~1 second)
-			if m.signInSpinner%5 == 0 {
-				return m, tea.Batch(
-					tea.Tick(200*time.Millisecond, func(t time.Time) tea.Msg {
-						return signInTickMsg{}
-					}),
-					checkSignIn,
-				)
-			}
-			return m, tea.Tick(200*time.Millisecond, func(t time.Time) tea.Msg {
-				return signInTickMsg{}
-			})
-
-		case signInCheckMsg:
-			if msg.signedIn {
-				if m.signInFromModal {
-					m.modalSelector.selected = m.signInModel
-					m.changeModel = true
-				} else {
-					m.selected = true
-				}
-				m.quitting = true
-				return m, tea.Quit
-			}
-		}
-		return m, nil
-	}
-
-	if m.showingModal {
-		switch msg := msg.(type) {
-		case tea.KeyMsg:
-			switch msg.Type {
-			case tea.KeyCtrlC, tea.KeyEsc, tea.KeyLeft:
-				m.showingModal = false
-				return m, nil
-
-			case tea.KeyEnter:
-				filtered := m.modalSelector.filteredItems()
-				if len(filtered) > 0 && m.modalSelector.cursor < len(filtered) {
-					m.modalSelector.selected = filtered[m.modalSelector.cursor].Name
-				}
-				if m.modalSelector.selected != "" {
-					if cmd := m.checkCloudSignIn(m.modalSelector.selected, true); cmd != nil {
-						return m, cmd
-					}
-					m.changeModel = true
-					m.quitting = true
-					return m, tea.Quit
-				}
-				return m, nil
-
-			default:
-				// Delegate navigation (up/down/pgup/pgdown/filter/backspace) to selectorModel
-				m.modalSelector.updateNavigation(msg)
-			}
-		}
-		return m, nil
-	}
-
-	switch msg := msg.(type) {
-	case tea.KeyMsg:
-		switch msg.String() {
-		case "ctrl+c", "q", "esc":
-			m.quitting = true
-			return m, tea.Quit
-
-		case "up", "k":
-			if m.cursor > 0 {
-				m.cursor--
-			}
-			// Auto-collapse "Others" when cursor moves back into pinned items
-			if m.showOthers && m.cursor < len(mainMenuItems) {
-				m.showOthers = false
-				m.buildItems()
-			}
-
-		case "down", "j":
-			if m.cursor < len(m.items)-1 {
-				m.cursor++
-			}
-			// Auto-expand "Others..." when cursor lands on it
-			if m.cursor < len(m.items) && m.items[m.cursor].isOthers && !m.showOthers {
-				m.showOthers = true
-				m.buildItems()
-				// cursor now points at the first "other" integration
-			}
-
-		case "enter", " ":
-			item := m.items[m.cursor]
-
-			if item.integration != "" && !config.IsIntegrationInstalled(item.integration) {
-				return m, nil
-			}
-
-			var configuredModel string
-			if item.isRunModel {
-				configuredModel = config.LastModel()
-			} else if item.integration != "" {
-				configuredModel = config.IntegrationModel(item.integration)
-			}
-			if cmd := m.checkCloudSignIn(configuredModel, false); cmd != nil {
-				return m, cmd
-			}
-
-			m.selected = true
-			m.quitting = true
-			return m, tea.Quit
-
-		case "right", "l":
-			item := m.items[m.cursor]
-			if item.integration != "" || item.isRunModel {
-				if item.integration != "" && !config.IsIntegrationInstalled(item.integration) {
-					return m, nil
-				}
-				m.openModelModal()
-			}
-		}
-	}
-
-	return m, nil
-}
-
-func (m model) View() string {
-	if m.quitting {
-		return ""
-	}
-
-	if m.showingSignIn {
-		return m.renderSignInDialog()
-	}
-
-	if m.showingModal {
-		return m.renderModal()
-	}
-
-	s := selectorTitleStyle.Render("Ollama "+versionStyle.Render(version.Version)) + "\n\n"
-
-	for i, item := range m.items {
-		cursor := ""
-		style := menuItemStyle
-		isInstalled := true
-
-		if item.integration != "" {
-			isInstalled = config.IsIntegrationInstalled(item.integration)
-		}
-
-		if m.cursor == i {
-			cursor = "▸ "
-			if isInstalled {
-				style = menuSelectedItemStyle
-			} else {
-				style = greyedSelectedStyle
-			}
-		} else if !isInstalled && item.integration != "" {
-			style = greyedStyle
-		}
-
-		title := item.title
-		var modelSuffix string
-		if item.integration != "" {
-			if !isInstalled {
-				title += " " + notInstalledStyle.Render("(not installed)")
-			} else if m.cursor == i {
-				if mdl := config.IntegrationModel(item.integration); mdl != "" && m.modelExists(mdl) {
-					modelSuffix = " " + modelStyle.Render("("+mdl+")")
-				}
-			}
-		} else if item.isRunModel && m.cursor == i {
-			if mdl := config.LastModel(); mdl != "" && m.modelExists(mdl) {
-				modelSuffix = " " + modelStyle.Render("("+mdl+")")
-			}
-		}
-
-		s += style.Render(cursor+title) + modelSuffix + "\n"
-
-		desc := item.description
-		if !isInstalled && item.integration != "" && m.cursor == i {
-			if hint := config.IntegrationInstallHint(item.integration); hint != "" {
-				desc = hint
-			} else {
-				desc = "not installed"
-			}
-		}
-		s += menuDescStyle.Render(desc) + "\n\n"
-	}
-
-	if m.statusMsg != "" {
-		s += "\n" + lipgloss.NewStyle().Foreground(lipgloss.AdaptiveColor{Light: "124", Dark: "210"}).Render(m.statusMsg) + "\n"
-	}
-
-	s += "\n" + selectorHelpStyle.Render("↑/↓ navigate • enter launch • → change model • esc quit")
-
-	if m.width > 0 {
-		return lipgloss.NewStyle().MaxWidth(m.width).Render(s)
-	}
-	return s
-}
-
-func (m model) renderModal() string {
-	modalStyle := lipgloss.NewStyle().
-		PaddingBottom(1).
-		PaddingRight(2)
-
-	s := modalStyle.Render(m.modalSelector.renderContent())
-	if m.width > 0 {
-		return lipgloss.NewStyle().MaxWidth(m.width).Render(s)
-	}
-	return s
-}
-
-func (m model) renderSignInDialog() string {
-	return renderSignIn(m.signInModel, m.signInURL, m.signInSpinner, m.width)
-}
-
-type Selection int
-
-const (
-	SelectionNone Selection = iota
-	SelectionRunModel
-	SelectionChangeRunModel
-	SelectionIntegration       // Generic integration selection
-	SelectionChangeIntegration // Generic change model for integration
-)
-
-type Result struct {
-	Selection   Selection
-	Integration string // integration name if applicable
-	Model       string // model name if selected from modal
-}
-
-func Run() (Result, error) {
-	m := initialModel()
-	p := tea.NewProgram(m)
-
-	finalModel, err := p.Run()
-	if err != nil {
-		return Result{Selection: SelectionNone}, fmt.Errorf("error running TUI: %w", err)
-	}
-
-	fm := finalModel.(model)
-	if fm.err != nil {
-		return Result{Selection: SelectionNone}, fm.err
-	}
-
-	if !fm.selected && !fm.changeModel {
-		return Result{Selection: SelectionNone}, nil
-	}
-
-	item := fm.items[fm.cursor]
-
-	if fm.changeModel {
-		if item.isRunModel {
-			return Result{
-				Selection: SelectionChangeRunModel,
-				Model:     fm.modalSelector.selected,
-			}, nil
-		}
-		return Result{
-			Selection:   SelectionChangeIntegration,
-			Integration: item.integration,
-			Model:       fm.modalSelector.selected,
-		}, nil
-	}
-
-	if item.isRunModel {
-		return Result{Selection: SelectionRunModel}, nil
-	}
-
-	return Result{
-		Selection:   SelectionIntegration,
-		Integration: item.integration,
-	}, nil
-}
--- a/convert/convert.go
+++ b/convert/convert.go
@@ -6,14 +6,11 @@ import (
 	"errors"
 	"fmt"
 	"io/fs"
-	"iter"
 	"log/slog"
-	"maps"
 	"os"
 	"slices"
 	"strings"

-	ofs "github.com/ollama/ollama/fs"
 	"github.com/ollama/ollama/fs/ggml"
 )

@@ -21,13 +18,8 @@ type ModelParameters struct {
 	Architectures []string `json:"architectures"`
 	VocabSize     uint32   `json:"vocab_size"`

-	// TODO is this needed?
-	ModelType string `json:"model_type"`
-
 	TextModel struct {
-		VocabSize  uint32 `json:"vocab_size"`
-		HiddenSize uint32 `json:"hidden_size"`
-		ModelType  string `json:"model_type"`
+		VocabSize uint32 `json:"vocab_size"`
 	} `json:"text_config"`
 }

@@ -41,94 +33,8 @@ type AdapterParameters struct {
 	} `json:"lora_parameters"`
 }

-type KV map[string]any
-
-func (kv KV) Architecture() string {
-	return kv.String("general.architecture", "unknown")
-}
-
-type valueTypes interface {
-	uint8 | int8 | uint16 | int16 |
-		uint32 | int32 | uint64 | int64 |
-		string | float32 | float64 | bool
-}
-
-type arrayValueTypes interface {
-	[]uint8 | []int8 | []uint16 | []int16 |
-		[]uint32 | []int32 | []uint64 | []int64 |
-		[]string | []float32 | []float64 | []bool
-}
-
-func keyValue[T valueTypes | arrayValueTypes](kv KV, key string, defaultValue ...T) (T, bool) {
-	if !strings.HasPrefix(key, "tokenizer.") && !strings.HasPrefix(key, "general.") {
-		key = kv.Architecture() + "." + key
-	}
-
-	if val, ok := kv[key].(T); ok {
-		return val, true
-	}
-	return defaultValue[0], false
-}
-
-func (kv KV) String(key string, defaultValue ...string) string {
-	val, _ := keyValue(kv, key, append(defaultValue, "")...)
-	return val
-}
-
-func (kv KV) Uint(key string, defaultValue ...uint32) uint32 {
-	val, _ := keyValue(kv, key, append(defaultValue, 0)...)
-	return val
-}
-
-func (kv KV) Float(key string, defaultValue ...float32) float32 {
-	val, _ := keyValue(kv, key, append(defaultValue, 0)...)
-	return val
-}
-
-func (kv KV) Bool(key string, defaultValue ...bool) bool {
-	val, _ := keyValue(kv, key, append(defaultValue, false)...)
-	return val
-}
-
-func (kv KV) Strings(key string, defaultValue ...[]string) []string {
-	val, _ := keyValue(kv, key, append(defaultValue, []string{""})...)
-	return val
-}
-
-func (kv KV) Ints(key string, defaultValue ...[]int32) []int32 {
-	val, _ := keyValue(kv, key, append(defaultValue, []int32{0})...)
-	return val
-}
-
-func (kv KV) Uints(key string, defaultValue ...[]uint32) []uint32 {
-	val, _ := keyValue(kv, key, append(defaultValue, []uint32{0})...)
-	return val
-}
-
-func (kv KV) Floats(key string, defaultValue ...[]float32) []float32 {
-	val, _ := keyValue(kv, key, append(defaultValue, []float32{0})...)
-	return val
-}
-
-func (kv KV) Bools(key string, defaultValue ...[]bool) []bool {
-	val, _ := keyValue(kv, key, append(defaultValue, []bool{false})...)
-	return val
-}
-
-func (kv KV) Len() int {
-	return len(kv)
-}
-
-func (kv KV) Keys() iter.Seq[string] {
-	return maps.Keys(kv)
-}
-
-func (kv KV) Value(key string) any {
-	return kv[key]
-}
-
-func (ModelParameters) KV(t *Tokenizer) KV {
-	kv := KV{
+func (ModelParameters) KV(t *Tokenizer) ggml.KV {
+	kv := ggml.KV{
 		"general.file_type":            uint32(1),
 		"general.quantization_version": uint32(2),
 		"tokenizer.ggml.pre":           t.Pre,
@@ -157,7 +63,7 @@ func (ModelParameters) KV(t *Tokenizer) KV {
 	return kv
 }

-func (p AdapterParameters) KV() KV {
+func (p AdapterParameters) KV() ggml.KV {
 	var alpha float32
 	if p.LoraParameters.Alpha == 0 {
 		alpha = float32(p.Alpha)
@@ -165,7 +71,7 @@ func (p AdapterParameters) KV() KV {
 		alpha = p.LoraParameters.Alpha
 	}

-	kv := KV{
+	kv := ggml.KV{
 		"adapter.lora.alpha": alpha,
 		"adapter.type":       "lora",
 		"general.file_type":  uint32(1),
@@ -182,14 +88,9 @@ func (ModelParameters) specialTokenTypes() []string {
 	}
 }

-type ModelKV interface {
-	// KV maps parameters to LLM key-values
-	KV(*Tokenizer) KV
-}
-
 type ModelConverter interface {
-	ModelKV
-
+	// KV maps parameters to LLM key-values
+	KV(*Tokenizer) ggml.KV
 	// Tensors maps input tensors to LLM tensors. Model specific modifications can be done here.
 	Tensors([]Tensor) []*ggml.Tensor
 	// Replacements returns a list of string pairs to replace in tensor names.
@@ -206,7 +107,7 @@ type moreParser interface {

 type AdapterConverter interface {
 	// KV maps parameters to LLM key-values
-	KV(ofs.Config) KV
+	KV(ggml.KV) ggml.KV
 	// Tensors maps input tensors to LLM tensors. Adapter specific modifications can be done here.
 	Tensors([]Tensor) []*ggml.Tensor
 	// Replacements returns a list of string pairs to replace in tensor names.
@@ -214,7 +115,7 @@ type AdapterConverter interface {
 	Replacements() []string
 }

-func ConvertAdapter(fsys fs.FS, f *os.File, baseKV ofs.Config) error {
+func ConvertAdapter(fsys fs.FS, f *os.File, baseKV ggml.KV) error {
 	bts, err := fs.ReadFile(fsys, "adapter_config.json")
 	if err != nil {
 		return err
@@ -225,8 +126,8 @@ func ConvertAdapter(fsys fs.FS, f *os.File, baseKV ofs.Config) error {
 		return err
 	}

-	arch := baseKV.Architecture()
-	if arch == "" {
+	arch, ok := baseKV["general.architecture"]
+	if !ok {
 		return errors.New("architecture not set for the base model")
 	}

@@ -252,19 +153,23 @@ func ConvertAdapter(fsys fs.FS, f *os.File, baseKV ofs.Config) error {
 	return writeFile(f, conv.KV(baseKV), conv.Tensors(ts))
 }

-func LoadModelMetadata(fsys fs.FS) (ModelKV, *Tokenizer, error) {
+// Convert writes an Ollama compatible model to the provided io.WriteSeeker based on configurations
+// and files it finds in the input path.
+// Supported input model formats include safetensors.
+// Supported input tokenizers files include tokenizer.json (preferred) and tokenizer.model.
+func ConvertModel(fsys fs.FS, f *os.File) error {
 	bts, err := fs.ReadFile(fsys, "config.json")
 	if err != nil {
-		return nil, nil, err
+		return err
 	}

 	var p ModelParameters
 	if err := json.Unmarshal(bts, &p); err != nil {
-		return nil, nil, err
+		return err
 	}

 	if len(p.Architectures) < 1 {
-		return nil, nil, errors.New("unknown architecture")
+		return errors.New("unknown architecture")
 	}

 	var conv ModelConverter
@@ -277,8 +182,6 @@ func LoadModelMetadata(fsys fs.FS) (ModelKV, *Tokenizer, error) {
 		conv = &llama4Model{}
 	case "Mistral3ForConditionalGeneration":
 		conv = &mistral3Model{}
-	case "Ministral3ForCausalLM":
-		conv = &mistral3CausalModel{}
 	case "MixtralForCausalLM":
 		conv = &mixtralModel{}
 	case "GemmaForCausalLM":
@@ -297,45 +200,29 @@ func LoadModelMetadata(fsys fs.FS) (ModelKV, *Tokenizer, error) {
 		conv = &qwen25VLModel{}
 	case "Qwen3VLForConditionalGeneration", "Qwen3VLMoeForConditionalGeneration":
 		conv = &qwen3VLModel{}
-	case "Olmo3ForCausalLM":
-		conv = &olmoModel{}
 	case "BertModel":
 		conv = &bertModel{}
-	case "NomicBertModel", "NomicBertMoEModel":
-		conv = &nomicbertModel{}
 	case "CohereForCausalLM":
 		conv = &commandrModel{}
 	case "GptOssForCausalLM":
 		conv = &gptossModel{}
-	case "DeepseekOCRForCausalLM":
-		conv = &deepseekocr{}
-	case "DeepseekV3ForCausalLM":
-		conv = &deepseek2Model{}
-	case "Glm4MoeLiteForCausalLM":
-		conv = &glm4MoeLiteModel{}
-	case "GlmOcrForConditionalGeneration":
-		conv = &glmOcrModel{}
-	case "Lfm2ForCausalLM":
-		conv = &lfm2Model{}
-	case "Qwen3NextForCausalLM":
-		conv = &qwen3NextModel{}
 	default:
-		return nil, nil, fmt.Errorf("unsupported architecture %q", p.Architectures[0])
+		return fmt.Errorf("unsupported architecture %q", p.Architectures[0])
 	}

 	if err := json.Unmarshal(bts, conv); err != nil {
-		return nil, nil, err
+		return err
 	}

 	if t, ok := conv.(moreParser); ok {
 		if err := t.parseMore(fsys); err != nil {
-			return nil, nil, err
+			return err
 		}
 	}

 	t, err := parseTokenizer(fsys, conv.specialTokenTypes())
 	if err != nil {
-		return nil, nil, err
+		return err
 	}

 	vocabSize := int(cmp.Or(p.VocabSize, p.TextModel.VocabSize))
@@ -357,19 +244,6 @@ func LoadModelMetadata(fsys fs.FS) (ModelKV, *Tokenizer, error) {
 	default:
 		slog.Debug("vocabulary", "size", len(t.Vocabulary.Tokens))
 	}
-	return conv, t, nil
-}
-
-// Convert writes an Ollama compatible model to the provided io.WriteSeeker based on configurations
-// and files it finds in the input path.
-// Supported input model formats include safetensors.
-// Supported input tokenizers files include tokenizer.json (preferred) and tokenizer.model.
-func ConvertModel(fsys fs.FS, f *os.File) error {
-	kv, t, err := LoadModelMetadata(fsys)
-	if err != nil {
-		return err
-	}
-	conv := kv.(ModelConverter)

 	ts, err := parseTensors(fsys, strings.NewReplacer(conv.Replacements()...))
 	if err != nil {
@@ -379,7 +253,7 @@ func ConvertModel(fsys fs.FS, f *os.File) error {
 	return writeFile(f, conv.KV(t), conv.Tensors(ts))
 }

-func writeFile(f *os.File, kv KV, ts []*ggml.Tensor) error {
+func writeFile(f *os.File, kv ggml.KV, ts []*ggml.Tensor) error {
 	for i := range ts {
 		ts[i].Shape = slices.Clone(ts[i].Shape)
 		slices.Reverse(ts[i].Shape)
--- a/convert/convert_bert.go
+++ b/convert/convert_bert.go
@@ -88,7 +88,7 @@ func (p *bertModel) parseMore(fsys fs.FS) error {
 	return nil
 }

-func (p *bertModel) KV(t *Tokenizer) KV {
+func (p *bertModel) KV(t *Tokenizer) ggml.KV {
 	kv := p.ModelParameters.KV(t)
 	kv["general.architecture"] = "bert"
 	kv["bert.attention.causal"] = false
--- a/convert/convert_commandr.go
+++ b/convert/convert_commandr.go
@@ -24,7 +24,7 @@ type commandrModel struct {

 var _ ModelConverter = (*commandrModel)(nil)

-func (p *commandrModel) KV(t *Tokenizer) KV {
+func (p *commandrModel) KV(t *Tokenizer) ggml.KV {
 	kv := p.ModelParameters.KV(t)
 	kv["general.architecture"] = "command-r"
 	kv["general.name"] = "command-r"
--- a/convert/convert_deepseek2.go
+++ b/convert/convert_deepseek2.go
@@ -1,173 +0,0 @@
-package convert
-
-import (
-	"cmp"
-	"fmt"
-	"log/slog"
-	"regexp"
-	"strconv"
-
-	"github.com/ollama/ollama/fs/ggml"
-)
-
-type deepseek2Model struct {
-	ModelParameters               // architectures, vocab_size
-	MaxPositionEmbeddings uint32  `json:"max_position_embeddings"`
-	HiddenSize            uint32  `json:"hidden_size"`
-	HiddenLayers          uint32  `json:"num_hidden_layers"`
-	IntermediateSize      uint32  `json:"intermediate_size"`
-	NumAttentionHeads     uint32  `json:"num_attention_heads"`
-	NumKeyValueHeads      uint32  `json:"num_key_value_heads"`
-	RMSNormEPS            float32 `json:"rms_norm_eps"`
-
-	RopeTheta     float32 `json:"rope_theta"`
-	QKNopeHeadDim uint32  `json:"qk_nope_head_dim"`
-	QKRopeHeadDim uint32  `json:"qk_rope_head_dim"`
-	KVLoraRank    uint32  `json:"kv_lora_rank"`
-	QLoraRank     uint32  `json:"q_lora_rank"`
-	VHeadDim      uint32  `json:"v_head_dim"`
-
-	ExpertCount            uint32  `json:"n_routed_experts"`
-	ExpertSharedCount      uint32  `json:"n_shared_experts"`
-	ExpertIntermediateSize uint32  `json:"moe_intermediate_size"`
-	ExpertUsedCount        uint32  `json:"num_experts_per_tok"`
-	ExpertWeightsNorm      bool    `json:"norm_topk_prob"`
-	ExpertWeightsScale     float32 `json:"routed_scaling_factor"`
-
-	ScoringFunc            string `json:"scoring_func"`
-	LeadingDenseBlockCount uint32 `json:"first_k_dense_replace"`
-
-	RopeScaling struct {
-		Factor                        float32 `json:"factor"`
-		OriginalMaxPositionEmbeddings uint32  `json:"original_max_position_embeddings"`
-		Type                          string  `json:"type"`
-		MScaleAllDim                  float32 `json:"mscale_all_dim"`
-	} `json:"rope_scaling"`
-
-	Architecture string
-}
-
-func (p *deepseek2Model) KV(t *Tokenizer) KV {
-	kv := p.ModelParameters.KV(t)
-	kv["general.architecture"] = "deepseek2"
-	kv["general.type"] = "model"
-	kv["deepseek2.block_count"] = p.HiddenLayers
-
-	numHeads := p.NumAttentionHeads
-	numKVHeads := p.NumKeyValueHeads
-
-	kv["deepseek2.attention.head_count"] = numHeads
-	kv["deepseek2.attention.head_count_kv"] = numKVHeads
-	kv["deepseek2.attention.key_length"] = p.QKNopeHeadDim + p.QKRopeHeadDim
-	kv["deepseek2.attention.kv_lora_rank"] = p.KVLoraRank
-	kv["deepseek2.attention.layer_norm_rms_epsilon"] = p.RMSNormEPS
-	kv["deepseek2.attention.q_lora_rank"] = p.QLoraRank
-	kv["deepseek2.attention.value_length"] = p.VHeadDim
-	kv["deepseek2.context_length"] = p.MaxPositionEmbeddings
-	kv["deepseek2.embedding_length"] = p.HiddenSize
-	kv["deepseek2.expert_count"] = p.ExpertCount
-	kv["deepseek2.expert_feed_forward_length"] = p.ExpertIntermediateSize
-	kv["deepseek2.expert_shared_count"] = p.ExpertSharedCount
-
-	var scoringFunc uint32
-	switch p.ScoringFunc {
-	case "softmax":
-		// not currently supported in the model, but needed for Deepseek-OCR
-		scoringFunc = 1
-	case "sigmoid":
-		scoringFunc = 2
-	}
-	kv["deepseek2.expert_gating_func"] = scoringFunc
-	kv["deepseek2.expert_used_count"] = p.ExpertUsedCount
-	kv["deepseek2.expert_weights_norm"] = p.ExpertWeightsNorm
-	kv["deepseek2.expert_weights_scale"] = p.ExpertWeightsScale
-	kv["deepseek2.feed_forward_length"] = p.IntermediateSize
-	kv["deepseek2.leading_dense_block_count"] = p.LeadingDenseBlockCount
-
-	kv["deepseek2.rope.dimension_count"] = p.QKRopeHeadDim
-	kv["deepseek2.rope.freq_base"] = cmp.Or(p.RopeTheta, 10000.0)
-	kv["deepseek2.rope.scaling.factor"] = p.RopeScaling.Factor
-	kv["deepseek2.rope.scaling.original_context_length"] = p.RopeScaling.OriginalMaxPositionEmbeddings
-	kv["deepseek2.rope.scaling.type"] = p.RopeScaling.Type
-	kv["deepseek2.rope.scaling.yarn_log_multiplier"] = 0.1 * p.RopeScaling.MScaleAllDim
-
-	kv["tokenizer.ggml.pre"] = "deepseek-v3"
-
-	return kv
-}
-
-func (p *deepseek2Model) Replacements() []string {
-	return []string{
-		"lm_head", "output",
-		"model.embed_tokens", "token_embd",
-		"model.norm", "output_norm",
-		"language_model.", "",
-		"model.layers", "blk",
-		"input_layernorm", "attn_norm",
-		"self_attn.kv_a_proj_with_mqa", "attn_kv_a_mqa",
-		"self_attn.kv_a_layernorm", "attn_kv_a_norm",
-		"self_attn.kv_b_proj", "attn_kv_b",
-		"self_attn.q_a_proj", "attn_q_a",
-		"self_attn.q_a_layernorm", "attn_q_a_norm",
-		"self_attn.q_b_proj", "attn_q_b",
-		"self_attn.o_proj", "attn_output",
-		"post_attention_layernorm", "ffn_norm",
-		"mlp.shared_experts.down_proj", "ffn_down_shexp",
-		"mlp.shared_experts.gate_proj", "ffn_gate_shexp",
-		"mlp.shared_experts.up_proj", "ffn_up_shexp",
-		"mlp.gate_proj", "ffn_gate",
-		"mlp.down_proj", "ffn_down",
-		"mlp.up_proj", "ffn_up",
-		"mlp.gate.e_score_correction_bias", "exp_probs_b.bias",
-		"mlp.gate", "ffn_gate_inp",
-	}
-}
-
-func (p *deepseek2Model) Tensors(s []Tensor) (out []*ggml.Tensor) {
-	merges := make([]merge, p.HiddenLayers*3)
-	for i := range p.HiddenLayers {
-		merges[i*3+0] = merge{
-			fmt.Sprintf("blk.%d.mlp.experts.*.gate_proj.weight", i),
-			fmt.Sprintf("blk.%d.ffn_gate_exps.weight", i),
-		}
-		merges[i*3+1] = merge{
-			fmt.Sprintf("blk.%d.mlp.experts.*.up_proj.weight", i),
-			fmt.Sprintf("blk.%d.ffn_up_exps.weight", i),
-		}
-		merges[i*3+2] = merge{
-			fmt.Sprintf("blk.%d.mlp.experts.*.down_proj.weight", i),
-			fmt.Sprintf("blk.%d.ffn_down_exps.weight", i),
-		}
-	}
-
-	skipLayer := func(n string, minValue uint32) bool {
-		re := regexp.MustCompile(`^blk\.(\d+)`)
-		matches := re.FindStringSubmatch(n)
-		if matches == nil {
-			return false
-		}
-
-		blkNum, err := strconv.Atoi(matches[1])
-		if err != nil {
-			return false
-		}
-
-		return uint32(blkNum) >= minValue
-	}
-
-	out, s = mergeTensors(s, merges...)
-	for _, t := range s {
-		// skip any additional layers (such as the Multi-Token Prediction layer)
-		if skipLayer(t.Name(), p.HiddenLayers) {
-			slog.Debug("skipping layer", "name", t.Name())
-			continue
-		}
-		out = append(out, &ggml.Tensor{
-			Name:     t.Name(),
-			Kind:     t.Kind(),
-			Shape:    t.Shape(),
-			WriterTo: t,
-		})
-	}
-	return out
-}
--- a/convert/convert_deepseekocr.go
+++ b/convert/convert_deepseekocr.go
@@ -1,136 +0,0 @@
-package convert
-
-import (
-	"fmt"
-
-	"github.com/ollama/ollama/fs/ggml"
-)
-
-type deepseekocr struct {
-	ModelParameters
-	LanguageConfig struct {
-		MaxPositionEmbeddings uint32 `json:"max_position_embeddings"`
-		HiddenSize            uint32 `json:"hidden_size"`
-		HiddenLayers          uint32 `json:"num_hidden_layers"`
-		IntermediateSize      uint32 `json:"intermediate_size"`
-		NumAttentionHeads     uint32 `json:"num_attention_heads"`
-		NumKeyValueHeads      uint32 `json:"num_key_value_heads"`
-		NumRoutedExperts      uint32 `json:"n_routed_experts"`
-		NumSharedExperts      uint32 `json:"n_shared_experts"`
-		NumExpertsPerToken    uint32 `json:"num_experts_per_tok"`
-		FirstKDenseReplace    uint32 `json:"first_k_dense_replace"`
-	} `json:"language_config"`
-
-	VisionConfig struct {
-		ImageSize uint32 `json:"image_size"`
-		Width     struct {
-			Vision struct {
-				Heads     uint32 `json:"heads"`
-				ImageSize uint32 `json:"image_size"`
-				Layers    uint32 `json:"layers"`
-				PatchSize uint32 `json:"patch_size"`
-				Width     uint32 `json:"width"`
-			} `json:"clip-l-14-224"`
-			Sam struct {
-				GlobalAttentionIndexes []int32 `json:"global_attn_indexes"`
-				Heads                  uint32  `json:"heads"`
-				Layers                 uint32  `json:"layers"`
-				Width                  uint32  `json:"width"`
-			} `json:"sam_vit_b"`
-		}
-	} `json:"vision_config"`
-}
-
-func (m *deepseekocr) KV(t *Tokenizer) KV {
-	kv := m.ModelParameters.KV(t)
-	kv["general.architecture"] = "deepseekocr"
-	kv["block_count"] = m.LanguageConfig.HiddenLayers
-	kv["context_length"] = m.LanguageConfig.MaxPositionEmbeddings
-	kv["embedding_length"] = m.LanguageConfig.HiddenSize
-	kv["feed_forward_length"] = m.LanguageConfig.IntermediateSize
-	kv["attention.head_count"] = m.LanguageConfig.NumAttentionHeads
-	kv["attention.head_count_kv"] = m.LanguageConfig.NumKeyValueHeads
-	kv["expert_count"] = m.LanguageConfig.NumRoutedExperts
-	kv["expert_used_count"] = m.LanguageConfig.NumExpertsPerToken
-	kv["leading_dense_block_count"] = m.LanguageConfig.FirstKDenseReplace
-
-	kv["vision.block_count"] = m.VisionConfig.Width.Vision.Layers
-	kv["vision.embedding_length"] = m.VisionConfig.Width.Vision.Width
-	kv["vision.head_count"] = m.VisionConfig.Width.Vision.Heads
-	kv["vision.image_size"] = m.VisionConfig.Width.Vision.ImageSize
-	kv["vision.patch_size"] = m.VisionConfig.Width.Vision.PatchSize
-
-	kv["sam.block_count"] = m.VisionConfig.Width.Sam.Layers
-	kv["sam.embedding_length"] = m.VisionConfig.Width.Sam.Width
-	kv["sam.head_count"] = m.VisionConfig.Width.Sam.Heads
-	kv["sam.global_attention_indexes"] = m.VisionConfig.Width.Sam.GlobalAttentionIndexes
-	return kv
-}
-
-func (m *deepseekocr) Tensors(s []Tensor) (out []*ggml.Tensor) {
-	merges := make([]merge, m.LanguageConfig.HiddenLayers*3)
-	for i := range m.LanguageConfig.HiddenLayers {
-		merges[i*3+0] = merge{
-			fmt.Sprintf("blk.%d.mlp.experts.*.gate_proj.weight", i),
-			fmt.Sprintf("blk.%d.ffn_gate_exps.weight", i),
-		}
-		merges[i*3+1] = merge{
-			fmt.Sprintf("blk.%d.mlp.experts.*.up_proj.weight", i),
-			fmt.Sprintf("blk.%d.ffn_up_exps.weight", i),
-		}
-		merges[i*3+2] = merge{
-			fmt.Sprintf("blk.%d.mlp.experts.*.down_proj.weight", i),
-			fmt.Sprintf("blk.%d.ffn_down_exps.weight", i),
-		}
-	}
-
-	out, s = mergeTensors(s, merges...)
-	for _, t := range s {
-		out = append(out, &ggml.Tensor{
-			Name:     t.Name(),
-			Kind:     t.Kind(),
-			Shape:    t.Shape(),
-			WriterTo: t,
-		})
-	}
-	return out
-}
-
-func (m *deepseekocr) Replacements() []string {
-	return []string{
-		"model.embed_tokens", "token_embd",
-		"model.layers", "blk",
-		"input_layernorm", "attn_norm",
-		"self_attn.q_proj", "attn_q",
-		"self_attn.k_proj", "attn_k",
-		"self_attn.v_proj", "attn_v",
-		"self_attn.o_proj", "attn_output",
-		"post_attention_layernorm", "ffn_norm",
-		"mlp.gate_proj", "ffn_gate",
-		"mlp.up_proj", "ffn_up",
-		"mlp.down_proj", "ffn_down",
-		"mlp.gate", "ffn_gate_inp",
-		"mlp.shared_experts.gate_proj", "ffn_gate_shexp",
-		"mlp.shared_experts.up_proj", "ffn_up_shexp",
-		"mlp.shared_experts.down_proj", "ffn_down_shexp",
-		"model.norm", "output_norm",
-		"lm_head", "output",
-
-		"model.vision_model", "v",
-		"embeddings.patch_embedding", "patch_embd",
-		"embeddings.class_embedding", "class_embd",
-		"embeddings.position_embedding", "position_embd",
-		"transformer.layers", "blk",
-
-		"model.projector", "mm",
-		"model.image_newline", "mm.image_newline",
-		//nolint:misspell // this misspelling is upstream. fixing it breaks the model
-		"model.view_seperator", "mm.view_seperator",
-
-		"model.sam_model.patch_embed.proj", "s.patch_embd",
-		"model.sam_model.pos_embed", "s.position_embd",
-		"model.sam_model.blocks", "s.blk",
-		"model.sam_model.neck", "s.neck",
-		"model.sam_model.net_", "s.net_",
-	}
-}
--- a/convert/convert_gemma.go
+++ b/convert/convert_gemma.go
@@ -23,7 +23,7 @@ type gemmaModel struct {

 var _ ModelConverter = (*gemmaModel)(nil)

-func (p *gemmaModel) KV(t *Tokenizer) KV {
+func (p *gemmaModel) KV(t *Tokenizer) ggml.KV {
 	kv := p.ModelParameters.KV(t)
 	kv["general.architecture"] = "gemma"
 	kv["gemma.context_length"] = p.MaxPositionEmbeddings
--- a/convert/convert_gemma2.go
+++ b/convert/convert_gemma2.go
@@ -1,5 +1,7 @@
 package convert

+import "github.com/ollama/ollama/fs/ggml"
+
 type gemma2Model struct {
 	gemmaModel
 	SlidingWindow         uint32  `json:"sliding_window"`
@@ -7,7 +9,7 @@ type gemma2Model struct {
 	FinalLogitSoftcap     float32 `json:"final_logit_softcapping"`
 }

-func (p *gemma2Model) KV(t *Tokenizer) KV {
+func (p *gemma2Model) KV(t *Tokenizer) ggml.KV {
 	kv := p.ModelParameters.KV(t)
 	kv["general.architecture"] = "gemma2"
 	kv["gemma2.context_length"] = p.MaxPositionEmbeddings
--- a/convert/convert_gemma2_adapter.go
+++ b/convert/convert_gemma2_adapter.go
@@ -6,7 +6,6 @@ import (
 	"github.com/pdevine/tensor"
 	"github.com/pdevine/tensor/native"

-	"github.com/ollama/ollama/fs"
 	"github.com/ollama/ollama/fs/ggml"
 )

@@ -16,7 +15,7 @@ type gemma2Adapter struct {

 var _ AdapterConverter = (*gemma2Adapter)(nil)

-func (p *gemma2Adapter) KV(baseKV fs.Config) KV {
+func (p *gemma2Adapter) KV(baseKV ggml.KV) ggml.KV {
 	kv := p.AdapterParameters.KV()
 	kv["general.architecture"] = "gemma2"
 	return kv
--- a/convert/convert_gemma3.go
+++ b/convert/convert_gemma3.go
@@ -2,7 +2,8 @@ package convert

 import (
 	"cmp"
-	"slices"
+
+	"github.com/ollama/ollama/fs/ggml"
 )

 type gemma3Model struct {
@@ -25,26 +26,16 @@ type gemma3Model struct {
 		NumChannels       uint32  `json:"num_channels"`        // num_channels 3
 		PatchSize         uint32  `json:"patch_size"`          // patch_size 14
 	} `json:"vision_config"`
-	MaxPositionEmbeddings    uint32   `json:"max_position_embeddings"`
-	NumAttentionHeads        uint32   `json:"num_attention_heads"`
-	NumKeyValueHeads         uint32   `json:"num_key_value_heads"`
-	RMSNormEPS               float32  `json:"rms_norm_eps"`
-	HeadDim                  uint32   `json:"head_dim"`
-	FinalLogitSoftcap        float32  `json:"final_logit_softcapping"`
-	RopeLocalTheta           float32  `json:"rope_local_base_freq"`
-	RopeTheta                float32  `json:"rope_theta"`
-	SlidingWindow            uint32   `json:"sliding_window"`
-	SlidingWindowPattern     *uint32  `json:"sliding_window_pattern"`
-	LayerTypes               []string `json:"layer_types"`
-	MultiModalTokensPerImage uint32   `json:"mm_tokens_per_image"`
-	RopeScaling              *struct {
-		Type                          string  `json:"rope_type"`
-		Factor                        float32 `json:"factor"`
-		OriginalMaxPositionEmbeddings uint32  `json:"original_max_position_embeddings"`
-		ExtrapolationFactor           float32 `json:"extrapolation_factor"`
-		BetaFast                      float32 `json:"beta_fast"`
-		BetaSlow                      float32 `json:"beta_slow"`
-	} `json:"rope_scaling"`
+	MaxPositionEmbeddings    uint32  `json:"max_position_embeddings"`
+	NumAttentionHeads        uint32  `json:"num_attention_heads"`
+	NumKeyValueHeads         uint32  `json:"num_key_value_heads"`
+	RMSNormEPS               float32 `json:"rms_norm_eps"`
+	HeadDim                  uint32  `json:"head_dim"`
+	FinalLogitSoftcap        float32 `json:"final_logit_softcapping"`
+	RopeLocalTheta           float32 `json:"rope_local_base_freq"`
+	RopeGlobalTheta          float32 `json:"rope_global_base_freq"`
+	SlidingWindow            uint32  `json:"sliding_window"`
+	MultiModalTokensPerImage uint32  `json:"mm_tokens_per_image"`
 }

 const (
@@ -53,7 +44,7 @@ const (
 	gemma27BLayerCount = 62
 )

-func (p *gemma3Model) KV(t *Tokenizer) KV {
+func (p *gemma3Model) KV(t *Tokenizer) ggml.KV {
 	kv := p.ModelParameters.KV(t)
 	kv["general.architecture"] = "gemma3"

@@ -90,38 +81,9 @@ func (p *gemma3Model) KV(t *Tokenizer) KV {
 		kv["gemma3.attention.key_length"] = p.HeadDim
 		kv["gemma3.attention.value_length"] = p.HeadDim
 		kv["gemma3.attention.sliding_window"] = p.SlidingWindow
-
-		// The sliding window pattern is either provided as the sliding_window_pattern
-		// key (an int) or as the layer_types key (a list of strings).
-		if p.SlidingWindowPattern != nil || len(p.LayerTypes) > 0 {
-			kv["gemma3.attention.sliding_window_pattern"] = slices.Collect(func(yield func(bool) bool) {
-				for i := range numBlocks {
-					var isLocal bool
-					if len(p.LayerTypes) > 0 && int(i) < len(p.LayerTypes) {
-						isLocal = p.LayerTypes[i] == "sliding_attention"
-					} else if p.SlidingWindowPattern != nil && *p.SlidingWindowPattern > 0 {
-						isLocal = (i+1)%*p.SlidingWindowPattern != 0
-					}
-					if !yield(isLocal) {
-						break
-					}
-				}
-			})
-		}
-		if p.FinalLogitSoftcap > 0 {
-			kv["gemma3.final_logit_softcapping"] = p.FinalLogitSoftcap
-		}
+		kv["gemma3.final_logit_softcapping"] = cmp.Or(p.FinalLogitSoftcap, 30)
 		kv["gemma3.rope.local.freq_base"] = cmp.Or(p.RopeLocalTheta, 10000.0)
-		kv["gemma3.rope.freq_base"] = cmp.Or(p.RopeTheta, 1000000.0)
-		if p.RopeScaling != nil && p.RopeScaling.Type == "yarn" && p.RopeScaling.Factor > 0 {
-			kv["gemma3.rope.scaling.type"] = "yarn"
-			kv["gemma3.rope.scaling.factor"] = p.RopeScaling.Factor
-			kv["gemma3.rope.scaling.original_context_length"] = p.RopeScaling.OriginalMaxPositionEmbeddings
-			kv["gemma3.rope.scaling.extrapolation_factor"] = cmp.Or(p.RopeScaling.ExtrapolationFactor, float32(1.0))
-			kv["gemma3.rope.scaling.beta_fast"] = cmp.Or(p.RopeScaling.BetaFast, float32(64.0))
-			kv["gemma3.rope.scaling.beta_slow"] = cmp.Or(p.RopeScaling.BetaSlow, float32(1.0))
-		}
-
+		kv["gemma3.rope.global.freq_base"] = cmp.Or(p.RopeGlobalTheta, 1000000.0)
 		kv["gemma3.embedding_length"] = p.HiddenSize
 		kv["gemma3.feed_forward_length"] = p.IntermediateSize
 	default:
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
jmorganca	1ab7631377	fix lint error	2025-11-06 13:55:25 -08:00
jmorganca	fed3665c70	fix tests	2025-11-06 13:49:17 -08:00
jmorganca	0a84939c11	api: add omitempty to required tool function parameter type	2025-11-06 12:58:15 -08:00