Additional thinking tags

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

.github/workflows/backend.yml

@@ -105,6 +105,19 @@ jobs:
             dockerfile: "./backend/Dockerfile.python"
             context: "./"
             ubuntu-version: '2404'
+          - build-type: 'cublas'
+            cuda-major-version: "12"
+            cuda-minor-version: "9"
+            platforms: 'linux/amd64'
+            tag-latest: 'auto'
+            tag-suffix: '-gpu-nvidia-cuda-12-pocket-tts'
+            runs-on: 'ubuntu-latest'
+            base-image: "ubuntu:24.04"
+            skip-drivers: 'false'
+            backend: "pocket-tts"
+            dockerfile: "./backend/Dockerfile.python"
+            context: "./"
+            ubuntu-version: '2404'
           - build-type: 'cublas'
             cuda-major-version: "12"
             cuda-minor-version: "0"
@@ -124,7 +137,7 @@ jobs:
             platforms: 'linux/amd64'
             tag-latest: 'auto'
             tag-suffix: '-gpu-nvidia-cuda-12-llama-cpp'
-            runs-on: 'ubuntu-latest'
+            runs-on: 'bigger-runner'
             base-image: "ubuntu:24.04"
             skip-drivers: 'false'
             backend: "llama-cpp"
@@ -340,6 +353,19 @@ jobs:
             dockerfile: "./backend/Dockerfile.python"
             context: "./"
             ubuntu-version: '2404'
+          - build-type: 'cublas'
+            cuda-major-version: "13"
+            cuda-minor-version: "0"
+            platforms: 'linux/amd64'
+            tag-latest: 'auto'
+            tag-suffix: '-gpu-nvidia-cuda-13-pocket-tts'
+            runs-on: 'ubuntu-latest'
+            base-image: "ubuntu:24.04"
+            skip-drivers: 'false'
+            backend: "pocket-tts"
+            dockerfile: "./backend/Dockerfile.python"
+            context: "./"
+            ubuntu-version: '2404'
           - build-type: 'cublas'
             cuda-major-version: "13"
             cuda-minor-version: "0"
@@ -405,6 +431,19 @@ jobs:
             backend: "vibevoice"
             dockerfile: "./backend/Dockerfile.python"
             context: "./"
+          - build-type: 'l4t'
+            cuda-major-version: "13"
+            cuda-minor-version: "0"
+            platforms: 'linux/arm64'
+            tag-latest: 'auto'
+            tag-suffix: '-nvidia-l4t-cuda-13-arm64-pocket-tts'
+            runs-on: 'ubuntu-24.04-arm'
+            base-image: "ubuntu:24.04"
+            skip-drivers: 'false'
+            ubuntu-version: '2404'
+            backend: "pocket-tts"
+            dockerfile: "./backend/Dockerfile.python"
+            context: "./"
           - build-type: 'l4t'
             cuda-major-version: "13"
             cuda-minor-version: "0"
@@ -641,13 +680,26 @@ jobs:
             dockerfile: "./backend/Dockerfile.python"
             context: "./"
             ubuntu-version: '2404'
+          - build-type: 'hipblas'
+            cuda-major-version: ""
+            cuda-minor-version: ""
+            platforms: 'linux/amd64'
+            tag-latest: 'auto'
+            tag-suffix: '-gpu-rocm-hipblas-pocket-tts'
+            runs-on: 'arc-runner-set'
+            base-image: "rocm/dev-ubuntu-24.04:6.4.4"
+            skip-drivers: 'false'
+            backend: "pocket-tts"
+            dockerfile: "./backend/Dockerfile.python"
+            context: "./"
+            ubuntu-version: '2404'
           - build-type: 'hipblas'
             cuda-major-version: ""
             cuda-minor-version: ""
             platforms: 'linux/amd64'
             tag-latest: 'auto'
             tag-suffix: '-gpu-rocm-hipblas-faster-whisper'
-            runs-on: 'ubuntu-latest'
+            runs-on: 'bigger-runner'
             base-image: "rocm/dev-ubuntu-24.04:6.4.4"
             skip-drivers: 'false'
             backend: "faster-whisper"
@@ -660,7 +712,7 @@ jobs:
             platforms: 'linux/amd64'
             tag-latest: 'auto'
             tag-suffix: '-gpu-rocm-hipblas-coqui'
-            runs-on: 'ubuntu-latest'
+            runs-on: 'bigger-runner'
             base-image: "rocm/dev-ubuntu-24.04:6.4.4"
             skip-drivers: 'false'
             backend: "coqui"
@@ -772,6 +824,19 @@ jobs:
             dockerfile: "./backend/Dockerfile.python"
             context: "./"
             ubuntu-version: '2204'
+          - build-type: 'l4t'
+            cuda-major-version: "12"
+            cuda-minor-version: "0"
+            platforms: 'linux/arm64'
+            tag-latest: 'auto'
+            tag-suffix: '-nvidia-l4t-pocket-tts'
+            runs-on: 'ubuntu-24.04-arm'
+            base-image: "nvcr.io/nvidia/l4t-jetpack:r36.4.0"
+            skip-drivers: 'true'
+            backend: "pocket-tts"
+            dockerfile: "./backend/Dockerfile.python"
+            context: "./"
+            ubuntu-version: '2204'
           - build-type: 'l4t'
             cuda-major-version: "12"
             cuda-minor-version: "0"
@@ -825,6 +890,19 @@ jobs:
             dockerfile: "./backend/Dockerfile.python"
             context: "./"
             ubuntu-version: '2404'
+          - build-type: 'intel'
+            cuda-major-version: ""
+            cuda-minor-version: ""
+            platforms: 'linux/amd64'
+            tag-latest: 'auto'
+            tag-suffix: '-gpu-intel-pocket-tts'
+            runs-on: 'arc-runner-set'
+            base-image: "intel/oneapi-basekit:2025.3.0-0-devel-ubuntu24.04"
+            skip-drivers: 'false'
+            backend: "pocket-tts"
+            dockerfile: "./backend/Dockerfile.python"
+            context: "./"
+            ubuntu-version: '2404'
           - build-type: 'intel'
             cuda-major-version: ""
             cuda-minor-version: ""
@@ -885,7 +963,7 @@ jobs:
             platforms: 'linux/amd64,linux/arm64'
             tag-latest: 'auto'
             tag-suffix: '-cpu-llama-cpp'
-            runs-on: 'ubuntu-latest'
+            runs-on: 'bigger-runner'
             base-image: "ubuntu:24.04"
             skip-drivers: 'false'
             backend: "llama-cpp"
@@ -911,7 +989,7 @@ jobs:
             platforms: 'linux/amd64,linux/arm64'
             tag-latest: 'auto'
             tag-suffix: '-gpu-vulkan-llama-cpp'
-            runs-on: 'ubuntu-latest'
+            runs-on: 'bigger-runner'
             base-image: "ubuntu:24.04"
             skip-drivers: 'false'
             backend: "llama-cpp"
@@ -1252,19 +1330,6 @@ jobs:
             dockerfile: "./backend/Dockerfile.python"
             context: "./"
             ubuntu-version: '2404'
-          - build-type: 'l4t'
-            cuda-major-version: "12"
-            cuda-minor-version: "0"
-            platforms: 'linux/arm64'
-            skip-drivers: 'true'
-            tag-latest: 'auto'
-            tag-suffix: '-nvidia-l4t-arm64-neutts'
-            base-image: "nvcr.io/nvidia/l4t-jetpack:r36.4.0"
-            runs-on: 'ubuntu-24.04-arm'
-            backend: "neutts"
-            dockerfile: "./backend/Dockerfile.python"
-            context: "./"
-            ubuntu-version: '2204'
           - build-type: ''
             cuda-major-version: ""
             cuda-minor-version: ""
@@ -1278,6 +1343,19 @@ jobs:
             dockerfile: "./backend/Dockerfile.python"
             context: "./"
             ubuntu-version: '2404'
+          - build-type: ''
+            cuda-major-version: ""
+            cuda-minor-version: ""
+            platforms: 'linux/amd64,linux/arm64'
+            tag-latest: 'auto'
+            tag-suffix: '-cpu-pocket-tts'
+            runs-on: 'ubuntu-latest'
+            base-image: "ubuntu:24.04"
+            skip-drivers: 'false'
+            backend: "pocket-tts"
+            dockerfile: "./backend/Dockerfile.python"
+            context: "./"
+            ubuntu-version: '2404'
   backend-jobs-darwin:
     uses: ./.github/workflows/backend_build_darwin.yml
     strategy:

.github/workflows/test-extra.yml

@@ -265,4 +265,23 @@ jobs:
       - name: Test moonshine
         run: |
           make --jobs=5 --output-sync=target -C backend/python/moonshine
-          make --jobs=5 --output-sync=target -C backend/python/moonshine test
+          make --jobs=5 --output-sync=target -C backend/python/moonshine test
+  tests-pocket-tts:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Clone
+        uses: actions/checkout@v6
+        with:
+          submodules: true
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install build-essential ffmpeg
+          sudo apt-get install -y ca-certificates cmake curl patch python3-pip
+          # Install UV
+          curl -LsSf https://astral.sh/uv/install.sh | sh
+          pip install --user --no-cache-dir grpcio-tools==1.64.1
+      - name: Test pocket-tts
+        run: |
+          make --jobs=5 --output-sync=target -C backend/python/pocket-tts
+          make --jobs=5 --output-sync=target -C backend/python/pocket-tts test

Dockerfile

@@ -42,22 +42,22 @@ RUN <<EOT bash
             ocaml-core ninja-build pkg-config libxml2-dev wayland-protocols python3-jsonschema \
             clang-format qtbase5-dev qt6-base-dev libxcb-glx0-dev sudo xz-utils mesa-vulkan-drivers
         if [ "amd64" = "$TARGETARCH" ]; then
-            wget "https://sdk.lunarg.com/sdk/download/1.4.328.1/linux/vulkansdk-linux-x86_64-1.4.328.1.tar.xz" && \
-            tar -xf vulkansdk-linux-x86_64-1.4.328.1.tar.xz && \
-            rm vulkansdk-linux-x86_64-1.4.328.1.tar.xz && \
+            wget "https://sdk.lunarg.com/sdk/download/1.4.335.0/linux/vulkansdk-linux-x86_64-1.4.335.0.tar.xz" && \
+            tar -xf vulkansdk-linux-x86_64-1.4.335.0.tar.xz && \
+            rm vulkansdk-linux-x86_64-1.4.335.0.tar.xz && \
             mkdir -p /opt/vulkan-sdk && \
-            mv 1.4.328.1 /opt/vulkan-sdk/ && \
-            cd /opt/vulkan-sdk/1.4.328.1 && \
+            mv 1.4.335.0 /opt/vulkan-sdk/ && \
+            cd /opt/vulkan-sdk/1.4.335.0 && \
             ./vulkansdk --no-deps --maxjobs \
                 vulkan-loader \
                 vulkan-validationlayers \
                 vulkan-extensionlayer \
                 vulkan-tools \
                 shaderc && \
-            cp -rfv /opt/vulkan-sdk/1.4.328.1/x86_64/bin/* /usr/bin/ && \
-            cp -rfv /opt/vulkan-sdk/1.4.328.1/x86_64/lib/* /usr/lib/x86_64-linux-gnu/ && \
-            cp -rfv /opt/vulkan-sdk/1.4.328.1/x86_64/include/* /usr/include/ && \
-            cp -rfv /opt/vulkan-sdk/1.4.328.1/x86_64/share/* /usr/share/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/bin/* /usr/bin/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/lib/* /usr/lib/x86_64-linux-gnu/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/include/* /usr/include/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/share/* /usr/share/ && \
             rm -rf /opt/vulkan-sdk
         fi
         if [ "arm64" = "$TARGETARCH" ]; then

Makefile

@@ -1,5 +1,5 @@
 # Disable parallel execution for backend builds
-.NOTPARALLEL: backends/diffusers backends/llama-cpp backends/piper backends/stablediffusion-ggml backends/whisper backends/faster-whisper backends/silero-vad backends/local-store backends/huggingface backends/rfdetr backends/kitten-tts backends/kokoro backends/chatterbox backends/llama-cpp-darwin backends/neutts build-darwin-python-backend build-darwin-go-backend backends/mlx backends/diffuser-darwin backends/mlx-vlm backends/mlx-audio backends/stablediffusion-ggml-darwin backends/vllm backends/moonshine
+.NOTPARALLEL: backends/diffusers backends/llama-cpp backends/piper backends/stablediffusion-ggml backends/whisper backends/faster-whisper backends/silero-vad backends/local-store backends/huggingface backends/rfdetr backends/kitten-tts backends/kokoro backends/chatterbox backends/llama-cpp-darwin backends/neutts build-darwin-python-backend build-darwin-go-backend backends/mlx backends/diffuser-darwin backends/mlx-vlm backends/mlx-audio backends/stablediffusion-ggml-darwin backends/vllm backends/moonshine backends/pocket-tts
 
 GOCMD=go
 GOTEST=$(GOCMD) test
@@ -9,7 +9,7 @@ LAUNCHER_BINARY_NAME=local-ai-launcher
 
 CUDA_MAJOR_VERSION?=13
 CUDA_MINOR_VERSION?=0
-UBUNTU_VERSION?=2204
+UBUNTU_VERSION?=2404
 UBUNTU_CODENAME?=noble
 
 GORELEASER?=
@@ -316,6 +316,7 @@ prepare-test-extra: protogen-python
 	$(MAKE) -C backend/python/vllm
 	$(MAKE) -C backend/python/vibevoice
 	$(MAKE) -C backend/python/moonshine
+	$(MAKE) -C backend/python/pocket-tts
 
 test-extra: prepare-test-extra
 	$(MAKE) -C backend/python/transformers test
@@ -324,6 +325,7 @@ test-extra: prepare-test-extra
 	$(MAKE) -C backend/python/vllm test
 	$(MAKE) -C backend/python/vibevoice test
 	$(MAKE) -C backend/python/moonshine test
+	$(MAKE) -C backend/python/pocket-tts test
 
 DOCKER_IMAGE?=local-ai
 DOCKER_AIO_IMAGE?=local-ai-aio
@@ -447,17 +449,16 @@ BACKEND_FASTER_WHISPER = faster-whisper|python|.|false|true
 BACKEND_COQUI = coqui|python|.|false|true
 BACKEND_BARK = bark|python|.|false|true
 BACKEND_EXLLAMA2 = exllama2|python|.|false|true
-
-# Python backends with ./backend context
-BACKEND_RFDETR = rfdetr|python|./backend|false|true
-BACKEND_KITTEN_TTS = kitten-tts|python|./backend|false|true
-BACKEND_NEUTTS = neutts|python|./backend|false|true
-BACKEND_KOKORO = kokoro|python|./backend|false|true
-BACKEND_VLLM = vllm|python|./backend|false|true
-BACKEND_DIFFUSERS = diffusers|python|./backend|--progress=plain|true
-BACKEND_CHATTERBOX = chatterbox|python|./backend|false|true
-BACKEND_VIBEVOICE = vibevoice|python|./backend|--progress=plain|true
-BACKEND_MOONSHINE = moonshine|python|./backend|false|true
+BACKEND_RFDETR = rfdetr|python|.|false|true
+BACKEND_KITTEN_TTS = kitten-tts|python|.|false|true
+BACKEND_NEUTTS = neutts|python|.|false|true
+BACKEND_KOKORO = kokoro|python|.|false|true
+BACKEND_VLLM = vllm|python|.|false|true
+BACKEND_DIFFUSERS = diffusers|python|.|--progress=plain|true
+BACKEND_CHATTERBOX = chatterbox|python|.|false|true
+BACKEND_VIBEVOICE = vibevoice|python|.|--progress=plain|true
+BACKEND_MOONSHINE = moonshine|python|.|false|true
+BACKEND_POCKET_TTS = pocket-tts|python|.|false|true
 
 # Helper function to build docker image for a backend
 # Usage: $(call docker-build-backend,BACKEND_NAME,DOCKERFILE_TYPE,BUILD_CONTEXT,PROGRESS_FLAG,NEEDS_BACKEND_ARG)
@@ -503,12 +504,13 @@ $(eval $(call generate-docker-build-target,$(BACKEND_DIFFUSERS)))
 $(eval $(call generate-docker-build-target,$(BACKEND_CHATTERBOX)))
 $(eval $(call generate-docker-build-target,$(BACKEND_VIBEVOICE)))
 $(eval $(call generate-docker-build-target,$(BACKEND_MOONSHINE)))
+$(eval $(call generate-docker-build-target,$(BACKEND_POCKET_TTS)))
 
 # Pattern rule for docker-save targets
 docker-save-%: backend-images
 	docker save local-ai-backend:$* -o backend-images/$*.tar
 
-docker-build-backends: docker-build-llama-cpp docker-build-rerankers docker-build-vllm docker-build-transformers docker-build-diffusers docker-build-kokoro docker-build-faster-whisper docker-build-coqui docker-build-bark docker-build-chatterbox docker-build-vibevoice docker-build-exllama2 docker-build-moonshine
+docker-build-backends: docker-build-llama-cpp docker-build-rerankers docker-build-vllm docker-build-transformers docker-build-diffusers docker-build-kokoro docker-build-faster-whisper docker-build-coqui docker-build-bark docker-build-chatterbox docker-build-vibevoice docker-build-exllama2 docker-build-moonshine docker-build-pocket-tts
 
 ########################################################
 ### END Backends

README.md

@@ -111,6 +111,8 @@
 
 ## 💻 Quickstart
 
+> ⚠️ **Note:** The `install.sh` script is currently experiencing issues due to the heavy changes currently undergoing in LocalAI and may produce broken or misconfigured installations. Please use Docker installation (see below) or manual binary installation until [issue #8032](https://github.com/mudler/LocalAI/issues/8032) is resolved.
+
 Run the installer script:
 
 ```bash
@@ -128,7 +130,7 @@ For more installation options, see [Installer Options](https://localai.io/instal
 
 > Note: the DMGs are not signed by Apple as quarantined. See https://github.com/mudler/LocalAI/issues/6268 for a workaround, fix is tracked here: https://github.com/mudler/LocalAI/issues/6244
 
-Or run with docker:
+### Containers (Docker, podman, ...)
 
 > **💡 Docker Run vs Docker Start**
 > 
@@ -137,13 +139,13 @@ Or run with docker:
 > 
 > If you've already run LocalAI before and want to start it again, use: `docker start -i local-ai`
 
-### CPU only image:
+#### CPU only image:
 
 ```bash
 docker run -ti --name local-ai -p 8080:8080 localai/localai:latest
 ```
 
-### NVIDIA GPU Images:
+#### NVIDIA GPU Images:
 
 ```bash
 # CUDA 13.0
@@ -160,25 +162,25 @@ docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-nv
 docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-nvidia-l4t-arm64-cuda-13
 ```
 
-### AMD GPU Images (ROCm):
+#### AMD GPU Images (ROCm):
 
 ```bash
 docker run -ti --name local-ai -p 8080:8080 --device=/dev/kfd --device=/dev/dri --group-add=video localai/localai:latest-gpu-hipblas
 ```
 
-### Intel GPU Images (oneAPI):
+#### Intel GPU Images (oneAPI):
 
 ```bash
 docker run -ti --name local-ai -p 8080:8080 --device=/dev/dri/card1 --device=/dev/dri/renderD128 localai/localai:latest-gpu-intel
 ```
 
-### Vulkan GPU Images:
+#### Vulkan GPU Images:
 
 ```bash
 docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-gpu-vulkan
 ```
 
-### AIO Images (pre-downloaded models):
+#### AIO Images (pre-downloaded models):
 
 ```bash
 # CPU version
@@ -295,6 +297,7 @@ LocalAI supports a comprehensive range of AI backends with multiple acceleration
 | **silero-vad** | Voice Activity Detection | CPU |
 | **neutts** | Text-to-speech with voice cloning | CUDA 12/13, ROCm, CPU |
 | **vibevoice** | Real-time TTS with voice cloning | CUDA 12/13, ROCm, Intel, CPU |
+| **pocket-tts** | Lightweight CPU-based TTS | CUDA 12/13, ROCm, Intel, CPU |
 
 ### Image & Video Generation
 | Backend | Description | Acceleration Support |
@@ -316,8 +319,8 @@ LocalAI supports a comprehensive range of AI backends with multiple acceleration
 |-------------------|-------------------|------------------|
 | **NVIDIA CUDA 12** | All CUDA-compatible backends | Nvidia hardware |
 | **NVIDIA CUDA 13** | All CUDA-compatible backends | Nvidia hardware |
-| **AMD ROCm** | llama.cpp, whisper, vllm, transformers, diffusers, rerankers, coqui, kokoro, bark, neutts, vibevoice | AMD Graphics |
-| **Intel oneAPI** | llama.cpp, whisper, stablediffusion, vllm, transformers, diffusers, rfdetr, rerankers, exllama2, coqui, kokoro, bark, vibevoice | Intel Arc, Intel iGPUs |
+| **AMD ROCm** | llama.cpp, whisper, vllm, transformers, diffusers, rerankers, coqui, kokoro, bark, neutts, vibevoice, pocket-tts | AMD Graphics |
+| **Intel oneAPI** | llama.cpp, whisper, stablediffusion, vllm, transformers, diffusers, rfdetr, rerankers, exllama2, coqui, kokoro, bark, vibevoice, pocket-tts | Intel Arc, Intel iGPUs |
 | **Apple Metal** | llama.cpp, whisper, diffusers, MLX, MLX-VLM, bark-cpp | Apple M1/M2/M3+ |
 | **Vulkan** | llama.cpp, whisper, stablediffusion | Cross-platform GPUs |
 | **NVIDIA Jetson (CUDA 12)** | llama.cpp, whisper, stablediffusion, diffusers, rfdetr | ARM64 embedded AI (AGX Orin, etc.) |

backend/Dockerfile.golang

@@ -47,22 +47,22 @@ RUN <<EOT bash
             ocaml-core ninja-build pkg-config libxml2-dev wayland-protocols python3-jsonschema \
             clang-format qtbase5-dev qt6-base-dev libxcb-glx0-dev sudo xz-utils
         if [ "amd64" = "$TARGETARCH" ]; then
-            wget "https://sdk.lunarg.com/sdk/download/1.4.328.1/linux/vulkansdk-linux-x86_64-1.4.328.1.tar.xz" && \
-            tar -xf vulkansdk-linux-x86_64-1.4.328.1.tar.xz && \
-            rm vulkansdk-linux-x86_64-1.4.328.1.tar.xz && \
+            wget "https://sdk.lunarg.com/sdk/download/1.4.335.0/linux/vulkansdk-linux-x86_64-1.4.335.0.tar.xz" && \
+            tar -xf vulkansdk-linux-x86_64-1.4.335.0.tar.xz && \
+            rm vulkansdk-linux-x86_64-1.4.335.0.tar.xz && \
             mkdir -p /opt/vulkan-sdk && \
-            mv 1.4.328.1 /opt/vulkan-sdk/ && \
-            cd /opt/vulkan-sdk/1.4.328.1 && \
+            mv 1.4.335.0 /opt/vulkan-sdk/ && \
+            cd /opt/vulkan-sdk/1.4.335.0 && \
             ./vulkansdk --no-deps --maxjobs \
                 vulkan-loader \
                 vulkan-validationlayers \
                 vulkan-extensionlayer \
                 vulkan-tools \
                 shaderc && \
-            cp -rfv /opt/vulkan-sdk/1.4.328.1/x86_64/bin/* /usr/bin/ && \
-            cp -rfv /opt/vulkan-sdk/1.4.328.1/x86_64/lib/* /usr/lib/x86_64-linux-gnu/ && \
-            cp -rfv /opt/vulkan-sdk/1.4.328.1/x86_64/include/* /usr/include/ && \
-            cp -rfv /opt/vulkan-sdk/1.4.328.1/x86_64/share/* /usr/share/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/bin/* /usr/bin/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/lib/* /usr/lib/x86_64-linux-gnu/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/include/* /usr/include/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/share/* /usr/share/ && \
             rm -rf /opt/vulkan-sdk
         fi
         if [ "arm64" = "$TARGETARCH" ]; then

backend/Dockerfile.llama-cpp

@@ -104,22 +104,22 @@ RUN <<EOT bash
             ocaml-core ninja-build pkg-config libxml2-dev wayland-protocols python3-jsonschema \
             clang-format qtbase5-dev qt6-base-dev libxcb-glx0-dev sudo xz-utils
         if [ "amd64" = "$TARGETARCH" ]; then
-            wget "https://sdk.lunarg.com/sdk/download/1.4.328.1/linux/vulkansdk-linux-x86_64-1.4.328.1.tar.xz" && \
-            tar -xf vulkansdk-linux-x86_64-1.4.328.1.tar.xz && \
-            rm vulkansdk-linux-x86_64-1.4.328.1.tar.xz && \
+            wget "https://sdk.lunarg.com/sdk/download/1.4.335.0/linux/vulkansdk-linux-x86_64-1.4.335.0.tar.xz" && \
+            tar -xf vulkansdk-linux-x86_64-1.4.335.0.tar.xz && \
+            rm vulkansdk-linux-x86_64-1.4.335.0.tar.xz && \
             mkdir -p /opt/vulkan-sdk && \
-            mv 1.4.328.1 /opt/vulkan-sdk/ && \
-            cd /opt/vulkan-sdk/1.4.328.1 && \
+            mv 1.4.335.0 /opt/vulkan-sdk/ && \
+            cd /opt/vulkan-sdk/1.4.335.0 && \
             ./vulkansdk --no-deps --maxjobs \
                 vulkan-loader \
                 vulkan-validationlayers \
                 vulkan-extensionlayer \
                 vulkan-tools \
                 shaderc && \
-            cp -rfv /opt/vulkan-sdk/1.4.328.1/x86_64/bin/* /usr/bin/ && \
-            cp -rfv /opt/vulkan-sdk/1.4.328.1/x86_64/lib/* /usr/lib/x86_64-linux-gnu/ && \
-            cp -rfv /opt/vulkan-sdk/1.4.328.1/x86_64/include/* /usr/include/ && \
-            cp -rfv /opt/vulkan-sdk/1.4.328.1/x86_64/share/* /usr/share/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/bin/* /usr/bin/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/lib/* /usr/lib/x86_64-linux-gnu/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/include/* /usr/include/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/share/* /usr/share/ && \
             rm -rf /opt/vulkan-sdk
         fi
         if [ "arm64" = "$TARGETARCH" ]; then

backend/Dockerfile.python

@@ -61,22 +61,22 @@ RUN <<EOT bash
             ocaml-core ninja-build pkg-config libxml2-dev wayland-protocols python3-jsonschema \
             clang-format qtbase5-dev qt6-base-dev libxcb-glx0-dev sudo xz-utils
         if [ "amd64" = "$TARGETARCH" ]; then
-            wget "https://sdk.lunarg.com/sdk/download/1.4.328.1/linux/vulkansdk-linux-x86_64-1.4.328.1.tar.xz" && \
-            tar -xf vulkansdk-linux-x86_64-1.4.328.1.tar.xz && \
-            rm vulkansdk-linux-x86_64-1.4.328.1.tar.xz && \
+            wget "https://sdk.lunarg.com/sdk/download/1.4.335.0/linux/vulkansdk-linux-x86_64-1.4.335.0.tar.xz" && \
+            tar -xf vulkansdk-linux-x86_64-1.4.335.0.tar.xz && \
+            rm vulkansdk-linux-x86_64-1.4.335.0.tar.xz && \
             mkdir -p /opt/vulkan-sdk && \
-            mv 1.4.328.1 /opt/vulkan-sdk/ && \
-            cd /opt/vulkan-sdk/1.4.328.1 && \
+            mv 1.4.335.0 /opt/vulkan-sdk/ && \
+            cd /opt/vulkan-sdk/1.4.335.0 && \
             ./vulkansdk --no-deps --maxjobs \
                 vulkan-loader \
                 vulkan-validationlayers \
                 vulkan-extensionlayer \
                 vulkan-tools \
                 shaderc && \
-            cp -rfv /opt/vulkan-sdk/1.4.328.1/x86_64/bin/* /usr/bin/ && \
-            cp -rfv /opt/vulkan-sdk/1.4.328.1/x86_64/lib/* /usr/lib/x86_64-linux-gnu/ && \
-            cp -rfv /opt/vulkan-sdk/1.4.328.1/x86_64/include/* /usr/include/ && \
-            cp -rfv /opt/vulkan-sdk/1.4.328.1/x86_64/share/* /usr/share/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/bin/* /usr/bin/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/lib/* /usr/lib/x86_64-linux-gnu/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/include/* /usr/include/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/share/* /usr/share/ && \
             rm -rf /opt/vulkan-sdk
         fi
         if [ "arm64" = "$TARGETARCH" ]; then

backend/cpp/llama-cpp/Makefile

@@ -1,5 +1,5 @@
 
-LLAMA_VERSION?=b1377188784f9aea26b8abde56d4aee8c733eec7
+LLAMA_VERSION?=959ecf7f234dc0bc0cd6829b25cb0ee1481aa78a
 LLAMA_REPO?=https://github.com/ggerganov/llama.cpp
 
 CMAKE_ARGS?=

backend/go/stablediffusion-ggml/Makefile

@@ -8,7 +8,7 @@ JOBS?=$(shell nproc --ignore=1)
 
 # stablediffusion.cpp (ggml)
 STABLEDIFFUSION_GGML_REPO?=https://github.com/leejet/stable-diffusion.cpp
-STABLEDIFFUSION_GGML_VERSION?=0e52afc6513cc2dea9a1a017afc4a008d5acf2b0
+STABLEDIFFUSION_GGML_VERSION?=a48b4a3ade9972faf0adcad47e51c6fc03f0e46d
 
 CMAKE_ARGS+=-DGGML_MAX_NAME=128
 

backend/go/whisper/Makefile

@@ -8,7 +8,7 @@ JOBS?=$(shell nproc --ignore=1)
 
 # whisper.cpp version
 WHISPER_REPO?=https://github.com/ggml-org/whisper.cpp
-WHISPER_CPP_VERSION?=679bdb53dbcbfb3e42685f50c7ff367949fd4d48
+WHISPER_CPP_VERSION?=f53dc74843e97f19f94a79241357f74ad5b691a6
 SO_TARGET?=libgowhisper.so
 
 CMAKE_ARGS+=-DBUILD_SHARED_LIBS=OFF

backend/index.yaml

@@ -428,6 +428,28 @@
     nvidia-l4t-cuda-12: "nvidia-l4t-vibevoice"
     nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-vibevoice"
   icon: https://avatars.githubusercontent.com/u/6154722?s=200&v=4
+- &pocket-tts
+  urls:
+    - https://github.com/kyutai-labs/pocket-tts
+  description: |
+    Pocket TTS is a lightweight text-to-speech model designed to run efficiently on CPUs.
+  tags:
+    - text-to-speech
+    - TTS
+  license: mit
+  name: "pocket-tts"
+  alias: "pocket-tts"
+  capabilities:
+    nvidia: "cuda12-pocket-tts"
+    intel: "intel-pocket-tts"
+    amd: "rocm-pocket-tts"
+    nvidia-l4t: "nvidia-l4t-pocket-tts"
+    default: "cpu-pocket-tts"
+    nvidia-cuda-13: "cuda13-pocket-tts"
+    nvidia-cuda-12: "cuda12-pocket-tts"
+    nvidia-l4t-cuda-12: "nvidia-l4t-pocket-tts"
+    nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-pocket-tts"
+  icon: https://avatars.githubusercontent.com/u/6154722?s=200&v=4
 - &piper
   name: "piper"
   uri: "quay.io/go-skynet/local-ai-backends:latest-piper"
@@ -515,18 +537,14 @@
     default: "cpu-neutts"
     nvidia: "cuda12-neutts"
     amd: "rocm-neutts"
-    nvidia-l4t: "nvidia-l4t-neutts"
     nvidia-cuda-12: "cuda12-neutts"
-    nvidia-l4t-cuda-12: "nvidia-l4t-arm64-neutts"
 - !!merge <<: *neutts
   name: "neutts-development"
   capabilities:
     default: "cpu-neutts-development"
     nvidia: "cuda12-neutts-development"
     amd: "rocm-neutts-development"
-    nvidia-l4t: "nvidia-l4t-neutts-development"
     nvidia-cuda-12: "cuda12-neutts-development"
-    nvidia-l4t-cuda-12: "nvidia-l4t-arm64-neutts-development"
 - !!merge <<: *llamacpp
   name: "llama-cpp-development"
   capabilities:
@@ -556,11 +574,6 @@
   uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-rocm-hipblas-neutts"
   mirrors:
     - localai/localai-backends:latest-gpu-rocm-hipblas-neutts
-- !!merge <<: *neutts
-  name: "nvidia-l4t-arm64-neutts"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-nvidia-l4t-arm64-neutts"
-  mirrors:
-    - localai/localai-backends:latest-nvidia-l4t-arm64-neutts
 - !!merge <<: *neutts
   name: "cpu-neutts-development"
   uri: "quay.io/go-skynet/local-ai-backends:master-cpu-neutts"
@@ -576,11 +589,6 @@
   uri: "quay.io/go-skynet/local-ai-backends:master-gpu-rocm-hipblas-neutts"
   mirrors:
     - localai/localai-backends:master-gpu-rocm-hipblas-neutts
-- !!merge <<: *neutts
-  name: "nvidia-l4t-arm64-neutts-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-nvidia-l4t-arm64-neutts"
-  mirrors:
-    - localai/localai-backends:master-nvidia-l4t-arm64-neutts
 - !!merge <<: *mlx
   name: "mlx-development"
   uri: "quay.io/go-skynet/local-ai-backends:master-metal-darwin-arm64-mlx"
@@ -1605,3 +1613,86 @@
   uri: "quay.io/go-skynet/local-ai-backends:master-nvidia-l4t-cuda-13-arm64-vibevoice"
   mirrors:
     - localai/localai-backends:master-nvidia-l4t-cuda-13-arm64-vibevoice
+## pocket-tts
+- !!merge <<: *pocket-tts
+  name: "pocket-tts-development"
+  capabilities:
+    nvidia: "cuda12-pocket-tts-development"
+    intel: "intel-pocket-tts-development"
+    amd: "rocm-pocket-tts-development"
+    nvidia-l4t: "nvidia-l4t-pocket-tts-development"
+    default: "cpu-pocket-tts-development"
+    nvidia-cuda-13: "cuda13-pocket-tts-development"
+    nvidia-cuda-12: "cuda12-pocket-tts-development"
+    nvidia-l4t-cuda-12: "nvidia-l4t-pocket-tts-development"
+    nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-pocket-tts-development"
+- !!merge <<: *pocket-tts
+  name: "cpu-pocket-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-cpu-pocket-tts"
+  mirrors:
+    - localai/localai-backends:latest-cpu-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "cpu-pocket-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-cpu-pocket-tts"
+  mirrors:
+    - localai/localai-backends:master-cpu-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "cuda12-pocket-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-pocket-tts"
+  mirrors:
+    - localai/localai-backends:latest-gpu-nvidia-cuda-12-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "cuda12-pocket-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-pocket-tts"
+  mirrors:
+    - localai/localai-backends:master-gpu-nvidia-cuda-12-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "cuda13-pocket-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-13-pocket-tts"
+  mirrors:
+    - localai/localai-backends:latest-gpu-nvidia-cuda-13-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "cuda13-pocket-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-13-pocket-tts"
+  mirrors:
+    - localai/localai-backends:master-gpu-nvidia-cuda-13-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "intel-pocket-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-intel-pocket-tts"
+  mirrors:
+    - localai/localai-backends:latest-gpu-intel-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "intel-pocket-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-intel-pocket-tts"
+  mirrors:
+    - localai/localai-backends:master-gpu-intel-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "rocm-pocket-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-rocm-hipblas-pocket-tts"
+  mirrors:
+    - localai/localai-backends:latest-gpu-rocm-hipblas-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "rocm-pocket-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-rocm-hipblas-pocket-tts"
+  mirrors:
+    - localai/localai-backends:master-gpu-rocm-hipblas-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "nvidia-l4t-pocket-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-nvidia-l4t-pocket-tts"
+  mirrors:
+    - localai/localai-backends:latest-nvidia-l4t-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "nvidia-l4t-pocket-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-nvidia-l4t-pocket-tts"
+  mirrors:
+    - localai/localai-backends:master-nvidia-l4t-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "cuda13-nvidia-l4t-arm64-pocket-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-nvidia-l4t-cuda-13-arm64-pocket-tts"
+  mirrors:
+    - localai/localai-backends:latest-nvidia-l4t-cuda-13-arm64-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "cuda13-nvidia-l4t-arm64-pocket-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-nvidia-l4t-cuda-13-arm64-pocket-tts"
+  mirrors:
+    - localai/localai-backends:master-nvidia-l4t-cuda-13-arm64-pocket-tts

backend/python/diffusers/backend.py

@@ -41,6 +41,14 @@ from optimum.quanto import freeze, qfloat8, quantize
 from transformers import T5EncoderModel
 from safetensors.torch import load_file
 
+# Import LTX-2 specific utilities
+try:
+    from diffusers.pipelines.ltx2.export_utils import encode_video as ltx2_encode_video
+    LTX2_AVAILABLE = True
+except ImportError:
+    LTX2_AVAILABLE = False
+    ltx2_encode_video = None
+
 _ONE_DAY_IN_SECONDS = 60 * 60 * 24
 COMPEL = os.environ.get("COMPEL", "0") == "1"
 XPU = os.environ.get("XPU", "0") == "1"
@@ -290,6 +298,20 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                 pipe.enable_model_cpu_offload()
             return pipe
 
+        # LTX2ImageToVideoPipeline - needs img2vid flag, CPU offload, and special handling
+        if pipeline_type == "LTX2ImageToVideoPipeline":
+            self.img2vid = True
+            self.ltx2_pipeline = True
+            pipe = load_diffusers_pipeline(
+                class_name="LTX2ImageToVideoPipeline",
+                model_id=request.Model,
+                torch_dtype=torchType,
+                variant=variant
+            )
+            if not DISABLE_CPU_OFFLOAD:
+                pipe.enable_model_cpu_offload()
+            return pipe
+
         # ================================================================
         # Dynamic pipeline loading - the default path for most pipelines
         # Uses the dynamic loader to instantiate any pipeline by class name
@@ -404,6 +426,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
             fromSingleFile = request.Model.startswith("http") or request.Model.startswith("/") or local
             self.img2vid = False
             self.txt2vid = False
+            self.ltx2_pipeline = False
 
             # Load pipeline using dynamic loader
             # Special cases that require custom initialization are handled first
@@ -686,7 +709,44 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
             print(f"Generating video with {kwargs=}", file=sys.stderr)
 
             # Generate video frames based on pipeline type
-            if self.PipelineType == "WanPipeline":
+            if self.ltx2_pipeline or self.PipelineType == "LTX2ImageToVideoPipeline":
+                # LTX-2 image-to-video generation with audio
+                if not LTX2_AVAILABLE:
+                    return backend_pb2.Result(success=False, message="LTX-2 pipeline requires diffusers.pipelines.ltx2.export_utils")
+                
+                # LTX-2 uses 'image' parameter instead of 'start_image'
+                if request.start_image:
+                    image = load_image(request.start_image)
+                    kwargs["image"] = image
+                    # Remove start_image if it was added
+                    kwargs.pop("start_image", None)
+                
+                # LTX-2 uses 'frame_rate' instead of 'fps'
+                frame_rate = float(fps)
+                kwargs["frame_rate"] = frame_rate
+                
+                # LTX-2 requires output_type="np" and return_dict=False
+                kwargs["output_type"] = "np"
+                kwargs["return_dict"] = False
+                
+                # Generate video and audio
+                video, audio = self.pipe(**kwargs)
+                
+                # Convert video to uint8 format
+                video = (video * 255).round().astype("uint8")
+                video = torch.from_numpy(video)
+                
+                # Use LTX-2's encode_video function which handles audio
+                ltx2_encode_video(
+                    video[0],
+                    fps=frame_rate,
+                    audio=audio[0].float().cpu(),
+                    audio_sample_rate=self.pipe.vocoder.config.output_sampling_rate,
+                    output_path=request.dst,
+                )
+                
+                return backend_pb2.Result(message="Video generated successfully", success=True)
+            elif self.PipelineType == "WanPipeline":
                 # WAN2.2 text-to-video generation
                 output = self.pipe(**kwargs)
                 frames = output.frames[0]  # WAN2.2 returns frames in this format
@@ -727,7 +787,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
             else:
                 return backend_pb2.Result(success=False, message=f"Pipeline {self.PipelineType} does not support video generation")
 
-            # Export video
+            # Export video (for non-LTX-2 pipelines)
             export_to_video(frames, request.dst, fps=fps)
             
             return backend_pb2.Result(message="Video generated successfully", success=True)

backend/python/pocket-tts/Makefile

@@ -0,0 +1,23 @@
+.PHONY: pocket-tts
+pocket-tts:
+	bash install.sh
+
+.PHONY: run
+run: pocket-tts
+	@echo "Running pocket-tts..."
+	bash run.sh
+	@echo "pocket-tts run."
+
+.PHONY: test
+test: pocket-tts
+	@echo "Testing pocket-tts..."
+	bash test.sh
+	@echo "pocket-tts tested."
+
+.PHONY: protogen-clean
+protogen-clean:
+	$(RM) backend_pb2_grpc.py backend_pb2.py
+
+.PHONY: clean
+clean: protogen-clean
+	rm -rf venv __pycache__

backend/python/pocket-tts/backend.py

@@ -0,0 +1,255 @@
+#!/usr/bin/env python3
+"""
+This is an extra gRPC server of LocalAI for Pocket TTS
+"""
+from concurrent import futures
+import time
+import argparse
+import signal
+import sys
+import os
+import traceback
+import scipy.io.wavfile
+import backend_pb2
+import backend_pb2_grpc
+import torch
+from pocket_tts import TTSModel
+
+import grpc
+
+def is_float(s):
+    """Check if a string can be converted to float."""
+    try:
+        float(s)
+        return True
+    except ValueError:
+        return False
+
+def is_int(s):
+    """Check if a string can be converted to int."""
+    try:
+        int(s)
+        return True
+    except ValueError:
+        return False
+
+_ONE_DAY_IN_SECONDS = 60 * 60 * 24
+
+# If MAX_WORKERS are specified in the environment use it, otherwise default to 1
+MAX_WORKERS = int(os.environ.get('PYTHON_GRPC_MAX_WORKERS', '1'))
+
+# Implement the BackendServicer class with the service methods
+class BackendServicer(backend_pb2_grpc.BackendServicer):
+    """
+    BackendServicer is the class that implements the gRPC service
+    """
+    def Health(self, request, context):
+        return backend_pb2.Reply(message=bytes("OK", 'utf-8'))
+    
+    def LoadModel(self, request, context):
+        # Get device
+        if torch.cuda.is_available():
+            print("CUDA is available", file=sys.stderr)
+            device = "cuda"
+        else:
+            print("CUDA is not available", file=sys.stderr)
+            device = "cpu"
+        mps_available = hasattr(torch.backends, "mps") and torch.backends.mps.is_available()
+        if mps_available:
+            device = "mps"
+        if not torch.cuda.is_available() and request.CUDA:
+            return backend_pb2.Result(success=False, message="CUDA is not available")
+
+        # Normalize potential 'mpx' typo to 'mps'
+        if device == "mpx":
+            print("Note: device 'mpx' detected, treating it as 'mps'.", file=sys.stderr)
+            device = "mps"
+        
+        # Validate mps availability if requested
+        if device == "mps" and not torch.backends.mps.is_available():
+            print("Warning: MPS not available. Falling back to CPU.", file=sys.stderr)
+            device = "cpu"
+
+        self.device = device
+
+        options = request.Options
+
+        # empty dict
+        self.options = {}
+
+        # The options are a list of strings in this form optname:optvalue
+        # We are storing all the options in a dict so we can use it later when
+        # generating the audio
+        for opt in options:
+            if ":" not in opt:
+                continue
+            key, value = opt.split(":", 1)  # Split only on first colon
+            # if value is a number, convert it to the appropriate type
+            if is_float(value):
+                value = float(value)
+            elif is_int(value):
+                value = int(value)
+            elif value.lower() in ["true", "false"]:
+                value = value.lower() == "true"
+            self.options[key] = value
+
+        # Default voice for caching
+        self.default_voice_url = self.options.get("default_voice", None)
+        self._voice_cache = {}
+
+        try:
+            print("Loading Pocket TTS model", file=sys.stderr)
+            self.tts_model = TTSModel.load_model()
+            print(f"Model loaded successfully. Sample rate: {self.tts_model.sample_rate}", file=sys.stderr)
+
+            # Pre-load default voice if specified
+            if self.default_voice_url:
+                try:
+                    print(f"Pre-loading default voice: {self.default_voice_url}", file=sys.stderr)
+                    voice_state = self.tts_model.get_state_for_audio_prompt(self.default_voice_url)
+                    self._voice_cache[self.default_voice_url] = voice_state
+                    print("Default voice loaded successfully", file=sys.stderr)
+                except Exception as e:
+                    print(f"Warning: Failed to pre-load default voice: {e}", file=sys.stderr)
+
+        except Exception as err:
+            return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
+        
+        return backend_pb2.Result(message="Model loaded successfully", success=True)
+
+    def _get_voice_state(self, voice_input):
+        """
+        Get voice state from cache or load it.
+        voice_input can be:
+        - HuggingFace URL (e.g., hf://kyutai/tts-voices/alba-mackenna/casual.wav)
+        - Local file path
+        - None (use default)
+        """
+        # Use default if no voice specified
+        if not voice_input:
+            voice_input = self.default_voice_url
+
+        if not voice_input:
+            return None
+
+        # Check cache first
+        if voice_input in self._voice_cache:
+            return self._voice_cache[voice_input]
+
+        # Load voice state
+        try:
+            print(f"Loading voice from: {voice_input}", file=sys.stderr)
+            voice_state = self.tts_model.get_state_for_audio_prompt(voice_input)
+            self._voice_cache[voice_input] = voice_state
+            return voice_state
+        except Exception as e:
+            print(f"Error loading voice from {voice_input}: {e}", file=sys.stderr)
+            return None
+
+    def TTS(self, request, context):
+        try:
+            # Determine voice input
+            # Priority: request.voice > AudioPath (from ModelOptions) > default
+            voice_input = None
+            
+            if request.voice:
+                voice_input = request.voice
+            elif hasattr(request, 'AudioPath') and request.AudioPath:
+                # Use AudioPath as voice file
+                if os.path.isabs(request.AudioPath):
+                    voice_input = request.AudioPath
+                elif hasattr(request, 'ModelFile') and request.ModelFile:
+                    model_file_base = os.path.dirname(request.ModelFile)
+                    voice_input = os.path.join(model_file_base, request.AudioPath)
+                elif hasattr(request, 'ModelPath') and request.ModelPath:
+                    voice_input = os.path.join(request.ModelPath, request.AudioPath)
+                else:
+                    voice_input = request.AudioPath
+
+            # Get voice state
+            voice_state = self._get_voice_state(voice_input)
+            if voice_state is None:
+                return backend_pb2.Result(
+                    success=False,
+                    message=f"Voice not found or failed to load: {voice_input}. Please provide a valid voice URL or file path."
+                )
+
+            # Prepare text
+            text = request.text.strip()
+
+            if not text:
+                return backend_pb2.Result(
+                    success=False,
+                    message="Text is empty"
+                )
+
+            print(f"Generating audio for text: {text[:50]}...", file=sys.stderr)
+
+            # Generate audio
+            audio = self.tts_model.generate_audio(voice_state, text)
+
+            # Audio is a 1D torch tensor containing PCM data
+            if audio is None or audio.numel() == 0:
+                return backend_pb2.Result(
+                    success=False,
+                    message="No audio generated"
+                )
+
+            # Save audio to file
+            output_path = request.dst
+            if not output_path:
+                output_path = "/tmp/pocket-tts-output.wav"
+
+            # Ensure output directory exists
+            output_dir = os.path.dirname(output_path)
+            if output_dir and not os.path.exists(output_dir):
+                os.makedirs(output_dir, exist_ok=True)
+
+            # Convert torch tensor to numpy and save
+            audio_numpy = audio.numpy()
+            scipy.io.wavfile.write(output_path, self.tts_model.sample_rate, audio_numpy)
+            print(f"Saved audio to {output_path}", file=sys.stderr)
+
+        except Exception as err:
+            print(f"Error in TTS: {err}", file=sys.stderr)
+            print(traceback.format_exc(), file=sys.stderr)
+            return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
+        
+        return backend_pb2.Result(success=True)
+
+def serve(address):
+    server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS),
+        options=[
+            ('grpc.max_message_length', 50 * 1024 * 1024),  # 50MB
+            ('grpc.max_send_message_length', 50 * 1024 * 1024),  # 50MB
+            ('grpc.max_receive_message_length', 50 * 1024 * 1024),  # 50MB
+        ])
+    backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server)
+    server.add_insecure_port(address)
+    server.start()
+    print("Server started. Listening on: " + address, file=sys.stderr)
+
+    # Define the signal handler function
+    def signal_handler(sig, frame):
+        print("Received termination signal. Shutting down...")
+        server.stop(0)
+        sys.exit(0)
+
+    # Set the signal handlers for SIGINT and SIGTERM
+    signal.signal(signal.SIGINT, signal_handler)
+    signal.signal(signal.SIGTERM, signal_handler)
+
+    try:
+        while True:
+            time.sleep(_ONE_DAY_IN_SECONDS)
+    except KeyboardInterrupt:
+        server.stop(0)
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Run the gRPC server.")
+    parser.add_argument(
+        "--addr", default="localhost:50051", help="The address to bind the server to."
+    )
+    args = parser.parse_args()
+
+    serve(args.addr)

backend/python/pocket-tts/install.sh

@@ -0,0 +1,30 @@
+#!/bin/bash
+set -e
+
+backend_dir=$(dirname $0)
+if [ -d $backend_dir/common ]; then
+    source $backend_dir/common/libbackend.sh
+else
+    source $backend_dir/../common/libbackend.sh
+fi
+
+# This is here because the Intel pip index is broken and returns 200 status codes for every package name, it just doesn't return any package links.
+# This makes uv think that the package exists in the Intel pip index, and by default it stops looking at other pip indexes once it finds a match.
+# We need uv to continue falling through to the pypi default index to find optimum[openvino] in the pypi index
+# the --upgrade actually allows us to *downgrade* torch to the version provided in the Intel pip index
+if [ "x${BUILD_PROFILE}" == "xintel" ]; then
+    EXTRA_PIP_INSTALL_FLAGS+=" --upgrade --index-strategy=unsafe-first-match"
+fi
+
+# Use python 3.12 for l4t
+if [ "x${BUILD_PROFILE}" == "xl4t13" ]; then
+  PYTHON_VERSION="3.12"
+  PYTHON_PATCH="12"
+  PY_STANDALONE_TAG="20251120"
+fi
+
+if [ "x${BUILD_PROFILE}" == "xl4t12" ]; then
+    USE_PIP=true
+fi
+
+installRequirements

backend/python/pocket-tts/protogen.sh

@@ -0,0 +1,11 @@
+#!/bin/bash
+set -e
+
+backend_dir=$(dirname $0)
+if [ -d $backend_dir/common ]; then
+    source $backend_dir/common/libbackend.sh
+else
+    source $backend_dir/../common/libbackend.sh
+fi
+
+python3 -m grpc_tools.protoc -I../.. -I./ --python_out=. --grpc_python_out=. backend.proto

backend/python/pocket-tts/requirements-cpu.txt

@@ -0,0 +1,4 @@
+--extra-index-url https://download.pytorch.org/whl/cpu
+pocket-tts
+scipy
+torch

backend/python/pocket-tts/requirements-cublas12.txt

@@ -0,0 +1,4 @@
+--extra-index-url https://download.pytorch.org/whl/cu121
+pocket-tts
+scipy
+torch

backend/python/pocket-tts/requirements-cublas13.txt

@@ -0,0 +1,4 @@
+--extra-index-url https://download.pytorch.org/whl/cu130
+pocket-tts
+scipy
+torch

backend/python/pocket-tts/requirements-hipblas.txt

@@ -0,0 +1,4 @@
+--extra-index-url https://download.pytorch.org/whl/rocm6.3
+pocket-tts
+scipy
+torch==2.7.1+rocm6.3

backend/python/pocket-tts/requirements-intel.txt

@@ -0,0 +1,4 @@
+--extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
+pocket-tts
+scipy
+torch==2.5.1+cxx11.abi

backend/python/pocket-tts/requirements-l4t12.txt

@@ -0,0 +1,4 @@
+--extra-index-url https://pypi.jetson-ai-lab.io/jp6/cu129/
+pocket-tts
+scipy
+torch

backend/python/pocket-tts/requirements-l4t13.txt

@@ -0,0 +1,4 @@
+--extra-index-url https://download.pytorch.org/whl/cu130
+pocket-tts
+scipy
+torch

backend/python/pocket-tts/requirements-mps.txt

@@ -0,0 +1,4 @@
+pocket-tts
+scipy
+torch==2.7.1
+torchvision==0.22.1

backend/python/pocket-tts/requirements.txt

@@ -0,0 +1,4 @@
+grpcio==1.71.0
+protobuf
+certifi
+packaging==24.1

backend/python/pocket-tts/run.sh

@@ -0,0 +1,9 @@
+#!/bin/bash
+backend_dir=$(dirname $0)
+if [ -d $backend_dir/common ]; then
+    source $backend_dir/common/libbackend.sh
+else
+    source $backend_dir/../common/libbackend.sh
+fi
+
+startBackend $@

backend/python/pocket-tts/test.py

@@ -0,0 +1,141 @@
+"""
+A test script to test the gRPC service
+"""
+import unittest
+import subprocess
+import time
+import os
+import tempfile
+import backend_pb2
+import backend_pb2_grpc
+
+import grpc
+
+
+class TestBackendServicer(unittest.TestCase):
+    """
+    TestBackendServicer is the class that tests the gRPC service
+    """
+    def setUp(self):
+        """
+        This method sets up the gRPC service by starting the server
+        """
+        self.service = subprocess.Popen(["python3", "backend.py", "--addr", "localhost:50051"])
+        time.sleep(30)
+
+    def tearDown(self) -> None:
+        """
+        This method tears down the gRPC service by terminating the server
+        """
+        self.service.terminate()
+        self.service.wait()
+
+    def test_server_startup(self):
+        """
+        This method tests if the server starts up successfully
+        """
+        try:
+            self.setUp()
+            with grpc.insecure_channel("localhost:50051") as channel:
+                stub = backend_pb2_grpc.BackendStub(channel)
+                response = stub.Health(backend_pb2.HealthMessage())
+                self.assertEqual(response.message, b'OK')
+        except Exception as err:
+            print(err)
+            self.fail("Server failed to start")
+        finally:
+            self.tearDown()
+
+    def test_load_model(self):
+        """
+        This method tests if the model is loaded successfully
+        """
+        try:
+            self.setUp()
+            with grpc.insecure_channel("localhost:50051") as channel:
+                stub = backend_pb2_grpc.BackendStub(channel)
+                response = stub.LoadModel(backend_pb2.ModelOptions())
+                print(response)
+                self.assertTrue(response.success)
+                self.assertEqual(response.message, "Model loaded successfully")
+        except Exception as err:
+            print(err)
+            self.fail("LoadModel service failed")
+        finally:
+            self.tearDown()
+
+    def test_tts_with_hf_voice(self):
+        """
+        This method tests TTS generation with HuggingFace voice URL
+        """
+        try:
+            self.setUp()
+            with grpc.insecure_channel("localhost:50051") as channel:
+                stub = backend_pb2_grpc.BackendStub(channel)
+                # Load model
+                response = stub.LoadModel(backend_pb2.ModelOptions())
+                self.assertTrue(response.success)
+                
+                # Create temporary output file
+                with tempfile.NamedTemporaryFile(suffix='.wav', delete=False) as tmp_file:
+                    output_path = tmp_file.name
+                
+                # Test TTS with HuggingFace voice URL
+                tts_request = backend_pb2.TTSRequest(
+                    text="Hello world, this is a test.",
+                    dst=output_path,
+                    voice="azelma"
+                )
+                tts_response = stub.TTS(tts_request)
+                self.assertTrue(tts_response.success)
+                
+                # Verify output file exists and is not empty
+                self.assertTrue(os.path.exists(output_path))
+                self.assertGreater(os.path.getsize(output_path), 0)
+                
+                # Cleanup
+                os.unlink(output_path)
+        except Exception as err:
+            print(err)
+            self.fail("TTS service failed")
+        finally:
+            self.tearDown()
+
+    def test_tts_with_default_voice(self):
+        """
+        This method tests TTS generation with default voice (via AudioPath in LoadModel)
+        """
+        try:
+            self.setUp()
+            with grpc.insecure_channel("localhost:50051") as channel:
+                stub = backend_pb2_grpc.BackendStub(channel)
+                # Load model with default voice
+                load_request = backend_pb2.ModelOptions(
+                    Options=["default_voice:azelma"]
+                )
+                response = stub.LoadModel(load_request)
+                self.assertTrue(response.success)
+                
+                # Create temporary output file
+                with tempfile.NamedTemporaryFile(suffix='.wav', delete=False) as tmp_file:
+                    output_path = tmp_file.name
+                
+                # Test TTS without specifying voice (should use default)
+                tts_request = backend_pb2.TTSRequest(
+                    text="Hello world, this is a test.",
+                    dst=output_path
+                )
+                tts_response = stub.TTS(tts_request)
+                self.assertTrue(tts_response.success)
+                
+                # Verify output file exists and is not empty
+                self.assertTrue(os.path.exists(output_path))
+                self.assertGreater(os.path.getsize(output_path), 0)
+                
+                # Cleanup
+                os.unlink(output_path)
+        except Exception as err:
+            print(err)
+            self.fail("TTS service with default voice failed")
+        finally:
+            self.tearDown()

backend/python/pocket-tts/test.sh

@@ -0,0 +1,11 @@
+#!/bin/bash
+set -e
+
+backend_dir=$(dirname $0)
+if [ -d $backend_dir/common ]; then
+    source $backend_dir/common/libbackend.sh
+else
+    source $backend_dir/../common/libbackend.sh
+fi
+
+runUnittests

backend/python/transformers/requirements-cpu.txt

@@ -6,4 +6,4 @@ transformers
 bitsandbytes
 outetts
 sentence-transformers==5.2.0
-protobuf==6.33.2
+protobuf==6.33.4

backend/python/transformers/requirements-cublas12.txt

@@ -6,4 +6,4 @@ transformers
 bitsandbytes
 outetts
 sentence-transformers==5.2.0
-protobuf==6.33.2
+protobuf==6.33.4

backend/python/transformers/requirements-cublas13.txt

@@ -6,4 +6,4 @@ transformers
 bitsandbytes
 outetts
 sentence-transformers==5.2.0
-protobuf==6.33.2
+protobuf==6.33.4

backend/python/transformers/requirements-hipblas.txt

@@ -8,4 +8,4 @@ bitsandbytes
 outetts
 bitsandbytes
 sentence-transformers==5.2.0
-protobuf==6.33.2
+protobuf==6.33.4

backend/python/transformers/requirements-intel.txt

@@ -10,4 +10,4 @@ intel-extension-for-transformers
 bitsandbytes
 outetts
 sentence-transformers==5.2.0
-protobuf==6.33.2
+protobuf==6.33.4

backend/python/transformers/requirements.txt

@@ -1,5 +1,5 @@
 grpcio==1.76.0
-protobuf==6.33.2
+protobuf==6.33.4
 certifi
 setuptools
 scipy==1.15.1

core/cli/run.go

@@ -56,7 +56,7 @@ type RunCMD struct {
 	UseSubtleKeyComparison             bool     `env:"LOCALAI_SUBTLE_KEY_COMPARISON" default:"false" help:"If true, API Key validation comparisons will be performed using constant-time comparisons rather than simple equality. This trades off performance on each request for resiliancy against timing attacks." group:"hardening"`
 	DisableApiKeyRequirementForHttpGet bool     `env:"LOCALAI_DISABLE_API_KEY_REQUIREMENT_FOR_HTTP_GET" default:"false" help:"If true, a valid API key is not required to issue GET requests to portions of the web ui. This should only be enabled in secure testing environments" group:"hardening"`
 	DisableMetricsEndpoint             bool     `env:"LOCALAI_DISABLE_METRICS_ENDPOINT,DISABLE_METRICS_ENDPOINT" default:"false" help:"Disable the /metrics endpoint" group:"api"`
-	HttpGetExemptedEndpoints           []string `env:"LOCALAI_HTTP_GET_EXEMPTED_ENDPOINTS" default:"^/$,^/browse/?$,^/talk/?$,^/p2p/?$,^/chat/?$,^/text2image/?$,^/tts/?$,^/static/.*$,^/swagger.*$" help:"If LOCALAI_DISABLE_API_KEY_REQUIREMENT_FOR_HTTP_GET is overriden to true, this is the list of endpoints to exempt. Only adjust this in case of a security incident or as a result of a personal security posture review" group:"hardening"`
+	HttpGetExemptedEndpoints           []string `env:"LOCALAI_HTTP_GET_EXEMPTED_ENDPOINTS" default:"^/$,^/browse/?$,^/talk/?$,^/p2p/?$,^/chat/?$,^/image/?$,^/text2image/?$,^/tts/?$,^/static/.*$,^/swagger.*$" help:"If LOCALAI_DISABLE_API_KEY_REQUIREMENT_FOR_HTTP_GET is overriden to true, this is the list of endpoints to exempt. Only adjust this in case of a security incident or as a result of a personal security posture review" group:"hardening"`
 	Peer2Peer                          bool     `env:"LOCALAI_P2P,P2P" name:"p2p" default:"false" help:"Enable P2P mode" group:"p2p"`
 	Peer2PeerDHTInterval               int      `env:"LOCALAI_P2P_DHT_INTERVAL,P2P_DHT_INTERVAL" default:"360" name:"p2p-dht-interval" help:"Interval for DHT refresh (used during token generation)" group:"p2p"`
 	Peer2PeerOTPInterval               int      `env:"LOCALAI_P2P_OTP_INTERVAL,P2P_OTP_INTERVAL" default:"9000" name:"p2p-otp-interval" help:"Interval for OTP refresh (used during token generation)" group:"p2p"`
@@ -83,6 +83,7 @@ type RunCMD struct {
 	EnableTracing                      bool     `env:"LOCALAI_ENABLE_TRACING,ENABLE_TRACING" help:"Enable API tracing" group:"api"`
 	TracingMaxItems                    int      `env:"LOCALAI_TRACING_MAX_ITEMS" default:"1024" help:"Maximum number of traces to keep" group:"api"`
 	AgentJobRetentionDays              int      `env:"LOCALAI_AGENT_JOB_RETENTION_DAYS,AGENT_JOB_RETENTION_DAYS" default:"30" help:"Number of days to keep agent job history (default: 30)" group:"api"`
+	OpenResponsesStoreTTL               string   `env:"LOCALAI_OPEN_RESPONSES_STORE_TTL,OPEN_RESPONSES_STORE_TTL" default:"0" help:"TTL for Open Responses store (e.g., 1h, 30m, 0 = no expiration)" group:"api"`
 
 	Version bool
 }
@@ -249,6 +250,15 @@ func (r *RunCMD) Run(ctx *cliContext.Context) error {
 		opts = append(opts, config.WithLRUEvictionRetryInterval(dur))
 	}
 
+	// Handle Open Responses store TTL
+	if r.OpenResponsesStoreTTL != "" && r.OpenResponsesStoreTTL != "0" {
+		dur, err := time.ParseDuration(r.OpenResponsesStoreTTL)
+		if err != nil {
+			return fmt.Errorf("invalid Open Responses store TTL: %w", err)
+		}
+		opts = append(opts, config.WithOpenResponsesStoreTTL(dur))
+	}
+
 	// split ":" to get backend name and the uri
 	for _, v := range r.ExternalGRPCBackends {
 		backend := v[:strings.IndexByte(v, ':')]

core/config/application_config.go

@@ -86,6 +86,8 @@ type ApplicationConfig struct {
 
 	AgentJobRetentionDays int // Default: 30 days
 
+	OpenResponsesStoreTTL time.Duration // TTL for Open Responses store (0 = no expiration)
+
 	PathWithoutAuth []string
 }
 
@@ -467,6 +469,12 @@ func WithAgentJobRetentionDays(days int) AppOption {
 	}
 }
 
+func WithOpenResponsesStoreTTL(ttl time.Duration) AppOption {
+	return func(o *ApplicationConfig) {
+		o.OpenResponsesStoreTTL = ttl
+	}
+}
+
 func WithEnforcedPredownloadScans(enforced bool) AppOption {
 	return func(o *ApplicationConfig) {
 		o.EnforcePredownloadScans = enforced
@@ -594,6 +602,12 @@ func (o *ApplicationConfig) ToRuntimeSettings() RuntimeSettings {
 	} else {
 		lruEvictionRetryInterval = "1s" // default
 	}
+	var openResponsesStoreTTL string
+	if o.OpenResponsesStoreTTL > 0 {
+		openResponsesStoreTTL = o.OpenResponsesStoreTTL.String()
+	} else {
+		openResponsesStoreTTL = "0" // default: no expiration
+	}
 
 	return RuntimeSettings{
 		WatchdogEnabled:          &watchdogEnabled,
@@ -628,6 +642,7 @@ func (o *ApplicationConfig) ToRuntimeSettings() RuntimeSettings {
 		AutoloadBackendGalleries: &autoloadBackendGalleries,
 		ApiKeys:                  &apiKeys,
 		AgentJobRetentionDays:    &agentJobRetentionDays,
+		OpenResponsesStoreTTL:     &openResponsesStoreTTL,
 	}
 }
 
@@ -769,6 +784,14 @@ func (o *ApplicationConfig) ApplyRuntimeSettings(settings *RuntimeSettings) (req
 	if settings.AgentJobRetentionDays != nil {
 		o.AgentJobRetentionDays = *settings.AgentJobRetentionDays
 	}
+	if settings.OpenResponsesStoreTTL != nil {
+		if *settings.OpenResponsesStoreTTL == "0" || *settings.OpenResponsesStoreTTL == "" {
+			o.OpenResponsesStoreTTL = 0 // No expiration
+		} else if dur, err := time.ParseDuration(*settings.OpenResponsesStoreTTL); err == nil {
+			o.OpenResponsesStoreTTL = dur
+		}
+		// This setting doesn't require restart, can be updated dynamically
+	}
 	// Note: ApiKeys requires special handling (merging with startup keys) - handled in caller
 
 	return requireRestart

core/config/model_config.go

@@ -10,6 +10,7 @@ import (
 	"github.com/mudler/LocalAI/core/schema"
 	"github.com/mudler/LocalAI/pkg/downloader"
 	"github.com/mudler/LocalAI/pkg/functions"
+	"github.com/mudler/LocalAI/pkg/reasoning"
 	"github.com/mudler/cogito"
 	"gopkg.in/yaml.v3"
 )
@@ -51,6 +52,7 @@ type ModelConfig struct {
 	ResponseFormatMap                          map[string]interface{} `yaml:"-" json:"-"`
 
 	FunctionsConfig functions.FunctionsConfig `yaml:"function,omitempty" json:"function,omitempty"`
+	ReasoningConfig reasoning.ReasoningConfig `yaml:"reasoning,omitempty" json:"reasoning,omitempty"`
 
 	FeatureFlag FeatureFlag `yaml:"feature_flags,omitempty" json:"feature_flags,omitempty"` // Feature Flag registry. We move fast, and features may break on a per model/backend basis. Registry for (usually temporary) flags that indicate aborting something early.
 	// LLM configs (GPT4ALL, Llama.cpp, ...)

core/config/runtime_settings.go

@@ -60,4 +60,7 @@ type RuntimeSettings struct {
 
 	// Agent settings
 	AgentJobRetentionDays *int `json:"agent_job_retention_days,omitempty"`
+
+	// Open Responses settings
+	OpenResponsesStoreTTL *string `json:"open_responses_store_ttl,omitempty"` // TTL for stored responses (e.g., "1h", "30m", "0" = no expiration)
 }

core/http/app.go

@@ -108,7 +108,15 @@ func API(application *application.Application) (*echo.Echo, error) {
 			req := c.Request()
 			res := c.Response()
 			err := next(c)
-			xlog.Info("HTTP request", "method", req.Method, "path", req.URL.Path, "status", res.Status)
+
+			// Fix for #7989: Reduce log verbosity of Web UI polling
+			// If the path is /api/operations and the request was successful (200),
+			// we log it at DEBUG level (hidden by default) instead of INFO.
+			if req.URL.Path == "/api/operations" && res.Status == 200 {
+				xlog.Debug("HTTP request", "method", req.Method, "path", req.URL.Path, "status", res.Status)
+			} else {
+				xlog.Info("HTTP request", "method", req.Method, "path", req.URL.Path, "status", res.Status)
+			}
 			return err
 		}
 	})
@@ -185,6 +193,8 @@ func API(application *application.Application) (*echo.Echo, error) {
 			corsConfig.AllowOrigins = strings.Split(application.ApplicationConfig().CORSAllowOrigins, ",")
 		}
 		e.Use(middleware.CORSWithConfig(corsConfig))
+	} else {
+		e.Use(middleware.CORS())
 	}
 
 	// CSRF middleware
@@ -206,6 +216,7 @@ func API(application *application.Application) (*echo.Echo, error) {
 	routes.RegisterLocalAIRoutes(e, requestExtractor, application.ModelConfigLoader(), application.ModelLoader(), application.ApplicationConfig(), application.GalleryService(), opcache, application.TemplatesEvaluator(), application)
 	routes.RegisterOpenAIRoutes(e, requestExtractor, application)
 	routes.RegisterAnthropicRoutes(e, requestExtractor, application)
+	routes.RegisterOpenResponsesRoutes(e, requestExtractor, application)
 	if !application.ApplicationConfig().DisableWebUI {
 		routes.RegisterUIAPIRoutes(e, application.ModelConfigLoader(), application.ModelLoader(), application.ApplicationConfig(), application.GalleryService(), opcache, application)
 		routes.RegisterUIRoutes(e, application.ModelConfigLoader(), application.ModelLoader(), application.ApplicationConfig(), application.GalleryService())

core/http/endpoints/localai/settings.go

@@ -11,6 +11,7 @@ import (
 	"github.com/labstack/echo/v4"
 	"github.com/mudler/LocalAI/core/application"
 	"github.com/mudler/LocalAI/core/config"
+	"github.com/mudler/LocalAI/core/http/endpoints/openresponses"
 	"github.com/mudler/LocalAI/core/p2p"
 	"github.com/mudler/LocalAI/core/schema"
 	"github.com/mudler/xlog"
@@ -84,6 +85,16 @@ func UpdateSettingsEndpoint(app *application.Application) echo.HandlerFunc {
 				})
 			}
 		}
+		if settings.OpenResponsesStoreTTL != nil {
+			if *settings.OpenResponsesStoreTTL != "0" && *settings.OpenResponsesStoreTTL != "" {
+				if _, err := time.ParseDuration(*settings.OpenResponsesStoreTTL); err != nil {
+					return c.JSON(http.StatusBadRequest, schema.SettingsResponse{
+						Success: false,
+						Error:   "Invalid open_responses_store_ttl format: " + err.Error(),
+					})
+				}
+			}
+		}
 
 		// Save to file
 		if appConfig.DynamicConfigsDir == "" {
@@ -144,6 +155,22 @@ func UpdateSettingsEndpoint(app *application.Application) echo.HandlerFunc {
 			xlog.Info("Updated LRU eviction retry settings", "maxRetries", maxRetries, "retryInterval", retryInterval)
 		}
 
+		// Update Open Responses store TTL dynamically
+		if settings.OpenResponsesStoreTTL != nil {
+			ttl := time.Duration(0)
+			if *settings.OpenResponsesStoreTTL != "0" && *settings.OpenResponsesStoreTTL != "" {
+				if dur, err := time.ParseDuration(*settings.OpenResponsesStoreTTL); err == nil {
+					ttl = dur
+				} else {
+					xlog.Warn("Invalid Open Responses store TTL format", "ttl", *settings.OpenResponsesStoreTTL, "error", err)
+				}
+			}
+			// Import the store package
+			store := openresponses.GetGlobalStore()
+			store.SetTTL(ttl)
+			xlog.Info("Updated Open Responses store TTL", "ttl", ttl)
+		}
+
 		// Check if agent job retention changed
 		agentJobChanged := settings.AgentJobRetentionDays != nil
 

core/http/endpoints/openai/chat.go

@@ -13,6 +13,7 @@ import (
 	"github.com/mudler/LocalAI/core/http/middleware"
 	"github.com/mudler/LocalAI/core/schema"
 	"github.com/mudler/LocalAI/pkg/functions"
+	"github.com/mudler/LocalAI/pkg/reasoning"
 
 	"github.com/mudler/LocalAI/core/templates"
 	"github.com/mudler/LocalAI/pkg/model"
@@ -43,10 +44,19 @@ func ChatEndpoint(cl *config.ModelConfigLoader, ml *model.ModelLoader, evaluator
 		lastEmittedReasoning := ""
 		lastEmittedCleanedContent := ""
 
+		// Configure reasoning extraction options
+		// Auto-detect if prompt ends with thinking tag
+		// or use explicit config setting
+		thinkingForcedOpen := config.ReasoningConfig.ThinkingForcedOpen || reasoning.DetectThinkingForcedOpen(s)
+
 		_, _, err := ComputeChoices(req, s, config, cl, startupOptions, loader, func(s string, c *[]schema.Choice) {}, func(s string, tokenUsage backend.TokenUsage) bool {
 			accumulatedContent += s
 			// Extract reasoning from accumulated content
-			currentReasoning, cleanedContent := functions.ExtractReasoning(accumulatedContent)
+			opts := []reasoning.Option{}
+			if thinkingForcedOpen {
+				opts = append(opts, reasoning.WithThinkingForcedOpen())
+			}
+			currentReasoning, cleanedContent := reasoning.Extract(accumulatedContent, opts...)
 
 			// Calculate new reasoning delta (what we haven't emitted yet)
 			var reasoningDelta *string
@@ -230,7 +240,13 @@ func ChatEndpoint(cl *config.ModelConfigLoader, ml *model.ModelLoader, evaluator
 			return err
 		}
 		// Extract reasoning before processing tool calls
-		reasoning, cleanedResult := functions.ExtractReasoning(result)
+		// Auto-detect if prompt ends with thinking tag or use explicit config
+		toolsThinkingForcedOpen := config.ReasoningConfig.ThinkingForcedOpen || reasoning.DetectThinkingForcedOpen(prompt)
+		opts := []reasoning.Option{}
+		if toolsThinkingForcedOpen {
+			opts = append(opts, reasoning.WithThinkingForcedOpen())
+		}
+		extractedReasoning, cleanedResult := reasoning.Extract(result, opts...)
 		result = cleanedResult
 
 		textContentToReturn = functions.ParseTextContent(result, config.FunctionsConfig)
@@ -266,8 +282,8 @@ func ChatEndpoint(cl *config.ModelConfigLoader, ml *model.ModelLoader, evaluator
 			}
 
 			var deltaReasoning *string
-			if reasoning != "" {
-				deltaReasoning = &reasoning
+			if extractedReasoning != "" {
+				deltaReasoning = &extractedReasoning
 			}
 			delta := &schema.Message{Content: &result}
 			if deltaReasoning != nil {
@@ -618,17 +634,24 @@ func ChatEndpoint(cl *config.ModelConfigLoader, ml *model.ModelLoader, evaluator
 		// no streaming mode
 		default:
 
+			// Auto-detect if prompt ends with thinking tag for non-streaming mode
+			nonStreamThinkingForcedOpen := config.ReasoningConfig.ThinkingForcedOpen || reasoning.DetectThinkingForcedOpen(predInput)
+
 			tokenCallback := func(s string, c *[]schema.Choice) {
 				// Extract reasoning from the response
-				reasoning, cleanedS := functions.ExtractReasoning(s)
-				s = cleanedS
+				var extractedReasoning string
+				opts := []reasoning.Option{}
+				if nonStreamThinkingForcedOpen {
+					opts = append(opts, reasoning.WithThinkingForcedOpen())
+				}
+				extractedReasoning, s = reasoning.Extract(s, opts...)
 
 				if !shouldUseFn {
 					// no function is called, just reply and use stop as finish reason
 					stopReason := FinishReasonStop
 					message := &schema.Message{Role: "assistant", Content: &s}
-					if reasoning != "" {
-						message.Reasoning = &reasoning
+					if extractedReasoning != "" {
+						message.Reasoning = &extractedReasoning
 					}
 					*c = append(*c, schema.Choice{FinishReason: &stopReason, Index: 0, Message: message})
 					return
@@ -650,8 +673,8 @@ func ChatEndpoint(cl *config.ModelConfigLoader, ml *model.ModelLoader, evaluator
 
 					stopReason := FinishReasonStop
 					message := &schema.Message{Role: "assistant", Content: &result}
-					if reasoning != "" {
-						message.Reasoning = &reasoning
+					if extractedReasoning != "" {
+						message.Reasoning = &extractedReasoning
 					}
 					*c = append(*c, schema.Choice{
 						FinishReason: &stopReason,
@@ -664,8 +687,8 @@ func ChatEndpoint(cl *config.ModelConfigLoader, ml *model.ModelLoader, evaluator
 							Role: "assistant",
 						},
 					}
-					if reasoning != "" {
-						toolChoice.Message.Reasoning = &reasoning
+					if extractedReasoning != "" {
+						toolChoice.Message.Reasoning = &extractedReasoning
 					}
 
 					for _, ss := range results {
@@ -695,8 +718,8 @@ func ChatEndpoint(cl *config.ModelConfigLoader, ml *model.ModelLoader, evaluator
 									"arguments": args,
 								},
 							}
-							if reasoning != "" {
-								message.Reasoning = &reasoning
+							if extractedReasoning != "" {
+								message.Reasoning = &extractedReasoning
 							}
 							*c = append(*c, schema.Choice{
 								FinishReason: &functionCallReason,

core/http/endpoints/openresponses/store.go

@@ -0,0 +1,453 @@
+package openresponses
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"sync"
+	"time"
+
+	"github.com/mudler/LocalAI/core/schema"
+	"github.com/mudler/xlog"
+)
+
+// ResponseStore provides thread-safe storage for Open Responses API responses
+type ResponseStore struct {
+	mu            sync.RWMutex
+	responses     map[string]*StoredResponse
+	ttl           time.Duration // Time-to-live for stored responses (0 = no expiration)
+	cleanupCtx    context.Context
+	cleanupCancel context.CancelFunc
+}
+
+// StreamedEvent represents a buffered SSE event for streaming resume
+type StreamedEvent struct {
+	SequenceNumber int    `json:"sequence_number"`
+	EventType      string `json:"event_type"`
+	Data           []byte `json:"data"` // JSON-serialized event
+}
+
+// StoredResponse contains a complete response with its input request and output items
+type StoredResponse struct {
+	Request   *schema.OpenResponsesRequest
+	Response  *schema.ORResponseResource
+	Items     map[string]*schema.ORItemField // item_id -> item mapping for quick lookup
+	StoredAt  time.Time
+	ExpiresAt *time.Time // nil if no expiration
+
+	// Background execution support
+	CancelFunc    context.CancelFunc // For cancellation of background tasks
+	StreamEvents  []StreamedEvent    // Buffered events for streaming resume
+	StreamEnabled bool               // Was created with stream=true
+	IsBackground  bool               // Was created with background=true
+	EventsChan    chan struct{}      // Signals new events for live subscribers
+	mu            sync.RWMutex       // Protect concurrent access to this response
+}
+
+var (
+	globalStore *ResponseStore
+	storeOnce   sync.Once
+)
+
+// GetGlobalStore returns the singleton response store instance
+func GetGlobalStore() *ResponseStore {
+	storeOnce.Do(func() {
+		globalStore = NewResponseStore(0) // Default: no TTL, will be updated from appConfig
+	})
+	return globalStore
+}
+
+// SetTTL updates the TTL for the store
+// This will affect all new responses stored after this call
+func (s *ResponseStore) SetTTL(ttl time.Duration) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	// Stop existing cleanup loop if running
+	if s.cleanupCancel != nil {
+		s.cleanupCancel()
+		s.cleanupCancel = nil
+		s.cleanupCtx = nil
+	}
+
+	s.ttl = ttl
+
+	// If TTL > 0, start cleanup loop
+	if ttl > 0 {
+		s.cleanupCtx, s.cleanupCancel = context.WithCancel(context.Background())
+		go s.cleanupLoop(s.cleanupCtx)
+	}
+
+	xlog.Debug("Updated Open Responses store TTL", "ttl", ttl, "cleanup_running", ttl > 0)
+}
+
+// NewResponseStore creates a new response store with optional TTL
+// If ttl is 0, responses are stored indefinitely
+func NewResponseStore(ttl time.Duration) *ResponseStore {
+	store := &ResponseStore{
+		responses: make(map[string]*StoredResponse),
+		ttl:       ttl,
+	}
+
+	// Start cleanup goroutine if TTL is set
+	if ttl > 0 {
+		store.cleanupCtx, store.cleanupCancel = context.WithCancel(context.Background())
+		go store.cleanupLoop(store.cleanupCtx)
+	}
+
+	return store
+}
+
+// Store stores a response with its request and items
+func (s *ResponseStore) Store(responseID string, request *schema.OpenResponsesRequest, response *schema.ORResponseResource) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	// Build item index for quick lookup
+	items := make(map[string]*schema.ORItemField)
+	for i := range response.Output {
+		item := &response.Output[i]
+		if item.ID != "" {
+			items[item.ID] = item
+		}
+	}
+
+	stored := &StoredResponse{
+		Request:   request,
+		Response:  response,
+		Items:     items,
+		StoredAt:  time.Now(),
+		ExpiresAt: nil,
+	}
+
+	// Set expiration if TTL is configured
+	if s.ttl > 0 {
+		expiresAt := time.Now().Add(s.ttl)
+		stored.ExpiresAt = &expiresAt
+	}
+
+	s.responses[responseID] = stored
+	xlog.Debug("Stored Open Responses response", "response_id", responseID, "items_count", len(items))
+}
+
+// Get retrieves a stored response by ID
+func (s *ResponseStore) Get(responseID string) (*StoredResponse, error) {
+	s.mu.RLock()
+	defer s.mu.RUnlock()
+
+	stored, exists := s.responses[responseID]
+	if !exists {
+		return nil, fmt.Errorf("response not found: %s", responseID)
+	}
+
+	// Check expiration
+	if stored.ExpiresAt != nil && time.Now().After(*stored.ExpiresAt) {
+		// Expired, but we'll return it anyway and let caller handle cleanup
+		return nil, fmt.Errorf("response expired: %s", responseID)
+	}
+
+	return stored, nil
+}
+
+// GetItem retrieves a specific item from a stored response
+func (s *ResponseStore) GetItem(responseID, itemID string) (*schema.ORItemField, error) {
+	stored, err := s.Get(responseID)
+	if err != nil {
+		return nil, err
+	}
+
+	item, exists := stored.Items[itemID]
+	if !exists {
+		return nil, fmt.Errorf("item not found: %s in response %s", itemID, responseID)
+	}
+
+	return item, nil
+}
+
+// FindItem searches for an item across all stored responses
+// Returns the item and the response ID it was found in
+func (s *ResponseStore) FindItem(itemID string) (*schema.ORItemField, string, error) {
+	s.mu.RLock()
+	defer s.mu.RUnlock()
+
+	now := time.Now()
+	for responseID, stored := range s.responses {
+		// Skip expired responses
+		if stored.ExpiresAt != nil && now.After(*stored.ExpiresAt) {
+			continue
+		}
+
+		if item, exists := stored.Items[itemID]; exists {
+			return item, responseID, nil
+		}
+	}
+
+	return nil, "", fmt.Errorf("item not found in any stored response: %s", itemID)
+}
+
+// Delete removes a response from storage
+func (s *ResponseStore) Delete(responseID string) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	delete(s.responses, responseID)
+	xlog.Debug("Deleted Open Responses response", "response_id", responseID)
+}
+
+// Cleanup removes expired responses
+func (s *ResponseStore) Cleanup() int {
+	if s.ttl == 0 {
+		return 0
+	}
+
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	now := time.Now()
+	count := 0
+	for id, stored := range s.responses {
+		if stored.ExpiresAt != nil && now.After(*stored.ExpiresAt) {
+			delete(s.responses, id)
+			count++
+		}
+	}
+
+	if count > 0 {
+		xlog.Debug("Cleaned up expired Open Responses", "count", count)
+	}
+
+	return count
+}
+
+// cleanupLoop runs periodic cleanup of expired responses
+func (s *ResponseStore) cleanupLoop(ctx context.Context) {
+	if s.ttl == 0 {
+		return
+	}
+
+	ticker := time.NewTicker(s.ttl / 2) // Cleanup at half TTL interval
+	defer ticker.Stop()
+
+	for {
+		select {
+		case <-ctx.Done():
+			xlog.Debug("Stopped Open Responses store cleanup loop")
+			return
+		case <-ticker.C:
+			s.Cleanup()
+		}
+	}
+}
+
+// Count returns the number of stored responses
+func (s *ResponseStore) Count() int {
+	s.mu.RLock()
+	defer s.mu.RUnlock()
+	return len(s.responses)
+}
+
+// StoreBackground stores a background response with cancel function and optional streaming support
+func (s *ResponseStore) StoreBackground(responseID string, request *schema.OpenResponsesRequest, response *schema.ORResponseResource, cancelFunc context.CancelFunc, streamEnabled bool) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	// Build item index for quick lookup
+	items := make(map[string]*schema.ORItemField)
+	for i := range response.Output {
+		item := &response.Output[i]
+		if item.ID != "" {
+			items[item.ID] = item
+		}
+	}
+
+	stored := &StoredResponse{
+		Request:       request,
+		Response:      response,
+		Items:         items,
+		StoredAt:      time.Now(),
+		ExpiresAt:     nil,
+		CancelFunc:    cancelFunc,
+		StreamEvents:  []StreamedEvent{},
+		StreamEnabled: streamEnabled,
+		IsBackground:  true,
+		EventsChan:    make(chan struct{}, 100), // Buffered channel for event notifications
+	}
+
+	// Set expiration if TTL is configured
+	if s.ttl > 0 {
+		expiresAt := time.Now().Add(s.ttl)
+		stored.ExpiresAt = &expiresAt
+	}
+
+	s.responses[responseID] = stored
+	xlog.Debug("Stored background Open Responses response", "response_id", responseID, "stream_enabled", streamEnabled)
+}
+
+// UpdateStatus updates the status of a stored response
+func (s *ResponseStore) UpdateStatus(responseID string, status string, completedAt *int64) error {
+	s.mu.RLock()
+	stored, exists := s.responses[responseID]
+	s.mu.RUnlock()
+
+	if !exists {
+		return fmt.Errorf("response not found: %s", responseID)
+	}
+
+	stored.mu.Lock()
+	defer stored.mu.Unlock()
+
+	stored.Response.Status = status
+	stored.Response.CompletedAt = completedAt
+
+	xlog.Debug("Updated response status", "response_id", responseID, "status", status)
+	return nil
+}
+
+// UpdateResponse updates the entire response object for a stored response
+func (s *ResponseStore) UpdateResponse(responseID string, response *schema.ORResponseResource) error {
+	s.mu.RLock()
+	stored, exists := s.responses[responseID]
+	s.mu.RUnlock()
+
+	if !exists {
+		return fmt.Errorf("response not found: %s", responseID)
+	}
+
+	stored.mu.Lock()
+	defer stored.mu.Unlock()
+
+	// Rebuild item index
+	items := make(map[string]*schema.ORItemField)
+	for i := range response.Output {
+		item := &response.Output[i]
+		if item.ID != "" {
+			items[item.ID] = item
+		}
+	}
+
+	stored.Response = response
+	stored.Items = items
+
+	xlog.Debug("Updated response", "response_id", responseID, "status", response.Status, "items_count", len(items))
+	return nil
+}
+
+// AppendEvent appends a streaming event to the buffer for resume support
+func (s *ResponseStore) AppendEvent(responseID string, event *schema.ORStreamEvent) error {
+	s.mu.RLock()
+	stored, exists := s.responses[responseID]
+	s.mu.RUnlock()
+
+	if !exists {
+		return fmt.Errorf("response not found: %s", responseID)
+	}
+
+	// Serialize the event
+	data, err := json.Marshal(event)
+	if err != nil {
+		return fmt.Errorf("failed to marshal event: %w", err)
+	}
+
+	stored.mu.Lock()
+	stored.StreamEvents = append(stored.StreamEvents, StreamedEvent{
+		SequenceNumber: event.SequenceNumber,
+		EventType:      event.Type,
+		Data:           data,
+	})
+	stored.mu.Unlock()
+
+	// Notify any subscribers of new event
+	select {
+	case stored.EventsChan <- struct{}{}:
+	default:
+		// Channel full, subscribers will catch up
+	}
+
+	return nil
+}
+
+// GetEventsAfter returns all events with sequence number greater than startingAfter
+func (s *ResponseStore) GetEventsAfter(responseID string, startingAfter int) ([]StreamedEvent, error) {
+	s.mu.RLock()
+	stored, exists := s.responses[responseID]
+	s.mu.RUnlock()
+
+	if !exists {
+		return nil, fmt.Errorf("response not found: %s", responseID)
+	}
+
+	stored.mu.RLock()
+	defer stored.mu.RUnlock()
+
+	var result []StreamedEvent
+	for _, event := range stored.StreamEvents {
+		if event.SequenceNumber > startingAfter {
+			result = append(result, event)
+		}
+	}
+
+	return result, nil
+}
+
+// Cancel cancels a background response if it's still in progress
+func (s *ResponseStore) Cancel(responseID string) (*schema.ORResponseResource, error) {
+	s.mu.RLock()
+	stored, exists := s.responses[responseID]
+	s.mu.RUnlock()
+
+	if !exists {
+		return nil, fmt.Errorf("response not found: %s", responseID)
+	}
+
+	stored.mu.Lock()
+	defer stored.mu.Unlock()
+
+	// If already in a terminal state, just return the response (idempotent)
+	status := stored.Response.Status
+	if status == schema.ORStatusCompleted || status == schema.ORStatusFailed ||
+		status == schema.ORStatusIncomplete || status == schema.ORStatusCancelled {
+		xlog.Debug("Response already in terminal state", "response_id", responseID, "status", status)
+		return stored.Response, nil
+	}
+
+	// Cancel the context if available
+	if stored.CancelFunc != nil {
+		stored.CancelFunc()
+		xlog.Debug("Cancelled background response", "response_id", responseID)
+	}
+
+	// Update status to cancelled
+	now := time.Now().Unix()
+	stored.Response.Status = schema.ORStatusCancelled
+	stored.Response.CompletedAt = &now
+
+	return stored.Response, nil
+}
+
+// GetEventsChan returns the events notification channel for a response
+func (s *ResponseStore) GetEventsChan(responseID string) (chan struct{}, error) {
+	s.mu.RLock()
+	stored, exists := s.responses[responseID]
+	s.mu.RUnlock()
+
+	if !exists {
+		return nil, fmt.Errorf("response not found: %s", responseID)
+	}
+
+	return stored.EventsChan, nil
+}
+
+// IsStreamEnabled checks if a response was created with streaming enabled
+func (s *ResponseStore) IsStreamEnabled(responseID string) (bool, error) {
+	s.mu.RLock()
+	stored, exists := s.responses[responseID]
+	s.mu.RUnlock()
+
+	if !exists {
+		return false, fmt.Errorf("response not found: %s", responseID)
+	}
+
+	stored.mu.RLock()
+	defer stored.mu.RUnlock()
+
+	return stored.StreamEnabled, nil
+}

core/http/endpoints/openresponses/store_suite_test.go

@@ -0,0 +1,13 @@
+package openresponses
+
+import (
+	"testing"
+
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+)
+
+func TestStore(t *testing.T) {
+	RegisterFailHandler(Fail)
+	RunSpecs(t, "ResponseStore Suite")
+}

core/http/endpoints/openresponses/store_test.go

@@ -0,0 +1,626 @@
+package openresponses
+
+import (
+	"context"
+	"fmt"
+	"time"
+
+	"github.com/mudler/LocalAI/core/schema"
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+)
+
+var _ = Describe("ResponseStore", func() {
+	var store *ResponseStore
+
+	BeforeEach(func() {
+		store = NewResponseStore(0) // No TTL for most tests
+	})
+
+	AfterEach(func() {
+		// Clean up
+	})
+
+	Describe("Store and Get", func() {
+		It("should store and retrieve a response", func() {
+			responseID := "resp_test123"
+			request := &schema.OpenResponsesRequest{
+				Model: "test-model",
+				Input: "Hello",
+			}
+			response := &schema.ORResponseResource{
+				ID:        responseID,
+				Object:    "response",
+				CreatedAt: time.Now().Unix(),
+				Status:    "completed",
+				Model:     "test-model",
+				Output: []schema.ORItemField{
+					{
+						Type:   "message",
+						ID:     "msg_123",
+						Status: "completed",
+						Role:   "assistant",
+						Content: []schema.ORContentPart{{
+							Type:        "output_text",
+							Text:        "Hello, world!",
+							Annotations: []schema.ORAnnotation{},
+							Logprobs:    []schema.ORLogProb{},
+						}},
+					},
+				},
+			}
+
+			store.Store(responseID, request, response)
+
+			stored, err := store.Get(responseID)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(stored).ToNot(BeNil())
+			Expect(stored.Response.ID).To(Equal(responseID))
+			Expect(stored.Request.Model).To(Equal("test-model"))
+			Expect(len(stored.Items)).To(Equal(1))
+			Expect(stored.Items["msg_123"]).ToNot(BeNil())
+			Expect(stored.Items["msg_123"].ID).To(Equal("msg_123"))
+		})
+
+		It("should return error for non-existent response", func() {
+			_, err := store.Get("nonexistent")
+			Expect(err).To(HaveOccurred())
+			Expect(err.Error()).To(ContainSubstring("not found"))
+		})
+
+		It("should index all items by ID", func() {
+			responseID := "resp_test456"
+			request := &schema.OpenResponsesRequest{
+				Model: "test-model",
+				Input: "Test",
+			}
+			response := &schema.ORResponseResource{
+				ID:     responseID,
+				Object: "response",
+				Output: []schema.ORItemField{
+					{
+						Type:   "message",
+						ID:     "msg_1",
+						Status: "completed",
+						Role:   "assistant",
+					},
+					{
+						Type:      "function_call",
+						ID:        "fc_1",
+						Status:    "completed",
+						CallID:    "fc_1",
+						Name:      "test_function",
+						Arguments: `{"arg": "value"}`,
+					},
+					{
+						Type:   "message",
+						ID:     "msg_2",
+						Status: "completed",
+						Role:   "assistant",
+					},
+				},
+			}
+
+			store.Store(responseID, request, response)
+
+			stored, err := store.Get(responseID)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(len(stored.Items)).To(Equal(3))
+			Expect(stored.Items["msg_1"]).ToNot(BeNil())
+			Expect(stored.Items["fc_1"]).ToNot(BeNil())
+			Expect(stored.Items["msg_2"]).ToNot(BeNil())
+		})
+
+		It("should handle items without IDs", func() {
+			responseID := "resp_test789"
+			request := &schema.OpenResponsesRequest{
+				Model: "test-model",
+				Input: "Test",
+			}
+			response := &schema.ORResponseResource{
+				ID:     responseID,
+				Object: "response",
+				Output: []schema.ORItemField{
+					{
+						Type:   "message",
+						ID:     "", // No ID
+						Status: "completed",
+						Role:   "assistant",
+					},
+					{
+						Type:   "message",
+						ID:     "msg_with_id",
+						Status: "completed",
+						Role:   "assistant",
+					},
+				},
+			}
+
+			store.Store(responseID, request, response)
+
+			stored, err := store.Get(responseID)
+			Expect(err).ToNot(HaveOccurred())
+			// Only items with IDs are indexed
+			Expect(len(stored.Items)).To(Equal(1))
+			Expect(stored.Items["msg_with_id"]).ToNot(BeNil())
+		})
+	})
+
+	Describe("GetItem", func() {
+		It("should retrieve a specific item by ID", func() {
+			responseID := "resp_item_test"
+			itemID := "msg_specific"
+			request := &schema.OpenResponsesRequest{
+				Model: "test-model",
+				Input: "Test",
+			}
+			response := &schema.ORResponseResource{
+				ID:     responseID,
+				Object: "response",
+				Output: []schema.ORItemField{
+					{
+						Type:   "message",
+						ID:     itemID,
+						Status: "completed",
+						Role:   "assistant",
+						Content: []schema.ORContentPart{{
+							Type:        "output_text",
+							Text:        "Specific message",
+							Annotations: []schema.ORAnnotation{},
+							Logprobs:    []schema.ORLogProb{},
+						}},
+					},
+				},
+			}
+
+			store.Store(responseID, request, response)
+
+			item, err := store.GetItem(responseID, itemID)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(item).ToNot(BeNil())
+			Expect(item.ID).To(Equal(itemID))
+			Expect(item.Type).To(Equal("message"))
+		})
+
+		It("should return error for non-existent item", func() {
+			responseID := "resp_item_test2"
+			request := &schema.OpenResponsesRequest{
+				Model: "test-model",
+				Input: "Test",
+			}
+			response := &schema.ORResponseResource{
+				ID:     responseID,
+				Object: "response",
+				Output: []schema.ORItemField{
+					{
+						Type:   "message",
+						ID:     "msg_existing",
+						Status: "completed",
+					},
+				},
+			}
+
+			store.Store(responseID, request, response)
+
+			_, err := store.GetItem(responseID, "nonexistent_item")
+			Expect(err).To(HaveOccurred())
+			Expect(err.Error()).To(ContainSubstring("item not found"))
+		})
+
+		It("should return error for non-existent response when getting item", func() {
+			_, err := store.GetItem("nonexistent_response", "any_item")
+			Expect(err).To(HaveOccurred())
+			Expect(err.Error()).To(ContainSubstring("response not found"))
+		})
+	})
+
+	Describe("FindItem", func() {
+		It("should find an item across all stored responses", func() {
+			// Store first response
+			responseID1 := "resp_find_1"
+			itemID1 := "msg_find_1"
+			store.Store(responseID1, &schema.OpenResponsesRequest{Model: "test"}, &schema.ORResponseResource{
+				ID:     responseID1,
+				Object: "response",
+				Output: []schema.ORItemField{
+					{Type: "message", ID: itemID1, Status: "completed"},
+				},
+			})
+
+			// Store second response
+			responseID2 := "resp_find_2"
+			itemID2 := "msg_find_2"
+			store.Store(responseID2, &schema.OpenResponsesRequest{Model: "test"}, &schema.ORResponseResource{
+				ID:     responseID2,
+				Object: "response",
+				Output: []schema.ORItemField{
+					{Type: "message", ID: itemID2, Status: "completed"},
+				},
+			})
+
+			// Find item from first response
+			item, foundResponseID, err := store.FindItem(itemID1)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(item).ToNot(BeNil())
+			Expect(item.ID).To(Equal(itemID1))
+			Expect(foundResponseID).To(Equal(responseID1))
+
+			// Find item from second response
+			item, foundResponseID, err = store.FindItem(itemID2)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(item).ToNot(BeNil())
+			Expect(item.ID).To(Equal(itemID2))
+			Expect(foundResponseID).To(Equal(responseID2))
+		})
+
+		It("should return error when item not found in any response", func() {
+			_, _, err := store.FindItem("nonexistent_item")
+			Expect(err).To(HaveOccurred())
+			Expect(err.Error()).To(ContainSubstring("item not found in any stored response"))
+		})
+	})
+
+	Describe("Delete", func() {
+		It("should delete a stored response", func() {
+			responseID := "resp_delete_test"
+			request := &schema.OpenResponsesRequest{Model: "test"}
+			response := &schema.ORResponseResource{
+				ID:     responseID,
+				Object: "response",
+			}
+
+			store.Store(responseID, request, response)
+			Expect(store.Count()).To(Equal(1))
+
+			store.Delete(responseID)
+			Expect(store.Count()).To(Equal(0))
+
+			_, err := store.Get(responseID)
+			Expect(err).To(HaveOccurred())
+		})
+
+		It("should handle deleting non-existent response gracefully", func() {
+			// Should not panic
+			store.Delete("nonexistent")
+			Expect(store.Count()).To(Equal(0))
+		})
+	})
+
+	Describe("Count", func() {
+		It("should return correct count of stored responses", func() {
+			Expect(store.Count()).To(Equal(0))
+
+			store.Store("resp_1", &schema.OpenResponsesRequest{Model: "test"}, &schema.ORResponseResource{ID: "resp_1", Object: "response"})
+			Expect(store.Count()).To(Equal(1))
+
+			store.Store("resp_2", &schema.OpenResponsesRequest{Model: "test"}, &schema.ORResponseResource{ID: "resp_2", Object: "response"})
+			Expect(store.Count()).To(Equal(2))
+
+			store.Delete("resp_1")
+			Expect(store.Count()).To(Equal(1))
+		})
+	})
+
+	Describe("TTL and Expiration", func() {
+		It("should set expiration when TTL is configured", func() {
+			ttlStore := NewResponseStore(100 * time.Millisecond)
+			responseID := "resp_ttl_test"
+			request := &schema.OpenResponsesRequest{Model: "test"}
+			response := &schema.ORResponseResource{ID: responseID, Object: "response"}
+
+			ttlStore.Store(responseID, request, response)
+
+			stored, err := ttlStore.Get(responseID)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(stored.ExpiresAt).ToNot(BeNil())
+			Expect(stored.ExpiresAt.After(time.Now())).To(BeTrue())
+		})
+
+		It("should not set expiration when TTL is 0", func() {
+			responseID := "resp_no_ttl"
+			request := &schema.OpenResponsesRequest{Model: "test"}
+			response := &schema.ORResponseResource{ID: responseID, Object: "response"}
+
+			store.Store(responseID, request, response)
+
+			stored, err := store.Get(responseID)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(stored.ExpiresAt).To(BeNil())
+		})
+
+		It("should clean up expired responses", func() {
+			ttlStore := NewResponseStore(50 * time.Millisecond)
+			responseID := "resp_expire_test"
+			request := &schema.OpenResponsesRequest{Model: "test"}
+			response := &schema.ORResponseResource{ID: responseID, Object: "response"}
+
+			ttlStore.Store(responseID, request, response)
+			Expect(ttlStore.Count()).To(Equal(1))
+
+			// Wait for expiration (longer than TTL and cleanup interval)
+			time.Sleep(150 * time.Millisecond)
+
+			// Cleanup should remove expired response (may have already been cleaned by goroutine)
+			count := ttlStore.Cleanup()
+			// Count might be 0 if cleanup goroutine already ran, or 1 if we're first
+			Expect(count).To(BeNumerically(">=", 0))
+			Expect(ttlStore.Count()).To(Equal(0))
+
+			_, err := ttlStore.Get(responseID)
+			Expect(err).To(HaveOccurred())
+		})
+
+		It("should return error for expired response", func() {
+			ttlStore := NewResponseStore(50 * time.Millisecond)
+			responseID := "resp_expire_error"
+			request := &schema.OpenResponsesRequest{Model: "test"}
+			response := &schema.ORResponseResource{ID: responseID, Object: "response"}
+
+			ttlStore.Store(responseID, request, response)
+
+			// Wait for expiration (but not long enough for cleanup goroutine to remove it)
+			time.Sleep(75 * time.Millisecond)
+
+			// Try to get before cleanup goroutine removes it
+			_, err := ttlStore.Get(responseID)
+			// Error could be "expired" or "not found" (if cleanup already ran)
+			Expect(err).To(HaveOccurred())
+			// Either error message is acceptable
+			errMsg := err.Error()
+			Expect(errMsg).To(Or(ContainSubstring("expired"), ContainSubstring("not found")))
+		})
+	})
+
+	Describe("Thread Safety", func() {
+		It("should handle concurrent stores and gets", func() {
+			// This is a basic concurrency test
+			done := make(chan bool, 10)
+			for i := 0; i < 10; i++ {
+				go func(id int) {
+					responseID := fmt.Sprintf("resp_concurrent_%d", id)
+					request := &schema.OpenResponsesRequest{Model: "test"}
+					response := &schema.ORResponseResource{
+						ID:     responseID,
+						Object: "response",
+						Output: []schema.ORItemField{
+							{Type: "message", ID: fmt.Sprintf("msg_%d", id), Status: "completed"},
+						},
+					}
+					store.Store(responseID, request, response)
+
+					// Retrieve immediately
+					stored, err := store.Get(responseID)
+					Expect(err).ToNot(HaveOccurred())
+					Expect(stored).ToNot(BeNil())
+					done <- true
+				}(i)
+			}
+
+			// Wait for all goroutines
+			for i := 0; i < 10; i++ {
+				<-done
+			}
+
+			Expect(store.Count()).To(Equal(10))
+		})
+	})
+
+	Describe("GetGlobalStore", func() {
+		It("should return singleton instance", func() {
+			store1 := GetGlobalStore()
+			store2 := GetGlobalStore()
+			Expect(store1).To(Equal(store2))
+		})
+
+		It("should persist data across GetGlobalStore calls", func() {
+			globalStore := GetGlobalStore()
+			responseID := "resp_global_test"
+			request := &schema.OpenResponsesRequest{Model: "test"}
+			response := &schema.ORResponseResource{ID: responseID, Object: "response"}
+
+			globalStore.Store(responseID, request, response)
+
+			// Get store again
+			globalStore2 := GetGlobalStore()
+			stored, err := globalStore2.Get(responseID)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(stored).ToNot(BeNil())
+		})
+	})
+
+	Describe("Background Mode Support", func() {
+		It("should store background response with cancel function", func() {
+			responseID := "resp_bg_test"
+			request := &schema.OpenResponsesRequest{Model: "test"}
+			response := &schema.ORResponseResource{
+				ID:     responseID,
+				Object: "response",
+				Status: schema.ORStatusQueued,
+			}
+
+			_, cancel := context.WithCancel(context.Background())
+			defer cancel()
+
+			store.StoreBackground(responseID, request, response, cancel, true)
+
+			stored, err := store.Get(responseID)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(stored).ToNot(BeNil())
+			Expect(stored.IsBackground).To(BeTrue())
+			Expect(stored.StreamEnabled).To(BeTrue())
+			Expect(stored.CancelFunc).ToNot(BeNil())
+		})
+
+		It("should update status of stored response", func() {
+			responseID := "resp_status_test"
+			request := &schema.OpenResponsesRequest{Model: "test"}
+			response := &schema.ORResponseResource{
+				ID:     responseID,
+				Object: "response",
+				Status: schema.ORStatusQueued,
+			}
+
+			store.Store(responseID, request, response)
+
+			err := store.UpdateStatus(responseID, schema.ORStatusInProgress, nil)
+			Expect(err).ToNot(HaveOccurred())
+
+			stored, err := store.Get(responseID)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(stored.Response.Status).To(Equal(schema.ORStatusInProgress))
+		})
+
+		It("should append and retrieve streaming events", func() {
+			responseID := "resp_events_test"
+			request := &schema.OpenResponsesRequest{Model: "test"}
+			response := &schema.ORResponseResource{
+				ID:     responseID,
+				Object: "response",
+				Status: schema.ORStatusInProgress,
+			}
+
+			_, cancel := context.WithCancel(context.Background())
+			defer cancel()
+
+			store.StoreBackground(responseID, request, response, cancel, true)
+
+			// Append events
+			event1 := &schema.ORStreamEvent{
+				Type:           "response.created",
+				SequenceNumber: 0,
+			}
+			event2 := &schema.ORStreamEvent{
+				Type:           "response.in_progress",
+				SequenceNumber: 1,
+			}
+			event3 := &schema.ORStreamEvent{
+				Type:           "response.output_text.delta",
+				SequenceNumber: 2,
+			}
+
+			err := store.AppendEvent(responseID, event1)
+			Expect(err).ToNot(HaveOccurred())
+			err = store.AppendEvent(responseID, event2)
+			Expect(err).ToNot(HaveOccurred())
+			err = store.AppendEvent(responseID, event3)
+			Expect(err).ToNot(HaveOccurred())
+
+			// Get all events after -1 (all events)
+			events, err := store.GetEventsAfter(responseID, -1)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(events).To(HaveLen(3))
+
+			// Get events after sequence 1
+			events, err = store.GetEventsAfter(responseID, 1)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(events).To(HaveLen(1))
+			Expect(events[0].SequenceNumber).To(Equal(2))
+		})
+
+		It("should cancel an in-progress response", func() {
+			responseID := "resp_cancel_test"
+			request := &schema.OpenResponsesRequest{Model: "test"}
+			response := &schema.ORResponseResource{
+				ID:     responseID,
+				Object: "response",
+				Status: schema.ORStatusInProgress,
+			}
+
+			_, cancel := context.WithCancel(context.Background())
+			defer cancel()
+
+			store.StoreBackground(responseID, request, response, cancel, false)
+
+			// Cancel the response
+			cancelledResponse, err := store.Cancel(responseID)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(cancelledResponse.Status).To(Equal(schema.ORStatusCancelled))
+			Expect(cancelledResponse.CompletedAt).ToNot(BeNil())
+		})
+
+		It("should be idempotent when cancelling already completed response", func() {
+			responseID := "resp_idempotent_cancel"
+			request := &schema.OpenResponsesRequest{Model: "test"}
+			completedAt := time.Now().Unix()
+			response := &schema.ORResponseResource{
+				ID:          responseID,
+				Object:      "response",
+				Status:      schema.ORStatusCompleted,
+				CompletedAt: &completedAt,
+			}
+
+			store.Store(responseID, request, response)
+
+			// Try to cancel a completed response
+			cancelledResponse, err := store.Cancel(responseID)
+			Expect(err).ToNot(HaveOccurred())
+			// Status should remain completed (not changed to cancelled)
+			Expect(cancelledResponse.Status).To(Equal(schema.ORStatusCompleted))
+		})
+
+		It("should check if streaming is enabled", func() {
+			responseID := "resp_stream_check"
+			request := &schema.OpenResponsesRequest{Model: "test"}
+			response := &schema.ORResponseResource{
+				ID:     responseID,
+				Object: "response",
+				Status: schema.ORStatusQueued,
+			}
+
+			_, cancel := context.WithCancel(context.Background())
+			defer cancel()
+
+			store.StoreBackground(responseID, request, response, cancel, true)
+
+			enabled, err := store.IsStreamEnabled(responseID)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(enabled).To(BeTrue())
+
+			// Store another without streaming
+			responseID2 := "resp_no_stream"
+			store.StoreBackground(responseID2, request, response, cancel, false)
+
+			enabled2, err := store.IsStreamEnabled(responseID2)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(enabled2).To(BeFalse())
+		})
+
+		It("should notify subscribers of new events", func() {
+			responseID := "resp_events_chan"
+			request := &schema.OpenResponsesRequest{Model: "test"}
+			response := &schema.ORResponseResource{
+				ID:     responseID,
+				Object: "response",
+				Status: schema.ORStatusInProgress,
+			}
+
+			_, cancel := context.WithCancel(context.Background())
+			defer cancel()
+
+			store.StoreBackground(responseID, request, response, cancel, true)
+
+			eventsChan, err := store.GetEventsChan(responseID)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(eventsChan).ToNot(BeNil())
+
+			// Append an event
+			event := &schema.ORStreamEvent{
+				Type:           "response.output_text.delta",
+				SequenceNumber: 0,
+			}
+
+			go func() {
+				time.Sleep(10 * time.Millisecond)
+				store.AppendEvent(responseID, event)
+			}()
+
+			// Wait for notification
+			select {
+			case <-eventsChan:
+				// Event received
+			case <-time.After(1 * time.Second):
+				Fail("Timeout waiting for event notification")
+			}
+		})
+	})
+})

core/http/http_suite_test.go

@@ -1,13 +1,33 @@
 package http_test
 
 import (
+	"os"
+	"path/filepath"
 	"testing"
 
 	. "github.com/onsi/ginkgo/v2"
 	. "github.com/onsi/gomega"
 )
 
+var (
+	tmpdir   string
+	modelDir string
+)
+
 func TestLocalAI(t *testing.T) {
 	RegisterFailHandler(Fail)
+
+	var err error
+	tmpdir, err = os.MkdirTemp("", "")
+	Expect(err).ToNot(HaveOccurred())
+	modelDir = filepath.Join(tmpdir, "models")
+	err = os.Mkdir(modelDir, 0750)
+	Expect(err).ToNot(HaveOccurred())
+
+	AfterSuite(func() {
+		err := os.RemoveAll(tmpdir)
+		Expect(err).ToNot(HaveOccurred())
+	})
+
 	RunSpecs(t, "LocalAI HTTP test suite")
 }

core/http/middleware/request.go

@@ -484,3 +484,103 @@ func mergeOpenAIRequestAndModelConfig(config *config.ModelConfig, input *schema.
 	}
 	return fmt.Errorf("unable to validate configuration after merging")
 }
+
+func (re *RequestExtractor) SetOpenResponsesRequest(c echo.Context) error {
+	input, ok := c.Get(CONTEXT_LOCALS_KEY_LOCALAI_REQUEST).(*schema.OpenResponsesRequest)
+	if !ok || input.Model == "" {
+		return echo.ErrBadRequest
+	}
+
+	cfg, ok := c.Get(CONTEXT_LOCALS_KEY_MODEL_CONFIG).(*config.ModelConfig)
+	if !ok || cfg == nil {
+		return echo.ErrBadRequest
+	}
+
+	// Extract or generate the correlation ID (Open Responses uses x-request-id)
+	correlationID := c.Request().Header.Get("x-request-id")
+	if correlationID == "" {
+		correlationID = uuid.New().String()
+	}
+	c.Response().Header().Set("x-request-id", correlationID)
+
+	// Use the request context directly - Echo properly supports context cancellation!
+	reqCtx := c.Request().Context()
+	c1, cancel := context.WithCancel(re.applicationConfig.Context)
+
+	// Cancel when request context is cancelled (client disconnects)
+	go func() {
+		select {
+		case <-reqCtx.Done():
+			cancel()
+		case <-c1.Done():
+			// Already cancelled
+		}
+	}()
+
+	// Add the correlation ID to the new context
+	ctxWithCorrelationID := context.WithValue(c1, CorrelationIDKey, correlationID)
+
+	input.Context = ctxWithCorrelationID
+	input.Cancel = cancel
+
+	err := mergeOpenResponsesRequestAndModelConfig(cfg, input)
+	if err != nil {
+		return err
+	}
+
+	if cfg.Model == "" {
+		xlog.Debug("replacing empty cfg.Model with input value", "input.Model", input.Model)
+		cfg.Model = input.Model
+	}
+
+	c.Set(CONTEXT_LOCALS_KEY_LOCALAI_REQUEST, input)
+	c.Set(CONTEXT_LOCALS_KEY_MODEL_CONFIG, cfg)
+
+	return nil
+}
+
+func mergeOpenResponsesRequestAndModelConfig(config *config.ModelConfig, input *schema.OpenResponsesRequest) error {
+	// Temperature
+	if input.Temperature != nil {
+		config.Temperature = input.Temperature
+	}
+
+	// TopP
+	if input.TopP != nil {
+		config.TopP = input.TopP
+	}
+
+	// MaxOutputTokens -> Maxtokens
+	if input.MaxOutputTokens != nil {
+		config.Maxtokens = input.MaxOutputTokens
+	}
+
+	// Convert tools to functions - this will be handled in the endpoint handler
+	// We just validate that tools are present if needed
+
+	// Handle tool_choice
+	if input.ToolChoice != nil {
+		switch tc := input.ToolChoice.(type) {
+		case string:
+			// "auto", "required", or "none"
+			if tc == "required" {
+				config.SetFunctionCallString("required")
+			} else if tc == "none" {
+				// Don't use tools - handled in endpoint
+			}
+			// "auto" is default - let model decide
+		case map[string]interface{}:
+			// Specific tool: {type:"function", name:"..."}
+			if tcType, ok := tc["type"].(string); ok && tcType == "function" {
+				if name, ok := tc["name"].(string); ok {
+					config.SetFunctionCallString(name)
+				}
+			}
+		}
+	}
+
+	if valid, _ := config.Validate(); valid {
+		return nil
+	}
+	return fmt.Errorf("unable to validate configuration after merging")
+}

core/http/routes/openresponses.go

@@ -0,0 +1,58 @@
+package routes
+
+import (
+	"github.com/labstack/echo/v4"
+	"github.com/mudler/LocalAI/core/application"
+	"github.com/mudler/LocalAI/core/config"
+	"github.com/mudler/LocalAI/core/http/endpoints/openresponses"
+	"github.com/mudler/LocalAI/core/http/middleware"
+	"github.com/mudler/LocalAI/core/schema"
+)
+
+func RegisterOpenResponsesRoutes(app *echo.Echo,
+	re *middleware.RequestExtractor,
+	application *application.Application) {
+
+	// Open Responses API endpoint
+	responsesHandler := openresponses.ResponsesEndpoint(
+		application.ModelConfigLoader(),
+		application.ModelLoader(),
+		application.TemplatesEvaluator(),
+		application.ApplicationConfig(),
+	)
+
+	responsesMiddleware := []echo.MiddlewareFunc{
+		middleware.TraceMiddleware(application),
+		re.BuildFilteredFirstAvailableDefaultModel(config.BuildUsecaseFilterFn(config.FLAG_CHAT)),
+		re.SetModelAndConfig(func() schema.LocalAIRequest { return new(schema.OpenResponsesRequest) }),
+		setOpenResponsesRequestContext(re),
+	}
+
+	// Main Open Responses endpoint
+	app.POST("/v1/responses", responsesHandler, responsesMiddleware...)
+
+	// Also support without version prefix for compatibility
+	app.POST("/responses", responsesHandler, responsesMiddleware...)
+
+	// GET /responses/:id - Retrieve a response (for polling background requests)
+	getResponseHandler := openresponses.GetResponseEndpoint()
+	app.GET("/v1/responses/:id", getResponseHandler, middleware.TraceMiddleware(application))
+	app.GET("/responses/:id", getResponseHandler, middleware.TraceMiddleware(application))
+
+	// POST /responses/:id/cancel - Cancel a background response
+	cancelResponseHandler := openresponses.CancelResponseEndpoint()
+	app.POST("/v1/responses/:id/cancel", cancelResponseHandler, middleware.TraceMiddleware(application))
+	app.POST("/responses/:id/cancel", cancelResponseHandler, middleware.TraceMiddleware(application))
+}
+
+// setOpenResponsesRequestContext sets up the context and cancel function for Open Responses requests
+func setOpenResponsesRequestContext(re *middleware.RequestExtractor) echo.MiddlewareFunc {
+	return func(next echo.HandlerFunc) echo.HandlerFunc {
+		return func(c echo.Context) error {
+			if err := re.SetOpenResponsesRequest(c); err != nil {
+				return err
+			}
+			return next(c)
+		}
+	}
+}

core/http/routes/ui.go

@@ -219,7 +219,7 @@ func RegisterUIRoutes(app *echo.Echo,
 		return c.Render(200, "views/chat", summary)
 	})
 
-	app.GET("/text2image/:model", func(c echo.Context) error {
+	app.GET("/image/:model", func(c echo.Context) error {
 		modelConfigs := cl.GetAllModelsConfigs()
 		modelsWithoutConfig, _ := services.ListModels(cl, ml, config.NoFilterFn, services.LOOSE_ONLY)
 
@@ -233,10 +233,10 @@ func RegisterUIRoutes(app *echo.Echo,
 		}
 
 		// Render index
-		return c.Render(200, "views/text2image", summary)
+		return c.Render(200, "views/image", summary)
 	})
 
-	app.GET("/text2image", func(c echo.Context) error {
+	app.GET("/image", func(c echo.Context) error {
 		modelConfigs := cl.GetAllModelsConfigs()
 		modelsWithoutConfig, _ := services.ListModels(cl, ml, config.NoFilterFn, services.LOOSE_ONLY)
 
@@ -266,7 +266,7 @@ func RegisterUIRoutes(app *echo.Echo,
 		}
 
 		// Render index
-		return c.Render(200, "views/text2image", summary)
+		return c.Render(200, "views/image", summary)
 	})
 
 	app.GET("/tts/:model", func(c echo.Context) error {
@@ -318,6 +318,56 @@ func RegisterUIRoutes(app *echo.Echo,
 		return c.Render(200, "views/tts", summary)
 	})
 
+	app.GET("/video/:model", func(c echo.Context) error {
+		modelConfigs := cl.GetAllModelsConfigs()
+		modelsWithoutConfig, _ := services.ListModels(cl, ml, config.NoFilterFn, services.LOOSE_ONLY)
+
+		summary := map[string]interface{}{
+			"Title":               "LocalAI - Generate videos with " + c.Param("model"),
+			"BaseURL":             middleware.BaseURL(c),
+			"ModelsConfig":        modelConfigs,
+			"ModelsWithoutConfig": modelsWithoutConfig,
+			"Model":               c.Param("model"),
+			"Version":             internal.PrintableVersion(),
+		}
+
+		// Render index
+		return c.Render(200, "views/video", summary)
+	})
+
+	app.GET("/video", func(c echo.Context) error {
+		modelConfigs := cl.GetAllModelsConfigs()
+		modelsWithoutConfig, _ := services.ListModels(cl, ml, config.NoFilterFn, services.LOOSE_ONLY)
+
+		if len(modelConfigs)+len(modelsWithoutConfig) == 0 {
+			// If no model is available redirect to the index which suggests how to install models
+			return c.Redirect(302, middleware.BaseURL(c))
+		}
+
+		modelThatCanBeUsed := ""
+		title := "LocalAI - Generate videos"
+
+		for _, b := range modelConfigs {
+			if b.HasUsecases(config.FLAG_VIDEO) {
+				modelThatCanBeUsed = b.Name
+				title = "LocalAI - Generate videos with " + modelThatCanBeUsed
+				break
+			}
+		}
+
+		summary := map[string]interface{}{
+			"Title":               title,
+			"BaseURL":             middleware.BaseURL(c),
+			"ModelsConfig":        modelConfigs,
+			"ModelsWithoutConfig": modelsWithoutConfig,
+			"Model":               modelThatCanBeUsed,
+			"Version":             internal.PrintableVersion(),
+		}
+
+		// Render index
+		return c.Render(200, "views/video", summary)
+	})
+
 	// Traces UI
 	app.GET("/traces", func(c echo.Context) error {
 		summary := map[string]interface{}{

core/http/static/video.js

@@ -0,0 +1,300 @@
+// Helper function to convert file to base64
+function fileToBase64(file) {
+  return new Promise((resolve, reject) => {
+    const reader = new FileReader();
+    reader.onload = () => {
+      // Remove data:image/...;base64, prefix if present
+      const base64 = reader.result.split(',')[1] || reader.result;
+      resolve(base64);
+    };
+    reader.onerror = reject;
+    reader.readAsDataURL(file);
+  });
+}
+
+function genVideo(event) {
+  event.preventDefault();
+  promptVideo();
+}
+
+async function promptVideo() {
+  const loader = document.getElementById("loader");
+  const input = document.getElementById("input");
+  const generateBtn = document.getElementById("generate-btn");
+  const resultDiv = document.getElementById("result");
+  const resultPlaceholder = document.getElementById("result-placeholder");
+
+  // Show loader and disable form
+  loader.classList.remove("hidden");
+  if (resultPlaceholder) {
+    resultPlaceholder.style.display = "none";
+  }
+  input.disabled = true;
+  generateBtn.disabled = true;
+
+  // Store the prompt for later restoration
+  const prompt = input.value.trim();
+  if (!prompt) {
+    alert("Please enter a prompt");
+    loader.classList.add("hidden");
+    if (resultPlaceholder) {
+      resultPlaceholder.style.display = "flex";
+    }
+    input.disabled = false;
+    generateBtn.disabled = false;
+    return;
+  }
+
+  // Collect all form values
+  const model = document.getElementById("video-model").value;
+  const size = document.getElementById("video-size").value;
+  const negativePrompt = document.getElementById("negative-prompt").value.trim();
+  
+  // Parse size into width and height
+  const sizeParts = size.split("x");
+  let width = 512;
+  let height = 512;
+  if (sizeParts.length === 2) {
+    width = parseInt(sizeParts[0]) || 512;
+    height = parseInt(sizeParts[1]) || 512;
+  }
+
+  // Video-specific parameters
+  const secondsInput = document.getElementById("video-seconds").value.trim();
+  const seconds = secondsInput ? secondsInput : undefined;
+  const fpsInput = document.getElementById("video-fps").value.trim();
+  const fps = fpsInput ? parseInt(fpsInput) : 16;
+  const framesInput = document.getElementById("video-frames").value.trim();
+  const numFrames = framesInput ? parseInt(framesInput) : undefined;
+
+  // Advanced parameters
+  const stepInput = document.getElementById("video-steps").value.trim();
+  const step = stepInput ? parseInt(stepInput) : undefined;
+  const seedInput = document.getElementById("video-seed").value.trim();
+  const seed = seedInput ? parseInt(seedInput) : undefined;
+  const cfgScaleInput = document.getElementById("video-cfg-scale").value.trim();
+  const cfgScale = cfgScaleInput ? parseFloat(cfgScaleInput) : undefined;
+
+  // Prepare request body
+  const requestBody = {
+    model: model,
+    prompt: prompt,
+    width: width,
+    height: height,
+    fps: fps,
+  };
+
+  if (negativePrompt) {
+    requestBody.negative_prompt = negativePrompt;
+  }
+
+  if (seconds !== undefined) {
+    requestBody.seconds = seconds;
+  }
+
+  if (numFrames !== undefined) {
+    requestBody.num_frames = numFrames;
+  }
+
+  if (step !== undefined) {
+    requestBody.step = step;
+  }
+
+  if (seed !== undefined) {
+    requestBody.seed = seed;
+  }
+
+  if (cfgScale !== undefined) {
+    requestBody.cfg_scale = cfgScale;
+  }
+
+  // Handle file inputs
+  try {
+    // Start image (for img2video)
+    const startImageInput = document.getElementById("start-image");
+    if (startImageInput.files.length > 0) {
+      const base64 = await fileToBase64(startImageInput.files[0]);
+      requestBody.start_image = base64;
+    }
+
+    // End image
+    const endImageInput = document.getElementById("end-image");
+    if (endImageInput.files.length > 0) {
+      const base64 = await fileToBase64(endImageInput.files[0]);
+      requestBody.end_image = base64;
+    }
+  } catch (error) {
+    console.error("Error processing image files:", error);
+    resultDiv.innerHTML = '<p class="text-xs text-red-500 p-2">Error processing image files: ' + error.message + '</p>';
+    loader.classList.add("hidden");
+    if (resultPlaceholder) {
+      resultPlaceholder.style.display = "none";
+    }
+    input.disabled = false;
+    generateBtn.disabled = false;
+    return;
+  }
+
+  // Make API request
+  try {
+    const response = await fetch("v1/videos/generations", {
+      method: "POST",
+      headers: {
+        "Content-Type": "application/json",
+      },
+      body: JSON.stringify(requestBody),
+    });
+
+    const json = await response.json();
+
+    if (json.error) {
+      // Display error
+      resultDiv.innerHTML = '<p class="text-xs text-red-500 p-2">Error: ' + json.error.message + '</p>';
+      loader.classList.add("hidden");
+      if (resultPlaceholder) {
+        resultPlaceholder.style.display = "none";
+      }
+      input.disabled = false;
+      generateBtn.disabled = false;
+      return;
+    }
+
+    // Clear result div and hide placeholder
+    resultDiv.innerHTML = '';
+    if (resultPlaceholder) {
+      resultPlaceholder.style.display = "none";
+    }
+
+    // Display generated video
+    if (json.data && json.data.length > 0) {
+      json.data.forEach((item, index) => {
+        const videoContainer = document.createElement("div");
+        videoContainer.className = "flex flex-col";
+
+        // Create video element
+        const video = document.createElement("video");
+        video.controls = true;
+        video.className = "w-full h-auto rounded-lg";
+        video.preload = "metadata";
+        
+        if (item.url) {
+          video.src = item.url;
+        } else if (item.b64_json) {
+          video.src = "data:video/mp4;base64," + item.b64_json;
+        } else {
+          return; // Skip invalid items
+        }
+        
+        videoContainer.appendChild(video);
+
+        // Create caption container
+        const captionDiv = document.createElement("div");
+        captionDiv.className = "mt-2 p-2 bg-[var(--color-bg-secondary)] rounded-lg text-xs";
+
+        // Prompt caption
+        const promptCaption = document.createElement("p");
+        promptCaption.className = "text-[var(--color-text-primary)] mb-1.5 break-words";
+        promptCaption.innerHTML = '<strong>Prompt:</strong> ' + escapeHtml(prompt);
+        captionDiv.appendChild(promptCaption);
+
+        // Negative prompt if provided
+        if (negativePrompt) {
+          const negativeCaption = document.createElement("p");
+          negativeCaption.className = "text-[var(--color-text-secondary)] mb-1.5 break-words";
+          negativeCaption.innerHTML = '<strong>Negative Prompt:</strong> ' + escapeHtml(negativePrompt);
+          captionDiv.appendChild(negativeCaption);
+        }
+
+        // Generation details
+        const detailsDiv = document.createElement("div");
+        detailsDiv.className = "flex flex-wrap gap-3 text-[10px] text-[var(--color-text-secondary)] mt-1.5";
+        detailsDiv.innerHTML = `
+          <span><strong>Size:</strong> ${width}x${height}</span>
+          ${fps ? `<span><strong>FPS:</strong> ${fps}</span>` : ''}
+          ${numFrames !== undefined ? `<span><strong>Frames:</strong> ${numFrames}</span>` : ''}
+          ${seconds !== undefined ? `<span><strong>Duration:</strong> ${seconds}s</span>` : ''}
+          ${step !== undefined ? `<span><strong>Steps:</strong> ${step}</span>` : ''}
+          ${seed !== undefined ? `<span><strong>Seed:</strong> ${seed}</span>` : ''}
+          ${cfgScale !== undefined ? `<span><strong>CFG Scale:</strong> ${cfgScale}</span>` : ''}
+        `;
+        captionDiv.appendChild(detailsDiv);
+
+        // Copy prompt button
+        const copyBtn = document.createElement("button");
+        copyBtn.className = "mt-1.5 px-2 py-0.5 text-[10px] bg-[var(--color-primary)] text-white rounded hover:opacity-80";
+        copyBtn.innerHTML = '<i class="fas fa-copy mr-1"></i>Copy Prompt';
+        copyBtn.onclick = () => {
+          navigator.clipboard.writeText(prompt).then(() => {
+            copyBtn.innerHTML = '<i class="fas fa-check mr-1"></i>Copied!';
+            setTimeout(() => {
+              copyBtn.innerHTML = '<i class="fas fa-copy mr-1"></i>Copy Prompt';
+            }, 2000);
+          });
+        };
+        captionDiv.appendChild(copyBtn);
+
+        videoContainer.appendChild(captionDiv);
+        resultDiv.appendChild(videoContainer);
+      });
+      // Hide placeholder when videos are displayed
+      if (resultPlaceholder) {
+        resultPlaceholder.style.display = "none";
+      }
+    } else {
+      resultDiv.innerHTML = '<p class="text-xs text-[var(--color-text-secondary)] p-2">No videos were generated.</p>';
+      if (resultPlaceholder) {
+        resultPlaceholder.style.display = "none";
+      }
+    }
+
+  } catch (error) {
+    console.error("Error generating video:", error);
+    resultDiv.innerHTML = '<p class="text-xs text-red-500 p-2">Error: ' + error.message + '</p>';
+    if (resultPlaceholder) {
+      resultPlaceholder.style.display = "none";
+    }
+  } finally {
+    // Hide loader and re-enable form
+    loader.classList.add("hidden");
+    input.disabled = false;
+    generateBtn.disabled = false;
+    input.focus();
+  }
+}
+
+// Helper function to escape HTML
+function escapeHtml(text) {
+  const div = document.createElement("div");
+  div.textContent = text;
+  return div.innerHTML;
+}
+
+// Initialize
+document.addEventListener("DOMContentLoaded", function() {
+  const input = document.getElementById("input");
+  const form = document.getElementById("genvideo");
+
+  if (input) {
+    input.focus();
+  }
+
+  if (form) {
+    form.addEventListener("submit", genVideo);
+  }
+
+  // Handle Enter key press in the prompt input (but allow Shift+Enter for new lines)
+  if (input) {
+    input.addEventListener("keydown", function(event) {
+      if (event.key === "Enter" && !event.shiftKey) {
+        event.preventDefault();
+        genVideo(event);
+      }
+    });
+  }
+
+  // Hide loader initially
+  const loader = document.getElementById("loader");
+  if (loader) {
+    loader.classList.add("hidden");
+  }
+});

core/http/views/image.html

@@ -28,19 +28,19 @@
                               {{ $cfg := . }}
                               {{ range .KnownUsecaseStrings }}
                                 {{ if eq . "FLAG_IMAGE" }}
-                                    <option value="text2image/{{$cfg.Name}}" {{ if eq $cfg.Name $model }} selected {{end}} class="bg-[var(--color-bg-primary)] text-[var(--color-text-primary)]">{{$cfg.Name}}</option>
+                                    <option value="image/{{$cfg.Name}}" {{ if eq $cfg.Name $model }} selected {{end}} class="bg-[var(--color-bg-primary)] text-[var(--color-text-primary)]">{{$cfg.Name}}</option>
                                 {{ end }}
                               {{ end }}
                             {{ end }}
                             {{ range .ModelsWithoutConfig }}
-                                <option value="text2image/{{.}}" {{ if eq . $model }} selected {{ end }} class="bg-[var(--color-bg-primary)] text-[var(--color-text-primary)]">{{.}}</option>
+                                <option value="image/{{.}}" {{ if eq . $model }} selected {{ end }} class="bg-[var(--color-bg-primary)] text-[var(--color-text-primary)]">{{.}}</option>
                             {{end}}
                         </select>
                     </div>
                     
                     <div class="relative">
                         <input id="image-model" type="hidden" value="{{.Model}}">
-                        <form id="genimage" action="text2image/{{.Model}}" method="get">
+                        <form id="genimage" action="image/{{.Model}}" method="get">
                             <!-- Basic Settings -->
                             <div class="space-y-2">
                                 <!-- Prompt -->
@@ -326,4 +326,4 @@
 </script>
 
 </body>
-</html>
+</html>

core/http/views/manage.html

@@ -315,7 +315,7 @@
                                     </a>
                                     {{ end }}
                                     {{ if eq . "FLAG_IMAGE" }}
-                                    <a href="text2image/{{$backendCfg.Name}}" class="inline-flex items-center px-1.5 py-0.5 rounded text-[10px] font-medium bg-[var(--color-success)]/10 text-green-300 hover:bg-[var(--color-success)]/20 transition-colors" title="Image">
+                                    <a href="image/{{$backendCfg.Name}}" class="inline-flex items-center px-1.5 py-0.5 rounded text-[10px] font-medium bg-[var(--color-success)]/10 text-green-300 hover:bg-[var(--color-success)]/20 transition-colors" title="Image">
                                         <i class="fas fa-image text-[8px] mr-1"></i>Image
                                     </a>
                                     {{ end }}

core/http/views/partials/navbar.html

@@ -25,9 +25,12 @@
                 <a href="chat/" class="text-[var(--color-text-secondary)] hover:text-[var(--color-text-primary)] px-2 py-2 rounded-lg transition duration-300 ease-in-out hover:bg-[var(--color-bg-secondary)] flex items-center group text-sm">
                     <i class="fa-solid fa-comments text-[var(--color-primary)] mr-1.5 text-sm group-hover:scale-110 transition-transform"></i>Chat
                 </a>
-                <a href="text2image/" class="text-[var(--color-text-secondary)] hover:text-[var(--color-text-primary)] px-2 py-2 rounded-lg transition duration-300 ease-in-out hover:bg-[var(--color-bg-secondary)] flex items-center group text-sm">
+                <a href="image/" class="text-[var(--color-text-secondary)] hover:text-[var(--color-text-primary)] px-2 py-2 rounded-lg transition duration-300 ease-in-out hover:bg-[var(--color-bg-secondary)] flex items-center group text-sm">
                     <i class="fas fa-image text-[var(--color-primary)] mr-1.5 text-sm group-hover:scale-110 transition-transform"></i>Images
                 </a>
+                <a href="video/" class="text-[var(--color-text-secondary)] hover:text-[var(--color-text-primary)] px-2 py-2 rounded-lg transition duration-300 ease-in-out hover:bg-[var(--color-bg-secondary)] flex items-center group text-sm">
+                    <i class="fas fa-video text-[var(--color-primary)] mr-1.5 text-sm group-hover:scale-110 transition-transform"></i>Video
+                </a>
                 <a href="tts/" class="text-[var(--color-text-secondary)] hover:text-[var(--color-text-primary)] px-2 py-2 rounded-lg transition duration-300 ease-in-out hover:bg-[var(--color-bg-secondary)] flex items-center group text-sm">
                     <i class="fa-solid fa-music text-[var(--color-primary)] mr-1.5 text-sm group-hover:scale-110 transition-transform"></i>TTS
                 </a>
@@ -85,9 +88,12 @@
                 <a href="chat/" class="block text-[var(--color-text-secondary)] hover:text-[var(--color-text-primary)] hover:bg-[var(--color-bg-secondary)] px-3 py-2 rounded-lg transition duration-300 ease-in-out flex items-center text-sm">
                     <i class="fa-solid fa-comments text-[var(--color-primary)] mr-3 w-5 text-center text-sm"></i>Chat
                 </a>
-                <a href="text2image/" class="block text-[var(--color-text-secondary)] hover:text-[var(--color-text-primary)] hover:bg-[var(--color-bg-secondary)] px-3 py-2 rounded-lg transition duration-300 ease-in-out flex items-center text-sm">
+                <a href="image/" class="block text-[var(--color-text-secondary)] hover:text-[var(--color-text-primary)] hover:bg-[var(--color-bg-secondary)] px-3 py-2 rounded-lg transition duration-300 ease-in-out flex items-center text-sm">
                     <i class="fas fa-image text-[var(--color-primary)] mr-3 w-5 text-center text-sm"></i>Images
                 </a>
+                <a href="video/" class="block text-[var(--color-text-secondary)] hover:text-[var(--color-text-primary)] hover:bg-[var(--color-bg-secondary)] px-3 py-2 rounded-lg transition duration-300 ease-in-out flex items-center text-sm">
+                    <i class="fas fa-video text-[var(--color-primary)] mr-3 w-5 text-center text-sm"></i>Video
+                </a>
                 <a href="tts/" class="block text-[var(--color-text-secondary)] hover:text-[var(--color-text-primary)] hover:bg-[var(--color-bg-secondary)] px-3 py-2 rounded-lg transition duration-300 ease-in-out flex items-center text-sm">
                     <i class="fa-solid fa-music text-[var(--color-primary)] mr-3 w-5 text-center text-sm"></i>TTS
                 </a>

core/http/views/settings.html

@@ -485,6 +485,28 @@
                 </div>
             </div>
 
+            <!-- Open Responses Settings Section -->
+            <div class="bg-[var(--color-bg-secondary)] border border-[var(--color-accent)]/20 rounded-lg p-6">
+                <h2 class="text-xl font-semibold text-[var(--color-text-primary)] mb-4 flex items-center">
+                    <i class="fas fa-database mr-2 text-[var(--color-accent)] text-sm"></i>
+                    Open Responses Settings
+                </h2>
+                <p class="text-xs text-[var(--color-text-secondary)] mb-4">
+                    Configure Open Responses API response storage
+                </p>
+
+                <div class="space-y-4">
+                    <!-- Store TTL -->
+                    <div>
+                        <label class="block text-sm font-medium text-[var(--color-text-primary)] mb-2">Response Store TTL</label>
+                        <p class="text-xs text-[var(--color-text-secondary)] mb-2">Time-to-live for stored responses (e.g., 1h, 30m, 0 = no expiration)</p>
+                        <input type="text" x-model="settings.open_responses_store_ttl"
+                               placeholder="0"
+                               class="w-full px-3 py-2 bg-[var(--color-bg-primary)] border border-[var(--color-accent)]/20 rounded text-sm text-[var(--color-text-primary)] focus:outline-none focus:ring-2 focus:ring-[var(--color-accent)]/50">
+                    </div>
+                </div>
+            </div>
+
             <!-- API Keys Settings Section -->
             <div class="bg-[var(--color-bg-secondary)] border border-[var(--color-error-light)] rounded-lg p-6">
                 <h2 class="text-xl font-semibold text-[var(--color-text-primary)] mb-4 flex items-center">
@@ -633,7 +655,8 @@ function settingsDashboard() {
             galleries_json: '[]',
             backend_galleries_json: '[]',
             api_keys_text: '',
-            agent_job_retention_days: 30
+            agent_job_retention_days: 30,
+            open_responses_store_ttl: '0'
         },
         sourceInfo: '',
         saving: false,
@@ -680,7 +703,8 @@ function settingsDashboard() {
                         galleries_json: JSON.stringify(data.galleries || [], null, 2),
                         backend_galleries_json: JSON.stringify(data.backend_galleries || [], null, 2),
                         api_keys_text: (data.api_keys || []).join('\n'),
-                        agent_job_retention_days: data.agent_job_retention_days || 30
+                        agent_job_retention_days: data.agent_job_retention_days || 30,
+                        open_responses_store_ttl: data.open_responses_store_ttl || '0'
                     };
                     this.sourceInfo = data.source || 'default';
                 } else {
@@ -838,6 +862,9 @@ function settingsDashboard() {
                 if (this.settings.agent_job_retention_days !== undefined) {
                     payload.agent_job_retention_days = parseInt(this.settings.agent_job_retention_days) || 30;
                 }
+                if (this.settings.open_responses_store_ttl !== undefined) {
+                    payload.open_responses_store_ttl = this.settings.open_responses_store_ttl;
+                }
 
                 const response = await fetch('/api/settings', {
                     method: 'POST',

core/http/views/video.html

@@ -0,0 +1,315 @@
+<!DOCTYPE html>
+<html lang="en">
+{{template "views/partials/head" .}}
+<script defer src="static/video.js"></script>
+
+<body class="bg-[var(--color-bg-primary)] text-[var(--color-text-primary)] flex flex-col h-screen">
+<div class="flex flex-col flex-1 overflow-hidden">
+   
+    {{template "views/partials/navbar" .}}
+    <div class="flex flex-1 overflow-hidden">
+        <!-- Two Column Layout: Settings on Left, Preview on Right -->
+        <div class="flex flex-col lg:flex-row flex-1 gap-4 p-4 overflow-hidden">
+            <!-- Left Column: Generation Settings -->
+            <div class="flex-shrink-0 lg:w-1/4 flex flex-col min-h-0">
+                <div class="card p-3 space-y-3 overflow-y-auto flex-1">
+                    <!-- Model Selection - Compact -->
+                    <div class="space-y-1.5">
+                        <div class="flex items-center justify-between gap-2">
+                            <label class="text-xs font-medium text-[var(--color-text-secondary)] uppercase tracking-wide flex-shrink-0">Model</label>
+                        </div>
+                        <select x-data="{ link : '' }" x-model="link" x-init="$watch('link', value => window.location = link)" 
+                            id="model-select"
+                            class="input w-full p-1.5 text-xs"
+                        >	
+                            <option value="" disabled class="text-[var(--color-text-secondary)]">Select a model</option>
+                            {{ $model:=.Model}}
+                            {{ range .ModelsConfig }}
+                              {{ $cfg := . }}
+                              {{ range .KnownUsecaseStrings }}
+                                {{ if eq . "FLAG_VIDEO" }}
+                                    <option value="video/{{$cfg.Name}}" {{ if eq $cfg.Name $model }} selected {{end}} class="bg-[var(--color-bg-primary)] text-[var(--color-text-primary)]">{{$cfg.Name}}</option>
+                                {{ end }}
+                              {{ end }}
+                            {{ end }}
+                            {{ range .ModelsWithoutConfig }}
+                                <option value="video/{{.}}" {{ if eq . $model }} selected {{ end }} class="bg-[var(--color-bg-primary)] text-[var(--color-text-primary)]">{{.}}</option>
+                            {{end}}
+                        </select>
+                    </div>
+                    
+                    <div class="relative">
+                        <input id="video-model" type="hidden" value="{{.Model}}">
+                        <form id="genvideo" action="video/{{.Model}}" method="get">
+                            <!-- Basic Settings -->
+                            <div class="space-y-2">
+                                <!-- Prompt -->
+                                <div class="space-y-1">
+                                    <label for="input" class="block text-xs font-medium text-[var(--color-text-secondary)] uppercase tracking-wide">
+                                        <i class="fas fa-magic mr-1.5 text-[var(--color-primary)]"></i>Prompt
+                                    </label>
+                                    <textarea
+                                        id="input"
+                                        name="input"
+                                        placeholder="Describe the video you want to generate..."
+                                        autocomplete="off"
+                                        rows="3"
+                                        class="input w-full p-1.5 text-xs resize-y"
+                                        required
+                                    ></textarea>
+                                </div>
+
+                                <!-- Negative Prompt -->
+                                <div class="space-y-1">
+                                    <label for="negative-prompt" class="block text-xs font-medium text-[var(--color-text-secondary)] uppercase tracking-wide">
+                                        <i class="fas fa-ban mr-1.5 text-[var(--color-primary)]"></i>Negative Prompt
+                                    </label>
+                                    <textarea
+                                        id="negative-prompt"
+                                        name="negative-prompt"
+                                        placeholder="Things to avoid in the video..."
+                                        rows="2"
+                                        class="input w-full p-1.5 text-xs resize-y"
+                                    ></textarea>
+                                </div>
+
+                                <!-- Size Selection with Presets -->
+                                <div class="space-y-1">
+                                    <label for="video-size" class="block text-xs font-medium text-[var(--color-text-secondary)] uppercase tracking-wide">
+                                        <i class="fas fa-expand-arrows-alt mr-1.5 text-[var(--color-primary)]"></i>Video Size
+                                    </label>
+                                    <div class="flex flex-wrap gap-1.5 mb-1.5">
+                                        <button type="button" class="size-preset px-2 py-0.5 text-[10px] rounded border border-[var(--color-border)] hover:bg-[var(--color-bg-secondary)]" data-size="256x256">256×256</button>
+                                        <button type="button" class="size-preset px-2 py-0.5 text-[10px] rounded border border-[var(--color-border)] hover:bg-[var(--color-bg-secondary)]" data-size="512x512">512×512</button>
+                                        <button type="button" class="size-preset px-2 py-0.5 text-[10px] rounded border border-[var(--color-border)] hover:bg-[var(--color-bg-secondary)]" data-size="768x768">768×768</button>
+                                        <button type="button" class="size-preset px-2 py-0.5 text-[10px] rounded border border-[var(--color-border)] hover:bg-[var(--color-bg-secondary)]" data-size="1024x1024">1024×1024</button>
+                                    </div>
+                                    <input
+                                        type="text"
+                                        id="video-size"
+                                        value="512x512"
+                                        placeholder="e.g., 256x256, 512x512, 1024x1024"
+                                        class="input p-1.5 text-xs w-full"
+                                    />
+                                </div>
+
+                                <!-- Video Duration / FPS / Frames -->
+                                <div class="space-y-1">
+                                    <label for="video-seconds" class="block text-xs font-medium text-[var(--color-text-secondary)] uppercase tracking-wide">
+                                        <i class="fas fa-clock mr-1.5 text-[var(--color-primary)]"></i>Duration (seconds)
+                                    </label>
+                                    <input
+                                        type="number"
+                                        id="video-seconds"
+                                        name="seconds"
+                                        min="1"
+                                        max="60"
+                                        placeholder="Leave empty for default"
+                                        class="input p-1.5 text-xs w-full"
+                                    />
+                                </div>
+
+                                <div class="space-y-1">
+                                    <label for="video-fps" class="block text-xs font-medium text-[var(--color-text-secondary)] uppercase tracking-wide">
+                                        <i class="fas fa-film mr-1.5 text-[var(--color-primary)]"></i>FPS
+                                    </label>
+                                    <input
+                                        type="number"
+                                        id="video-fps"
+                                        name="fps"
+                                        min="1"
+                                        max="60"
+                                        value="16"
+                                        placeholder="Frames per second"
+                                        class="input p-1.5 text-xs w-full"
+                                    />
+                                </div>
+
+                                <div class="space-y-1">
+                                    <label for="video-frames" class="block text-xs font-medium text-[var(--color-text-secondary)] uppercase tracking-wide">
+                                        <i class="fas fa-images mr-1.5 text-[var(--color-primary)]"></i>Number of Frames
+                                    </label>
+                                    <input
+                                        type="number"
+                                        id="video-frames"
+                                        name="num_frames"
+                                        min="1"
+                                        max="500"
+                                        placeholder="Leave empty for default"
+                                        class="input p-1.5 text-xs w-full"
+                                    />
+                                </div>
+                            </div>
+
+                            <!-- Advanced Settings (Collapsible) -->
+                            <div class="space-y-2">
+                                <button type="button" id="advanced-toggle" class="w-full flex items-center justify-between px-2 py-1.5 text-xs rounded text-[var(--color-text-secondary)] hover:text-[var(--color-text-primary)] hover:bg-[var(--color-bg-secondary)] transition-colors">
+                                    <span><i class="fa-solid fa-sliders mr-1.5 text-[var(--color-primary)]"></i> Advanced Settings</span>
+                                    <i class="fas fa-chevron-down text-[10px]" id="advanced-chevron"></i>
+                                </button>
+                                <div id="advanced-settings" class="hidden p-2 bg-[var(--color-bg-secondary)] border border-[var(--color-primary-border)]/20 rounded pl-4 border-l-2 border-[var(--color-bg-secondary)] space-y-2">
+                                    <!-- Steps -->
+                                    <div class="space-y-1">
+                                        <label for="video-steps" class="block text-xs text-[var(--color-text-secondary)]">
+                                            <i class="fas fa-step-forward mr-1.5 text-[var(--color-primary)]"></i>Steps
+                                        </label>
+                                        <input
+                                            type="number"
+                                            id="video-steps"
+                                            name="step"
+                                            min="1"
+                                            max="100"
+                                            placeholder="Leave empty for default"
+                                            class="input p-1.5 text-xs w-full"
+                                        />
+                                    </div>
+
+                                    <!-- Seed -->
+                                    <div class="space-y-1">
+                                        <label for="video-seed" class="block text-xs text-[var(--color-text-secondary)]">
+                                            <i class="fas fa-seedling mr-1.5 text-[var(--color-primary)]"></i>Seed
+                                        </label>
+                                        <input
+                                            type="number"
+                                            id="video-seed"
+                                            name="seed"
+                                            min="0"
+                                            placeholder="Leave empty for random"
+                                            class="input p-1.5 text-xs w-full"
+                                        />
+                                    </div>
+
+                                    <!-- CFG Scale -->
+                                    <div class="space-y-1">
+                                        <label for="video-cfg-scale" class="block text-xs text-[var(--color-text-secondary)]">
+                                            <i class="fas fa-sliders-h mr-1.5 text-[var(--color-primary)]"></i>CFG Scale
+                                        </label>
+                                        <input
+                                            type="number"
+                                            id="video-cfg-scale"
+                                            name="cfg_scale"
+                                            min="0"
+                                            max="20"
+                                            step="0.1"
+                                            placeholder="Leave empty for default"
+                                            class="input p-1.5 text-xs w-full"
+                                        />
+                                    </div>
+                                </div>
+                            </div>
+
+                            <!-- Image Inputs (Collapsible) -->
+                            <div class="space-y-2">
+                                <button type="button" id="image-inputs-toggle" class="w-full flex items-center justify-between px-2 py-1.5 text-xs rounded text-[var(--color-text-secondary)] hover:text-[var(--color-text-primary)] hover:bg-[var(--color-bg-secondary)] transition-colors">
+                                    <span><i class="fa-solid fa-image mr-1.5 text-[var(--color-primary)]"></i> Image Inputs</span>
+                                    <i class="fas fa-chevron-down text-[10px]" id="image-inputs-chevron"></i>
+                                </button>
+                                <div id="image-inputs-settings" class="hidden p-2 bg-[var(--color-bg-secondary)] border border-[var(--color-primary-border)]/20 rounded pl-4 border-l-2 border-[var(--color-bg-secondary)] space-y-2">
+                                    <!-- Start Image (img2video) -->
+                                    <div class="space-y-1">
+                                        <label for="start-image" class="block text-xs text-[var(--color-text-secondary)]">
+                                            <i class="fas fa-play-circle mr-1.5 text-[var(--color-primary)]"></i>Start Image (img2video)
+                                        </label>
+                                        <input
+                                            type="file"
+                                            id="start-image"
+                                            name="start_image"
+                                            accept="image/*"
+                                            class="input p-1.5 text-xs w-full"
+                                        />
+                                    </div>
+
+                                    <!-- End Image -->
+                                    <div class="space-y-1">
+                                        <label for="end-image" class="block text-xs text-[var(--color-text-secondary)]">
+                                            <i class="fas fa-stop-circle mr-1.5 text-[var(--color-primary)]"></i>End Image
+                                        </label>
+                                        <input
+                                            type="file"
+                                            id="end-image"
+                                            name="end_image"
+                                            accept="image/*"
+                                            class="input p-1.5 text-xs w-full"
+                                        />
+                                    </div>
+                                </div>
+                            </div>
+                            
+                            <!-- Submit Button -->
+                            <div>
+                                <button
+                                    type="submit"
+                                    id="generate-btn"
+                                    class="w-full px-2 py-1.5 text-xs rounded text-[var(--color-bg-primary)] bg-[var(--color-primary)] hover:bg-[var(--color-primary)]/90 transition-colors font-medium"
+                                >
+                                    <i class="fas fa-video mr-1.5"></i>Generate Video
+                                </button>
+                            </div>
+                        </form>
+                    </div>
+                </div>
+            </div>
+
+            <!-- Right Column: Video Preview -->
+            <div class="flex-grow lg:w-3/4 flex flex-col min-h-0">
+                <div class="relative flex-1 min-h-0 overflow-y-auto">
+                    <!-- Loading Animation -->
+                    <div id="loader" class="hidden absolute inset-0 flex items-center justify-center bg-[var(--color-bg-primary)]/80 rounded-xl z-10">
+                        <div class="text-center">
+                            <svg class="animate-spin h-10 w-10 text-[var(--color-primary)] mx-auto mb-3" xmlns="http://www.w3.org/2000/svg" fill="none" viewBox="0 0 24 24">
+                                <circle class="opacity-25" cx="12" cy="12" r="10" stroke="currentColor" stroke-width="4"></circle>
+                                <path class="opacity-75" fill="currentColor" d="M4 12a8 8 0 018-8V0C5.373 0 0 5.373 0 12h4zm2 5.291A7.962 7.962 0 014 12H0c0 3.042 1.135 5.824 3 7.938l3-2.647z"></path>
+                            </svg>
+                            <p class="text-xs text-[var(--color-text-secondary)]">Generating video...</p>
+                        </div>
+                    </div>
+                    <!-- Placeholder when no videos -->
+                    <div id="result-placeholder" class="min-h-[400px] flex items-center justify-center flex-shrink-0">
+                        <p class="text-xs text-[var(--color-text-secondary)] italic text-center">Your generated videos will appear here</p>
+                    </div>
+                    <!-- Results container -->
+                    <div id="result" class="grid grid-cols-1 gap-4 pb-4"></div>
+                </div>
+            </div>
+        </div>
+    </div>
+
+</div>
+
+<script>
+    // Collapsible sections
+    document.getElementById('advanced-toggle').addEventListener('click', function() {
+        const settings = document.getElementById('advanced-settings');
+        const chevron = document.getElementById('advanced-chevron');
+        settings.classList.toggle('hidden');
+        chevron.classList.toggle('fa-chevron-down');
+        chevron.classList.toggle('fa-chevron-up');
+    });
+
+    document.getElementById('image-inputs-toggle').addEventListener('click', function() {
+        const settings = document.getElementById('image-inputs-settings');
+        const chevron = document.getElementById('image-inputs-chevron');
+        settings.classList.toggle('hidden');
+        chevron.classList.toggle('fa-chevron-down');
+        chevron.classList.toggle('fa-chevron-up');
+    });
+
+    // Size preset buttons
+    document.querySelectorAll('.size-preset').forEach(button => {
+        button.addEventListener('click', function() {
+            const size = this.getAttribute('data-size');
+            document.getElementById('video-size').value = size;
+            // Update active state
+            document.querySelectorAll('.size-preset').forEach(btn => {
+                btn.classList.remove('bg-[var(--color-primary)]', 'text-white');
+            });
+            this.classList.add('bg-[var(--color-primary)]', 'text-white');
+        });
+    });
+
+    // Set initial active size preset
+    document.querySelector('.size-preset[data-size="512x512"]').classList.add('bg-[var(--color-primary)]', 'text-white');
+</script>
+
+</body>
+</html>

core/schema/openresponses.go

@@ -0,0 +1,306 @@
+package schema
+
+import (
+	"context"
+)
+
+// Open Responses status constants
+const (
+	ORStatusQueued     = "queued"
+	ORStatusInProgress = "in_progress"
+	ORStatusCompleted  = "completed"
+	ORStatusFailed     = "failed"
+	ORStatusIncomplete = "incomplete"
+	ORStatusCancelled  = "cancelled"
+)
+
+// OpenResponsesRequest represents a request to the Open Responses API
+// https://www.openresponses.org/specification
+type OpenResponsesRequest struct {
+	Model              string            `json:"model"`
+	Input              interface{}       `json:"input"` // string or []ORItemParam
+	Tools              []ORFunctionTool  `json:"tools,omitempty"`
+	ToolChoice         interface{}       `json:"tool_choice,omitempty"` // "auto"|"required"|"none"|{type:"function",name:"..."}
+	Stream             bool              `json:"stream,omitempty"`
+	MaxOutputTokens    *int              `json:"max_output_tokens,omitempty"`
+	Temperature        *float64          `json:"temperature,omitempty"`
+	TopP               *float64          `json:"top_p,omitempty"`
+	Truncation         string            `json:"truncation,omitempty"` // "auto"|"disabled"
+	Instructions       string            `json:"instructions,omitempty"`
+	Reasoning          *ORReasoningParam `json:"reasoning,omitempty"`
+	Metadata           map[string]string `json:"metadata,omitempty"`
+	PreviousResponseID string            `json:"previous_response_id,omitempty"`
+
+	// Additional parameters from spec
+	TextFormat        interface{} `json:"text_format,omitempty"`         // TextResponseFormat or JsonSchemaResponseFormatParam
+	ServiceTier       string      `json:"service_tier,omitempty"`        // "auto"|"default"|priority hint
+	AllowedTools      []string    `json:"allowed_tools,omitempty"`       // Restrict which tools can be invoked
+	Store             *bool       `json:"store,omitempty"`               // Whether to store the response
+	Include           []string    `json:"include,omitempty"`             // What to include in response
+	ParallelToolCalls *bool       `json:"parallel_tool_calls,omitempty"` // Allow parallel tool calls
+	PresencePenalty   *float64    `json:"presence_penalty,omitempty"`    // Presence penalty (-2.0 to 2.0)
+	FrequencyPenalty  *float64    `json:"frequency_penalty,omitempty"`   // Frequency penalty (-2.0 to 2.0)
+	TopLogprobs       *int        `json:"top_logprobs,omitempty"`        // Number of top logprobs to return
+	Background        *bool       `json:"background,omitempty"`          // Run request in background
+	MaxToolCalls      *int        `json:"max_tool_calls,omitempty"`      // Maximum number of tool calls
+
+	// OpenAI-compatible extensions (not in Open Responses spec)
+	LogitBias map[string]float64 `json:"logit_bias,omitempty"` // Map of token IDs to bias values (-100 to 100)
+
+	// Internal fields (like OpenAIRequest)
+	Context context.Context    `json:"-"`
+	Cancel  context.CancelFunc `json:"-"`
+}
+
+// ModelName implements the LocalAIRequest interface
+func (r *OpenResponsesRequest) ModelName(s *string) string {
+	if s != nil {
+		r.Model = *s
+	}
+	return r.Model
+}
+
+// ORFunctionTool represents a function tool definition
+type ORFunctionTool struct {
+	Type        string                 `json:"type"` // always "function"
+	Name        string                 `json:"name"`
+	Description string                 `json:"description,omitempty"`
+	Parameters  map[string]interface{} `json:"parameters,omitempty"`
+	Strict      bool                   `json:"strict"` // Always include in response
+}
+
+// ORReasoningParam represents reasoning configuration
+type ORReasoningParam struct {
+	Effort  string `json:"effort,omitempty"`  // "none"|"low"|"medium"|"high"|"xhigh"
+	Summary string `json:"summary,omitempty"` // "auto"|"concise"|"detailed"
+}
+
+// ORItemParam represents an input/output item (discriminated union by type)
+type ORItemParam struct {
+	Type   string `json:"type"`             // message|function_call|function_call_output|reasoning|item_reference
+	ID     string `json:"id,omitempty"`     // Present for all output items
+	Status string `json:"status,omitempty"` // in_progress|completed|incomplete
+
+	// Message fields
+	Role    string      `json:"role,omitempty"`    // user|assistant|system|developer
+	Content interface{} `json:"content,omitempty"` // string or []ORContentPart for messages
+
+	// Function call fields
+	CallID    string `json:"call_id,omitempty"`
+	Name      string `json:"name,omitempty"`
+	Arguments string `json:"arguments,omitempty"`
+
+	// Function call output fields
+	Output interface{} `json:"output,omitempty"` // string or []ORContentPart
+
+	// Note: For item_reference type, use the ID field above to reference the item
+}
+
+// ORContentPart represents a content block (discriminated union by type)
+// For output_text: type, text, annotations, logprobs are ALL REQUIRED per Open Responses spec
+type ORContentPart struct {
+	Type        string         `json:"type"`        // input_text|input_image|input_file|output_text|refusal
+	Text        string         `json:"text"`        // REQUIRED for output_text - must always be present (even if empty)
+	Annotations []ORAnnotation `json:"annotations"` // REQUIRED for output_text - must always be present (use [])
+	Logprobs    []ORLogProb    `json:"logprobs"`    // REQUIRED for output_text - must always be present (use [])
+	ImageURL    string         `json:"image_url,omitempty"`
+	FileURL     string         `json:"file_url,omitempty"`
+	Filename    string         `json:"filename,omitempty"`
+	FileData    string         `json:"file_data,omitempty"`
+	Refusal     string         `json:"refusal,omitempty"`
+	Detail      string         `json:"detail,omitempty"` // low|high|auto for images
+}
+
+// OROutputTextContentPart is an alias for ORContentPart used specifically for output_text
+type OROutputTextContentPart = ORContentPart
+
+// ORItemField represents an output item (same structure as ORItemParam)
+type ORItemField = ORItemParam
+
+// ORResponseResource represents the main response object
+type ORResponseResource struct {
+	ID                 string               `json:"id"`
+	Object             string               `json:"object"` // always "response"
+	CreatedAt          int64                `json:"created_at"`
+	CompletedAt        *int64               `json:"completed_at"` // Required: present as number or null
+	Status             string               `json:"status"`       // in_progress|completed|failed|incomplete
+	Model              string               `json:"model"`
+	Output             []ORItemField        `json:"output"`
+	Error              *ORError             `json:"error"`              // Always present, null if no error
+	IncompleteDetails  *ORIncompleteDetails `json:"incomplete_details"` // Always present, null if complete
+	PreviousResponseID *string              `json:"previous_response_id"`
+	Instructions       *string              `json:"instructions"`
+
+	// Tool-related fields
+	Tools             []ORFunctionTool `json:"tools"` // Always present, empty array if no tools
+	ToolChoice        interface{}      `json:"tool_choice"`
+	ParallelToolCalls bool             `json:"parallel_tool_calls"`
+	MaxToolCalls      *int             `json:"max_tool_calls"` // nullable
+
+	// Sampling parameters (always required)
+	Temperature      float64 `json:"temperature"`
+	TopP             float64 `json:"top_p"`
+	PresencePenalty  float64 `json:"presence_penalty"`
+	FrequencyPenalty float64 `json:"frequency_penalty"`
+	TopLogprobs      int     `json:"top_logprobs"` // Default to 0
+	MaxOutputTokens  *int    `json:"max_output_tokens"`
+
+	// Text format configuration
+	Text *ORTextConfig `json:"text"`
+
+	// Truncation and reasoning
+	Truncation string       `json:"truncation"`
+	Reasoning  *ORReasoning `json:"reasoning"` // nullable
+
+	// Usage statistics
+	Usage *ORUsage `json:"usage"` // nullable
+
+	// Metadata and operational flags
+	Metadata    map[string]string `json:"metadata"`
+	Store       bool              `json:"store"`
+	Background  bool              `json:"background"`
+	ServiceTier string            `json:"service_tier"`
+
+	// Safety and caching
+	SafetyIdentifier *string `json:"safety_identifier"` // nullable
+	PromptCacheKey   *string `json:"prompt_cache_key"`  // nullable
+}
+
+// ORTextConfig represents text format configuration
+type ORTextConfig struct {
+	Format *ORTextFormat `json:"format,omitempty"`
+}
+
+// ORTextFormat represents the text format type
+type ORTextFormat struct {
+	Type string `json:"type"` // "text" or "json_schema"
+}
+
+// ORError represents an error in the response
+type ORError struct {
+	Type    string `json:"type"` // invalid_request|not_found|server_error|model_error|too_many_requests
+	Code    string `json:"code,omitempty"`
+	Message string `json:"message"`
+	Param   string `json:"param,omitempty"`
+}
+
+// ORUsage represents token usage statistics
+type ORUsage struct {
+	InputTokens         int                    `json:"input_tokens"`
+	OutputTokens        int                    `json:"output_tokens"`
+	TotalTokens         int                    `json:"total_tokens"`
+	InputTokensDetails  *ORInputTokensDetails  `json:"input_tokens_details"`  // Always present
+	OutputTokensDetails *OROutputTokensDetails `json:"output_tokens_details"` // Always present
+}
+
+// ORInputTokensDetails represents input token breakdown
+type ORInputTokensDetails struct {
+	CachedTokens int `json:"cached_tokens"` // Always include, even if 0
+}
+
+// OROutputTokensDetails represents output token breakdown
+type OROutputTokensDetails struct {
+	ReasoningTokens int `json:"reasoning_tokens"` // Always include, even if 0
+}
+
+// ORReasoning represents reasoning configuration and metadata
+type ORReasoning struct {
+	Effort  string `json:"effort,omitempty"`
+	Summary string `json:"summary,omitempty"`
+}
+
+// ORIncompleteDetails represents details about why a response was incomplete
+type ORIncompleteDetails struct {
+	Reason string `json:"reason"`
+}
+
+// ORStreamEvent represents a streaming event
+// Note: Fields like delta, text, logprobs should be set explicitly for events that require them
+// The sendSSEEvent function uses a custom serializer to handle conditional field inclusion
+type ORStreamEvent struct {
+	Type            string              `json:"type"`
+	SequenceNumber  int                 `json:"sequence_number"`
+	Response        *ORResponseResource `json:"response,omitempty"`
+	OutputIndex     *int                `json:"output_index,omitempty"`
+	ContentIndex    *int                `json:"content_index,omitempty"`
+	SummaryIndex    *int                `json:"summary_index,omitempty"`
+	ItemID          string              `json:"item_id,omitempty"`
+	Item            *ORItemField        `json:"item,omitempty"`
+	Part            *ORContentPart      `json:"part,omitempty"`
+	Delta           *string             `json:"delta,omitempty"`     // Pointer to distinguish unset from empty
+	Text            *string             `json:"text,omitempty"`      // Pointer to distinguish unset from empty
+	Arguments       *string             `json:"arguments,omitempty"` // Pointer to distinguish unset from empty
+	Refusal         string              `json:"refusal,omitempty"`
+	Error           *ORErrorPayload     `json:"error,omitempty"`
+	Logprobs        *[]ORLogProb        `json:"logprobs,omitempty"` // Pointer to distinguish unset from empty
+	Obfuscation     string              `json:"obfuscation,omitempty"`
+	Annotation      *ORAnnotation       `json:"annotation,omitempty"`
+	AnnotationIndex *int                `json:"annotation_index,omitempty"`
+}
+
+// ORErrorPayload represents an error payload in streaming events
+type ORErrorPayload struct {
+	Type    string            `json:"type"`
+	Code    string            `json:"code,omitempty"`
+	Message string            `json:"message"`
+	Param   string            `json:"param,omitempty"`
+	Headers map[string]string `json:"headers,omitempty"`
+}
+
+// ORLogProb represents log probability information
+type ORLogProb struct {
+	Token       string         `json:"token"`
+	Logprob     float64        `json:"logprob"`
+	Bytes       []int          `json:"bytes"`
+	TopLogprobs []ORTopLogProb `json:"top_logprobs,omitempty"`
+}
+
+// ORTopLogProb represents a top log probability
+type ORTopLogProb struct {
+	Token   string  `json:"token"`
+	Logprob float64 `json:"logprob"`
+	Bytes   []int   `json:"bytes"`
+}
+
+// ORAnnotation represents an annotation (e.g., URL citation)
+type ORAnnotation struct {
+	Type       string `json:"type"` // url_citation
+	StartIndex int    `json:"start_index"`
+	EndIndex   int    `json:"end_index"`
+	URL        string `json:"url"`
+	Title      string `json:"title"`
+}
+
+// ORContentPartWithLogprobs creates an output_text content part with logprobs converted from OpenAI format
+func ORContentPartWithLogprobs(text string, logprobs *Logprobs) ORContentPart {
+	orLogprobs := []ORLogProb{}
+
+	// Convert OpenAI-style logprobs to Open Responses format
+	if logprobs != nil && len(logprobs.Content) > 0 {
+		for _, lp := range logprobs.Content {
+			// Convert top logprobs
+			topLPs := []ORTopLogProb{}
+			for _, tlp := range lp.TopLogprobs {
+				topLPs = append(topLPs, ORTopLogProb{
+					Token:   tlp.Token,
+					Logprob: tlp.Logprob,
+					Bytes:   tlp.Bytes,
+				})
+			}
+
+			orLogprobs = append(orLogprobs, ORLogProb{
+				Token:       lp.Token,
+				Logprob:     lp.Logprob,
+				Bytes:       lp.Bytes,
+				TopLogprobs: topLPs,
+			})
+		}
+	}
+
+	return ORContentPart{
+		Type:        "output_text",
+		Text:        text,
+		Annotations: []ORAnnotation{}, // REQUIRED - must always be present as array (empty if none)
+		Logprobs:    orLogprobs,       // REQUIRED - must always be present as array (empty if none)
+	}
+}

docs/content/features/text-generation.md

@@ -72,6 +72,359 @@ You can list all the models available with:
 curl http://localhost:8080/v1/models
 ```
 
+### Anthropic Messages API
+
+LocalAI supports the Anthropic Messages API, which is compatible with Claude clients. This endpoint provides a structured way to send messages and receive responses, with support for tools, streaming, and multimodal content.
+
+**Endpoint:** `POST /v1/messages` or `POST /messages`
+
+**Reference:** https://docs.anthropic.com/claude/reference/messages_post
+
+#### Basic Usage
+
+```bash
+curl http://localhost:8080/v1/messages \
+  -H "Content-Type: application/json" \
+  -H "anthropic-version: 2023-06-01" \
+  -d '{
+    "model": "ggml-koala-7b-model-q4_0-r2.bin",
+    "max_tokens": 1024,
+    "messages": [
+      {"role": "user", "content": "Say this is a test!"}
+    ]
+  }'
+```
+
+#### Request Parameters
+
+| Parameter | Type | Required | Description |
+|-----------|------|----------|-------------|
+| `model` | string | Yes | The model identifier |
+| `messages` | array | Yes | Array of message objects with `role` and `content` |
+| `max_tokens` | integer | Yes | Maximum number of tokens to generate (must be > 0) |
+| `system` | string | No | System message to set the assistant's behavior |
+| `temperature` | float | No | Sampling temperature (0.0 to 1.0) |
+| `top_p` | float | No | Nucleus sampling parameter |
+| `top_k` | integer | No | Top-k sampling parameter |
+| `stop_sequences` | array | No | Array of strings that will stop generation |
+| `stream` | boolean | No | Enable streaming responses |
+| `tools` | array | No | Array of tool definitions for function calling |
+| `tool_choice` | string/object | No | Tool choice strategy: "auto", "any", "none", or specific tool |
+| `metadata` | object | No | Custom metadata to attach to the request |
+
+#### Message Format
+
+Messages can contain text or structured content blocks:
+
+```bash
+curl http://localhost:8080/v1/messages \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "ggml-koala-7b-model-q4_0-r2.bin",
+    "max_tokens": 1024,
+    "messages": [
+      {
+        "role": "user",
+        "content": [
+          {
+            "type": "text",
+            "text": "What is in this image?"
+          },
+          {
+            "type": "image",
+            "source": {
+              "type": "base64",
+              "media_type": "image/jpeg",
+              "data": "base64_encoded_image_data"
+            }
+          }
+        ]
+      }
+    ]
+  }'
+```
+
+#### Tool Calling
+
+The Anthropic API supports function calling through tools:
+
+```bash
+curl http://localhost:8080/v1/messages \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "ggml-koala-7b-model-q4_0-r2.bin",
+    "max_tokens": 1024,
+    "tools": [
+      {
+        "name": "get_weather",
+        "description": "Get the current weather",
+        "input_schema": {
+          "type": "object",
+          "properties": {
+            "location": {
+              "type": "string",
+              "description": "The city and state"
+            }
+          },
+          "required": ["location"]
+        }
+      }
+    ],
+    "tool_choice": "auto",
+    "messages": [
+      {"role": "user", "content": "What is the weather in San Francisco?"}
+    ]
+  }'
+```
+
+#### Streaming
+
+Enable streaming responses by setting `stream: true`:
+
+```bash
+curl http://localhost:8080/v1/messages \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "ggml-koala-7b-model-q4_0-r2.bin",
+    "max_tokens": 1024,
+    "stream": true,
+    "messages": [
+      {"role": "user", "content": "Tell me a story"}
+    ]
+  }'
+```
+
+Streaming responses use Server-Sent Events (SSE) format with event types: `message_start`, `content_block_start`, `content_block_delta`, `content_block_stop`, `message_delta`, and `message_stop`.
+
+#### Response Format
+
+```json
+{
+  "id": "msg_abc123",
+  "type": "message",
+  "role": "assistant",
+  "content": [
+    {
+      "type": "text",
+      "text": "This is a test!"
+    }
+  ],
+  "model": "ggml-koala-7b-model-q4_0-r2.bin",
+  "stop_reason": "end_turn",
+  "usage": {
+    "input_tokens": 10,
+    "output_tokens": 5
+  }
+}
+```
+
+### Open Responses API
+
+LocalAI supports the Open Responses API specification, which provides a standardized interface for AI model interactions with support for background processing, streaming, tool calling, and advanced features like reasoning.
+
+**Endpoint:** `POST /v1/responses` or `POST /responses`
+
+**Reference:** https://www.openresponses.org/specification
+
+#### Basic Usage
+
+```bash
+curl http://localhost:8080/v1/responses \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "ggml-koala-7b-model-q4_0-r2.bin",
+    "input": "Say this is a test!",
+    "max_output_tokens": 1024
+  }'
+```
+
+#### Request Parameters
+
+| Parameter | Type | Required | Description |
+|-----------|------|----------|-------------|
+| `model` | string | Yes | The model identifier |
+| `input` | string/array | Yes | Input text or array of input items |
+| `max_output_tokens` | integer | No | Maximum number of tokens to generate |
+| `temperature` | float | No | Sampling temperature |
+| `top_p` | float | No | Nucleus sampling parameter |
+| `instructions` | string | No | System instructions |
+| `tools` | array | No | Array of tool definitions |
+| `tool_choice` | string/object | No | Tool choice: "auto", "required", "none", or specific tool |
+| `stream` | boolean | No | Enable streaming responses |
+| `background` | boolean | No | Run request in background (returns immediately) |
+| `store` | boolean | No | Whether to store the response |
+| `reasoning` | object | No | Reasoning configuration with `effort` and `summary` |
+| `parallel_tool_calls` | boolean | No | Allow parallel tool calls |
+| `max_tool_calls` | integer | No | Maximum number of tool calls |
+| `presence_penalty` | float | No | Presence penalty (-2.0 to 2.0) |
+| `frequency_penalty` | float | No | Frequency penalty (-2.0 to 2.0) |
+| `top_logprobs` | integer | No | Number of top logprobs to return |
+| `truncation` | string | No | Truncation mode: "auto" or "disabled" |
+| `text_format` | object | No | Text format configuration |
+| `metadata` | object | No | Custom metadata |
+
+#### Input Format
+
+Input can be a simple string or an array of structured items:
+
+```bash
+curl http://localhost:8080/v1/responses \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "ggml-koala-7b-model-q4_0-r2.bin",
+    "input": [
+      {
+        "type": "message",
+        "role": "user",
+        "content": "What is the weather?"
+      }
+    ],
+    "max_output_tokens": 1024
+  }'
+```
+
+#### Background Processing
+
+Run requests in the background for long-running tasks:
+
+```bash
+curl http://localhost:8080/v1/responses \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "ggml-koala-7b-model-q4_0-r2.bin",
+    "input": "Generate a long story",
+    "max_output_tokens": 4096,
+    "background": true
+  }'
+```
+
+The response will include a response ID that can be used to poll for completion:
+
+```json
+{
+  "id": "resp_abc123",
+  "object": "response",
+  "status": "in_progress",
+  "created_at": 1234567890
+}
+```
+
+#### Retrieving Background Responses
+
+Use the GET endpoint to retrieve background responses:
+
+```bash
+# Get response by ID
+curl http://localhost:8080/v1/responses/resp_abc123
+
+# Resume streaming with query parameters
+curl "http://localhost:8080/v1/responses/resp_abc123?stream=true&starting_after=10"
+```
+
+#### Canceling Background Responses
+
+Cancel a background response that's still in progress:
+
+```bash
+curl -X POST http://localhost:8080/v1/responses/resp_abc123/cancel
+```
+
+#### Tool Calling
+
+Open Responses API supports function calling with tools:
+
+```bash
+curl http://localhost:8080/v1/responses \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "ggml-koala-7b-model-q4_0-r2.bin",
+    "input": "What is the weather in San Francisco?",
+    "tools": [
+      {
+        "type": "function",
+        "name": "get_weather",
+        "description": "Get the current weather",
+        "parameters": {
+          "type": "object",
+          "properties": {
+            "location": {
+              "type": "string",
+              "description": "The city and state"
+            }
+          },
+          "required": ["location"]
+        }
+      }
+    ],
+    "tool_choice": "auto",
+    "max_output_tokens": 1024
+  }'
+```
+
+#### Reasoning Configuration
+
+Configure reasoning effort and summary style:
+
+```bash
+curl http://localhost:8080/v1/responses \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "ggml-koala-7b-model-q4_0-r2.bin",
+    "input": "Solve this complex problem step by step",
+    "reasoning": {
+      "effort": "high",
+      "summary": "detailed"
+    },
+    "max_output_tokens": 2048
+  }'
+```
+
+#### Response Format
+
+```json
+{
+  "id": "resp_abc123",
+  "object": "response",
+  "created_at": 1234567890,
+  "completed_at": 1234567895,
+  "status": "completed",
+  "model": "ggml-koala-7b-model-q4_0-r2.bin",
+  "output": [
+    {
+      "type": "message",
+      "id": "msg_001",
+      "role": "assistant",
+      "content": [
+        {
+          "type": "output_text",
+          "text": "This is a test!",
+          "annotations": [],
+          "logprobs": []
+        }
+      ],
+      "status": "completed"
+    }
+  ],
+  "error": null,
+  "incomplete_details": null,
+  "temperature": 0.7,
+  "top_p": 1.0,
+  "presence_penalty": 0.0,
+  "frequency_penalty": 0.0,
+  "usage": {
+    "input_tokens": 10,
+    "output_tokens": 5,
+    "total_tokens": 15,
+    "input_tokens_details": {
+      "cached_tokens": 0
+    },
+    "output_tokens_details": {
+      "reasoning_tokens": 0
+    }
+  }
+}
+```
+
 ## Backends
 
 ### RWKV

docs/content/features/text-to-audio.md

@@ -164,6 +164,57 @@ curl http://localhost:8080/tts -H "Content-Type: application/json" -d '{
    }' | aplay
 ```
 
+### Pocket TTS
+
+[Pocket TTS](https://github.com/kyutai-labs/pocket-tts) is a lightweight text-to-speech model designed to run efficiently on CPUs. It supports voice cloning through HuggingFace voice URLs or local audio files.
+
+#### Setup
+
+Install the `pocket-tts` model in the Model gallery or run `local-ai run models install pocket-tts`.
+
+#### Usage
+
+Use the tts endpoint by specifying the pocket-tts backend:
+
+```
+curl http://localhost:8080/tts -H "Content-Type: application/json" -d '{         
+     "model": "pocket-tts",
+     "input":"Hello world, this is a test."
+   }' | aplay
+```
+
+#### Voice cloning
+
+Pocket TTS supports voice cloning through built-in voice names, HuggingFace URLs, or local audio files. You can configure a model with a specific voice:
+
+```yaml
+name: pocket-tts
+backend: pocket-tts
+tts:
+  voice: "azelma"  # Built-in voice name
+  # Or use HuggingFace URL: "hf://kyutai/tts-voices/alba-mackenna/casual.wav"
+  # Or use local file path: "path/to/voice.wav"
+  # Available built-in voices: alba, marius, javert, jean, fantine, cosette, eponine, azelma
+```
+
+You can also pre-load a default voice for faster first generation:
+
+```yaml
+name: pocket-tts
+backend: pocket-tts
+options:
+  - "default_voice:azelma"  # Pre-load this voice when model loads
+```
+
+Then you can use the model:
+
+```
+curl http://localhost:8080/tts -H "Content-Type: application/json" -d '{         
+     "model": "pocket-tts",
+     "input":"Hello world, this is a test."
+   }' | aplay
+```
+
 ### Vall-E-X
 
 [VALL-E-X](https://github.com/Plachtaa/VALL-E-X) is an open source implementation of Microsoft's VALL-E X zero-shot TTS model.

docs/content/getting-started/try-it-out.md

@@ -112,6 +112,66 @@ curl http://localhost:8080/v1/chat/completions \
 
 </details>
 
+### Anthropic Messages API
+
+LocalAI supports the Anthropic Messages API for Claude-compatible models. [Anthropic documentation](https://docs.anthropic.com/claude/reference/messages_post).
+
+<details>
+
+```bash
+curl http://localhost:8080/v1/messages \
+  -H "Content-Type: application/json" \
+  -H "anthropic-version: 2023-06-01" \
+  -d '{
+    "model": "gpt-4",
+    "max_tokens": 1024,
+    "messages": [
+      {"role": "user", "content": "How are you doing?"}
+    ],
+    "temperature": 0.7
+  }'
+```
+
+</details>
+
+### Open Responses API
+
+LocalAI supports the Open Responses API specification with support for background processing, streaming, and advanced features. [Open Responses documentation](https://www.openresponses.org/specification).
+
+<details>
+
+```bash
+curl http://localhost:8080/v1/responses \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "gpt-4",
+    "input": "Say this is a test!",
+    "max_output_tokens": 1024,
+    "temperature": 0.7
+  }'
+```
+
+For background processing:
+
+```bash
+curl http://localhost:8080/v1/responses \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "gpt-4",
+    "input": "Generate a long story",
+    "max_output_tokens": 4096,
+    "background": true
+  }'
+```
+
+Then retrieve the response:
+
+```bash
+curl http://localhost:8080/v1/responses/<response_id>
+```
+
+</details>
+
 ### Image Generation
 
 Creates an image given a prompt. [OpenAI documentation](https://platform.openai.com/docs/api-reference/images/create).

docs/content/installation/_index.en.md

@@ -20,7 +20,7 @@ Choose the installation method that best suits your needs:
 
 1. **[Docker](docker/)** ⭐ **Recommended** - Works on all platforms, easiest setup
 2. **[macOS](macos/)** - Download and install the DMG application
-3. **[Linux](linux/)** - Install on Linux using the one-liner script or binaries
+3. **[Linux](linux/)** - Install on Linux using binaries (install.sh script currently has issues - see [issue #8032](https://github.com/mudler/LocalAI/issues/8032))
 4. **[Kubernetes](kubernetes/)** - Deploy LocalAI on Kubernetes clusters
 5. **[Build from Source](build/)** - Build LocalAI from source code
 
@@ -36,6 +36,6 @@ This will start LocalAI. The API will be available at `http://localhost:8080`. F
 
 For other platforms:
 - **macOS**: Download the [DMG](macos/)
-- **Linux**: Use the `curl https://localai.io/install.sh | sh` [one-liner](linux/)
+- **Linux**: See the [Linux installation guide](linux/) for installation options. **Note:** The `install.sh` script is currently experiencing issues - see [issue #8032](https://github.com/mudler/LocalAI/issues/8032) for details.
 
 For detailed instructions, see the [Docker installation guide](docker/).

docs/content/installation/linux.md

@@ -6,7 +6,11 @@ url: '/installation/linux/'
 ---
 
 
-## One-Line Installer (Recommended)
+## One-Line Installer
+
+{{% notice warning %}}
+**The `install.sh` script is currently experiencing issues and may produce broken or misconfigured installations. Please use alternative installation methods (Docker or manual binary installation) until [issue #8032](https://github.com/mudler/LocalAI/issues/8032) is resolved.**
+{{% /notice %}}
 
 The fastest way to install LocalAI on Linux is with the installation script:
 

docs/content/reference/cli-reference.md

@@ -97,7 +97,7 @@ For more information on VRAM management, see [VRAM and Memory Management]({{%rel
 | `--opaque-errors` | `false` | If true, all error responses are replaced with blank 500 errors. This is intended only for hardening against information leaks and is normally not recommended | `$LOCALAI_OPAQUE_ERRORS` |
 | `--use-subtle-key-comparison` | `false` | If true, API Key validation comparisons will be performed using constant-time comparisons rather than simple equality. This trades off performance on each request for resilience against timing attacks | `$LOCALAI_SUBTLE_KEY_COMPARISON` |
 | `--disable-api-key-requirement-for-http-get` | `false` | If true, a valid API key is not required to issue GET requests to portions of the web UI. This should only be enabled in secure testing environments | `$LOCALAI_DISABLE_API_KEY_REQUIREMENT_FOR_HTTP_GET` |
-| `--http-get-exempted-endpoints` | `^/$,^/browse/?$,^/talk/?$,^/p2p/?$,^/chat/?$,^/text2image/?$,^/tts/?$,^/static/.*$,^/swagger.*$` | If `--disable-api-key-requirement-for-http-get` is overridden to true, this is the list of endpoints to exempt. Only adjust this in case of a security incident or as a result of a personal security posture review | `$LOCALAI_HTTP_GET_EXEMPTED_ENDPOINTS` |
+| `--http-get-exempted-endpoints` | `^/$,^/browse/?$,^/talk/?$,^/p2p/?$,^/chat/?$,^/image/?$,^/text2image/?$,^/tts/?$,^/static/.*$,^/swagger.*$` | If `--disable-api-key-requirement-for-http-get` is overridden to true, this is the list of endpoints to exempt. Only adjust this in case of a security incident or as a result of a personal security posture review | `$LOCALAI_HTTP_GET_EXEMPTED_ENDPOINTS` |
 
 ## P2P Flags
 

docs/content/reference/compatibility-table.md

@@ -42,6 +42,7 @@ LocalAI will attempt to automatically load models which are not explicitly confi
 | [silero-vad](https://github.com/snakers4/silero-vad) with [Golang bindings](https://github.com/streamer45/silero-vad-go) | Silero VAD    | no                       | Voice Activity Detection    | no                               | no                   | CPU |
 | [neutts](https://github.com/neuphonic/neuttsair) | NeuTTSAir    | no                       | Text-to-speech with voice cloning    | no                               | no                   | CUDA 12/13, ROCm, CPU |
 | [vibevoice](https://github.com/microsoft/VibeVoice) | VibeVoice-Realtime    | no                       | Real-time text-to-speech with voice cloning    | no                               | no                   | CUDA 12/13, ROCm, Intel, CPU |
+| [pocket-tts](https://github.com/kyutai-labs/pocket-tts) | Pocket TTS    | no                       | Lightweight CPU-based text-to-speech with voice cloning    | no                               | no                   | CUDA 12/13, ROCm, Intel, CPU |
 | [mlx-audio](https://github.com/Blaizzy/mlx-audio) | MLX | no                       | Text-tospeech    | no                               | no                   | Metal (Apple Silicon) |
 
 ## Image & Video Generation

docs/data/version.json

@@ -1,3 +1,3 @@
 {
-  "version": "v3.9.0"
+  "version": "v3.10.0"
 }

gallery/index.yaml

@@ -29,6 +29,7 @@
 
     This description emphasizes its capabilities, efficiency, and versatility for multimodal search tasks.
   overrides:
+    reranking: true
     parameters:
       model: llama-cpp/models/Qwen3-VL-Reranker-8B.Q4_K_M.gguf
     name: Qwen3-VL-Reranker-8B-GGUF
@@ -478,6 +479,16 @@
     - filename: voices/streaming_model/en-Davis_man.pt
       uri: https://raw.githubusercontent.com/microsoft/VibeVoice/main/demo/voices/streaming_model/en-Davis_man.pt
       sha256: 67561d63bfa2153616e4c02fd967007c182593fc53738a6ad94bf5f84e8832ac
+- &pocket-tts
+  url: "github:mudler/LocalAI/gallery/pocket-tts.yaml@master"
+  icon: https://avatars.githubusercontent.com/u/6154722?s=200&v=4
+  license: mit
+  tags:
+    - text-to-speech
+    - TTS
+  name: "pocket-tts"
+  urls:
+    - https://github.com/kyutai-labs/pocket-tts
 - &qwen3vl
   url: "github:mudler/LocalAI/gallery/qwen3.yaml@master"
   icon: https://cdn-avatars.huggingface.co/v1/production/uploads/620760a26e3b7210c2ff1943/-s1gyJfvbE1RgO5iBeNOi.png
@@ -1237,6 +1248,63 @@
       cuda: true
       pipeline_type: QwenImageEditPipeline
       enable_parameters: num_inference_steps,image
+- &ltx2
+  name: "ltx-2"
+  url: "github:mudler/LocalAI/gallery/virtual.yaml@master"
+  urls:
+    - https://huggingface.co/Lightricks/LTX-2
+  license: ltx-2-community-license-agreement
+  tags:
+    - diffusers
+    - gpu
+    - image-to-video
+    - video-generation
+    - audio-video
+  description: |
+    **LTX-2** is a DiT-based audio-video foundation model designed to generate synchronized video and audio within a single model. It brings together the core building blocks of modern video generation, with open weights and a focus on practical, local execution.
+
+    **Key Features:**
+    - **Joint Audio-Video Generation**: Generates synchronized video and audio in a single model
+    - **Image-to-Video**: Converts static images into dynamic videos with matching audio
+    - **High Quality**: Produces realistic video with natural motion and synchronized audio
+    - **Open Weights**: Available under the LTX-2 Community License Agreement
+
+    **Model Details:**
+    - **Model Type**: Diffusion-based audio-video foundation model
+    - **Architecture**: DiT (Diffusion Transformer) based
+    - **Developed by**: Lightricks
+    - **Paper**: [LTX-2: Efficient Joint Audio-Visual Foundation Model](https://arxiv.org/abs/2601.03233)
+
+    **Usage Tips:**
+    - Width & height settings must be divisible by 32
+    - Frame count must be divisible by 8 + 1 (e.g., 9, 17, 25, 33, 41, 49, 57, 65, 73, 81, 89, 97, 105, 113, 121)
+    - Recommended settings: width=768, height=512, num_frames=121, frame_rate=24.0
+    - For best results, use detailed prompts describing motion and scene dynamics
+
+    **Limitations:**
+    - This model is not intended or able to provide factual information
+    - Prompt following is heavily influenced by the prompting-style
+    - When generating audio without speech, the audio may be of lower quality
+
+    **Citation:**
+    ```bibtex
+    @article{hacohen2025ltx2,
+      title={LTX-2: Efficient Joint Audio-Visual Foundation Model},
+      author={HaCohen, Yoav and Brazowski, Benny and Chiprut, Nisan and others},
+      journal={arXiv preprint arXiv:2601.03233},
+      year={2025}
+    }
+    ```
+  overrides:
+    backend: diffusers
+    low_vram: true
+    parameters:
+      model: Lightricks/LTX-2
+    diffusers:
+      cuda: true
+      pipeline_type: LTX2ImageToVideoPipeline
+    options:
+      - torch_dtype:bf16
 - &gptoss
   name: "gpt-oss-20b"
   url: "github:mudler/LocalAI/gallery/harmony.yaml@master"
@@ -3755,6 +3823,41 @@
     - filename: boomerang-qwen3-4.9B.Q4_K_M.gguf
       sha256: 11e6c068351d104dee31dd63550e5e2fc9be70467c1cfc07a6f84030cb701537
       uri: huggingface://mradermacher/boomerang-qwen3-4.9B-GGUF/boomerang-qwen3-4.9B.Q4_K_M.gguf
+- !!merge <<: *qwen3
+  name: "qwen3-coder-30b-a3b-instruct"
+  icon: https://cdn-avatars.huggingface.co/v1/production/uploads/620760a26e3b7210c2ff1943/-s1gyJfvbE1RgO5iBeNOi.png
+  url: "github:mudler/LocalAI/gallery/qwen3.yaml@master"
+  urls:
+    - https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct
+    - https://huggingface.co/unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF
+  description: |
+    Qwen3-Coder is available in multiple sizes. Today, we're excited to introduce Qwen3-Coder-30B-A3B-Instruct. This streamlined model maintains impressive performance and efficiency, featuring the following key enhancements:
+
+        - Significant Performance among open models on Agentic Coding, Agentic Browser-Use, and other foundational coding tasks.
+        - Long-context Capabilities with native support for 256K tokens, extendable up to 1M tokens using Yarn, optimized for repository-scale understanding.
+        - Agentic Coding supporting for most platform such as Qwen Code, CLINE, featuring a specially designed function call format.
+
+
+    Model Overview:
+    Qwen3-Coder-30B-A3B-Instruct has the following features:
+
+        - Type: Causal Language Models
+        - Training Stage: Pretraining & Post-training
+        - Number of Parameters: 30.5B in total and 3.3B activated
+        - Number of Layers: 48
+        - Number of Attention Heads (GQA): 32 for Q and 4 for KV
+        - Number of Experts: 128
+        - Number of Activated Experts: 8
+        - Context Length: 262,144 natively.
+
+    NOTE: This model supports only non-thinking mode and does not generate <think></think> blocks in its output. Meanwhile, specifying enable_thinking=False is no longer required.
+  overrides:
+    parameters:
+      model: Qwen3-Coder-30B-A3B-Instruct-Q4_K_M.gguf
+  files:
+    - filename: Qwen3-Coder-30B-A3B-Instruct-Q4_K_M.gguf
+      sha256: fadc3e5f8d42bf7e894a785b05082e47daee4df26680389817e2093056f088ad
+      uri: huggingface://unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF/Qwen3-Coder-30B-A3B-Instruct-Q4_K_M.gguf
 - &gemma3
   url: "github:mudler/LocalAI/gallery/gemma.yaml@master"
   name: "gemma-3-27b-it"

gallery/pocket-tts.yaml

@@ -0,0 +1,34 @@
+---
+name: localai
+
+config_file: |-
+    name: pocket-tts
+    backend: pocket-tts
+    description: |
+      Pocket TTS is a lightweight text-to-speech model designed to run efficiently on CPUs.
+      This model supports voice cloning through HuggingFace voice URLs or local audio files.
+
+    parameters:
+      model: ""
+
+    # TTS configuration
+    tts:
+      # Voice selection - can be:
+      # 1. Built-in voice name (e.g., "alba", "marius", "javert", "jean", "fantine", "cosette", "eponine", "azelma")
+      # 2. HuggingFace URL (e.g., "hf://kyutai/tts-voices/alba-mackenna/casual.wav")
+      # 3. Local file path (relative to model directory or absolute)
+      # voice: "azelma"
+      # Alternative: use audio_path to specify a voice file directly
+      # audio_path: "hf://kyutai/tts-voices/alba-mackenna/casual.wav"
+
+    known_usecases:
+      - tts
+
+    # Backend-specific options
+    # These are passed as "key:value" strings to the backend
+    options:
+      # Default voice to pre-load (optional)
+      # Can be a voice name or HuggingFace URL
+      # If set, this voice will be loaded when the model loads for faster first generation
+      - "default_voice:azelma"
+      # - "default_voice:hf://kyutai/tts-voices/alba-mackenna/casual.wav"

go.mod

@@ -6,7 +6,7 @@ toolchain go1.24.5
 
 require (
 	dario.cat/mergo v1.0.2
-	fyne.io/fyne/v2 v2.7.1
+	fyne.io/fyne/v2 v2.7.2
 	github.com/Masterminds/sprig/v3 v3.3.0
 	github.com/alecthomas/kong v1.13.0
 	github.com/anthropics/anthropic-sdk-go v1.19.0
@@ -20,7 +20,7 @@ require (
 	github.com/gofrs/flock v0.13.0
 	github.com/google/go-containerregistry v0.20.7
 	github.com/google/uuid v1.6.0
-	github.com/gpustack/gguf-parser-go v0.22.1
+	github.com/gpustack/gguf-parser-go v0.23.1
 	github.com/hpcloud/tail v1.0.0
 	github.com/ipfs/go-log v1.0.5
 	github.com/jaypipes/ghw v0.21.2
@@ -32,13 +32,13 @@ require (
 	github.com/mholt/archiver/v3 v3.5.1
 	github.com/microcosm-cc/bluemonday v1.0.27
 	github.com/modelcontextprotocol/go-sdk v1.2.0
-	github.com/mudler/cogito v0.7.2
+	github.com/mudler/cogito v0.8.1
 	github.com/mudler/edgevpn v0.31.1
-	github.com/mudler/go-processmanager v0.0.0-20240820160718-8b802d3ecf82
+	github.com/mudler/go-processmanager v0.1.0
 	github.com/mudler/memory v0.0.0-20251216220809-d1256471a6c2
 	github.com/mudler/xlog v0.0.5
-	github.com/onsi/ginkgo/v2 v2.27.3
-	github.com/onsi/gomega v1.38.3
+	github.com/onsi/ginkgo/v2 v2.27.5
+	github.com/onsi/gomega v1.39.0
 	github.com/otiai10/copy v1.14.1
 	github.com/otiai10/openaigo v1.7.0
 	github.com/phayes/freeport v0.0.0-20220201140144-74d24b5ae9f5
@@ -59,7 +59,6 @@ require (
 	go.opentelemetry.io/otel/metric v1.39.0
 	go.opentelemetry.io/otel/sdk/metric v1.39.0
 	google.golang.org/grpc v1.78.0
-	google.golang.org/protobuf v1.36.10
 	gopkg.in/yaml.v2 v2.4.0
 	gopkg.in/yaml.v3 v3.0.1
 	oras.land/oras-go/v2 v2.6.0
@@ -74,10 +73,11 @@ require (
 	github.com/tidwall/pretty v1.2.1 // indirect
 	github.com/tidwall/sjson v1.2.5 // indirect
 	github.com/valyala/fasttemplate v1.2.2 // indirect
+	google.golang.org/protobuf v1.36.10 // indirect
 )
 
 require (
-	fyne.io/systray v1.11.1-0.20250603113521-ca66a66d8b58 // indirect
+	fyne.io/systray v1.12.0 // indirect
 	github.com/BurntSushi/toml v1.5.0 // indirect
 	github.com/charmbracelet/colorprofile v0.2.3-0.20250311203215-f60798e515dc // indirect
 	github.com/charmbracelet/lipgloss v1.1.1-0.20250404203927-76690c660834 // indirect
@@ -324,7 +324,7 @@ require (
 	golang.org/x/exp v0.0.0-20250606033433-dcc06ee1d476 // indirect
 	golang.org/x/mod v0.30.0 // indirect
 	golang.org/x/sync v0.19.0 // indirect
-	golang.org/x/sys v0.39.0 // indirect
+	golang.org/x/sys v0.40.0 // indirect
 	golang.org/x/term v0.38.0 // indirect
 	golang.org/x/text v0.32.0 // indirect
 	golang.org/x/tools v0.39.0 // indirect

go.sum

@@ -8,10 +8,10 @@ dmitri.shuralyov.com/app/changes v0.0.0-20180602232624-0a106ad413e3/go.mod h1:Yl
 dmitri.shuralyov.com/html/belt v0.0.0-20180602232347-f7d459c86be0/go.mod h1:JLBrvjyP0v+ecvNYvCpyZgu5/xkfAUhi6wJj28eUfSU=
 dmitri.shuralyov.com/service/change v0.0.0-20181023043359-a85b471d5412/go.mod h1:a1inKt/atXimZ4Mv927x+r7UpyzRUf4emIoiiSC2TN4=
 dmitri.shuralyov.com/state v0.0.0-20180228185332-28bcc343414c/go.mod h1:0PRwlb0D6DFvNNtx+9ybjezNCa8XF0xaYcETyp6rHWU=
-fyne.io/fyne/v2 v2.7.1 h1:ja7rNHWWEooha4XBIZNnPP8tVFwmTfwMJdpZmLxm2Zc=
-fyne.io/fyne/v2 v2.7.1/go.mod h1:xClVlrhxl7D+LT+BWYmcrW4Nf+dJTvkhnPgji7spAwE=
-fyne.io/systray v1.11.1-0.20250603113521-ca66a66d8b58 h1:eA5/u2XRd8OUkoMqEv3IBlFYSruNlXD8bRHDiqm0VNI=
-fyne.io/systray v1.11.1-0.20250603113521-ca66a66d8b58/go.mod h1:RVwqP9nYMo7h5zViCBHri2FgjXF7H2cub7MAq4NSoLs=
+fyne.io/fyne/v2 v2.7.2 h1:XiNpWkn0PzX43ZCjbb0QYGg1RCxVbugwfVgikWZBCMw=
+fyne.io/fyne/v2 v2.7.2/go.mod h1:PXbqY3mQmJV3J1NRUR2VbVgUUx3vgvhuFJxyjRK/4Ug=
+fyne.io/systray v1.12.0 h1:CA1Kk0e2zwFlxtc02L3QFSiIbxJ/P0n582YrZHT7aTM=
+fyne.io/systray v1.12.0/go.mod h1:RVwqP9nYMo7h5zViCBHri2FgjXF7H2cub7MAq4NSoLs=
 git.apache.org/thrift.git v0.0.0-20180902110319-2566ecd5d999/go.mod h1:fPE2ZNJGynbRyZ4dJvy6G277gSllfV2HJqblrnkyeyg=
 github.com/AdaLogics/go-fuzz-headers v0.0.0-20240806141605-e8a1dd7889d6 h1:He8afgbRMd7mFxO99hRNu+6tazq8nFF9lIwo9JFroBk=
 github.com/AdaLogics/go-fuzz-headers v0.0.0-20240806141605-e8a1dd7889d6/go.mod h1:8o94RPi1/7XTJvwPpRSzSUedZrtlirdB3r9Z20bi2f8=
@@ -292,8 +292,8 @@ github.com/gorilla/css v1.0.1 h1:ntNaBIghp6JmvWnxbZKANoLyuXTPZ4cAMlo6RyhlbO8=
 github.com/gorilla/css v1.0.1/go.mod h1:BvnYkspnSzMmwRK+b8/xgNPLiIuNZr6vbZBTPQ2A3b0=
 github.com/gorilla/websocket v1.5.3 h1:saDtZ6Pbx/0u+bgYQ3q96pZgCzfhKXGPqt7kZ72aNNg=
 github.com/gorilla/websocket v1.5.3/go.mod h1:YR8l580nyteQvAITg2hZ9XVh4b55+EU/adAjf1fMHhE=
-github.com/gpustack/gguf-parser-go v0.22.1 h1:FRnEDWqT0Rcplr/R9ctCRSN2+3DhVsf6dnR5/i9JA4E=
-github.com/gpustack/gguf-parser-go v0.22.1/go.mod h1:y4TwTtDqFWTK+xvprOjRUh+dowgU2TKCX37vRKvGiZ0=
+github.com/gpustack/gguf-parser-go v0.23.1 h1:0U7DOrsi7ryx2L/dlMy+BSQ5bJV4AuMEIgGBs4RK46A=
+github.com/gpustack/gguf-parser-go v0.23.1/go.mod h1:y4TwTtDqFWTK+xvprOjRUh+dowgU2TKCX37vRKvGiZ0=
 github.com/gregjones/httpcache v0.0.0-20180305231024-9cad4c3443a7/go.mod h1:FecbI9+v66THATjSRHfNgh1IVFe/9kFxbXtjV0ctIMA=
 github.com/grpc-ecosystem/grpc-gateway v1.5.0/go.mod h1:RSKVYQBd5MCa4OVpNdGskqpgL2+G+NZTnrVHpWWfpdw=
 github.com/grpc-ecosystem/grpc-gateway v1.16.0 h1:gmcG1KaJ57LophUzW0Hy8NmPhnMZb4M0+kPpLofRdBo=
@@ -507,14 +507,14 @@ github.com/morikuni/aec v1.0.0/go.mod h1:BbKIizmSmc5MMPqRYbxO4ZU0S0+P200+tUnFx7P
 github.com/mr-tron/base58 v1.1.2/go.mod h1:BinMc/sQntlIE1frQmRFPUoPA1Zkr8VRgBdjWI2mNwc=
 github.com/mr-tron/base58 v1.2.0 h1:T/HDJBh4ZCPbU39/+c3rRvE0uKBQlU27+QI8LJ4t64o=
 github.com/mr-tron/base58 v1.2.0/go.mod h1:BinMc/sQntlIE1frQmRFPUoPA1Zkr8VRgBdjWI2mNwc=
-github.com/mudler/cogito v0.7.2 h1:J5eHZPsxpoKcnYUfogje5u0nnzGww7ytv7nSn1DMpms=
-github.com/mudler/cogito v0.7.2/go.mod h1:6sfja3lcu2nWRzEc0wwqGNu/eCG3EWgij+8s7xyUeQ4=
+github.com/mudler/cogito v0.8.1 h1:66qPJkAMrq/Vo8AC/PvXWuVxYPhi7X2DQuJIilL8+3I=
+github.com/mudler/cogito v0.8.1/go.mod h1:6sfja3lcu2nWRzEc0wwqGNu/eCG3EWgij+8s7xyUeQ4=
 github.com/mudler/edgevpn v0.31.1 h1:7qegiDWd0kAg6ljhNHxqvp8hbo/6BbzSdbb7/2WZfiY=
 github.com/mudler/edgevpn v0.31.1/go.mod h1:ftV5B0nKFzm4R8vR80UYnCb2nf7lxCRgAALxUEEgCf8=
 github.com/mudler/go-piper v0.0.0-20241023091659-2494246fd9fc h1:RxwneJl1VgvikiX28EkpdAyL4yQVnJMrbquKospjHyA=
 github.com/mudler/go-piper v0.0.0-20241023091659-2494246fd9fc/go.mod h1:O7SwdSWMilAWhBZMK9N9Y/oBDyMMzshE3ju8Xkexwig=
-github.com/mudler/go-processmanager v0.0.0-20240820160718-8b802d3ecf82 h1:FVT07EI8njvsD4tC2Hw8Xhactp5AWhsQWD4oTeQuSAU=
-github.com/mudler/go-processmanager v0.0.0-20240820160718-8b802d3ecf82/go.mod h1:Urp7LG5jylKoDq0663qeBh0pINGcRl35nXdKx82PSoU=
+github.com/mudler/go-processmanager v0.1.0 h1:fcSKgF9U/a1Z7KofAFeZnke5YseadCI5GqL9oT0LS3E=
+github.com/mudler/go-processmanager v0.1.0/go.mod h1:h6kmHUZeafr+k5hRYpGLMzJFH4hItHffgpRo2QIkP+o=
 github.com/mudler/memory v0.0.0-20251216220809-d1256471a6c2 h1:+WHsL/j6EWOMUiMVIOJNKOwSKiQt/qDPc9fePCf87fA=
 github.com/mudler/memory v0.0.0-20251216220809-d1256471a6c2/go.mod h1:EA8Ashhd56o32qN7ouPKFSRUs/Z+LrRCF4v6R2Oarm8=
 github.com/mudler/water v0.0.0-20250808092830-dd90dcf09025 h1:WFLP5FHInarYGXi6B/Ze204x7Xy6q/I4nCZnWEyPHK0=
@@ -561,10 +561,10 @@ github.com/nxadm/tail v1.4.8 h1:nPr65rt6Y5JFSKQO7qToXr7pePgD6Gwiw05lkbyAQTE=
 github.com/nxadm/tail v1.4.8/go.mod h1:+ncqLTQzXmGhMZNUePPaPqPvBxHAIsmXswZKocGu+AU=
 github.com/onsi/ginkgo v1.16.5 h1:8xi0RTUf59SOSfEtZMvwTvXYMzG4gV23XVHOZiXNtnE=
 github.com/onsi/ginkgo v1.16.5/go.mod h1:+E8gABHa3K6zRBolWtd+ROzc/U5bkGt0FwiG042wbpU=
-github.com/onsi/ginkgo/v2 v2.27.3 h1:ICsZJ8JoYafeXFFlFAG75a7CxMsJHwgKwtO+82SE9L8=
-github.com/onsi/ginkgo/v2 v2.27.3/go.mod h1:ArE1D/XhNXBXCBkKOLkbsb2c81dQHCRcF5zwn/ykDRo=
-github.com/onsi/gomega v1.38.3 h1:eTX+W6dobAYfFeGC2PV6RwXRu/MyT+cQguijutvkpSM=
-github.com/onsi/gomega v1.38.3/go.mod h1:ZCU1pkQcXDO5Sl9/VVEGlDyp+zm0m1cmeG5TOzLgdh4=
+github.com/onsi/ginkgo/v2 v2.27.5 h1:ZeVgZMx2PDMdJm/+w5fE/OyG6ILo1Y3e+QX4zSR0zTE=
+github.com/onsi/ginkgo/v2 v2.27.5/go.mod h1:ArE1D/XhNXBXCBkKOLkbsb2c81dQHCRcF5zwn/ykDRo=
+github.com/onsi/gomega v1.39.0 h1:y2ROC3hKFmQZJNFeGAMeHZKkjBL65mIZcvrLQBF9k6Q=
+github.com/onsi/gomega v1.39.0/go.mod h1:ZCU1pkQcXDO5Sl9/VVEGlDyp+zm0m1cmeG5TOzLgdh4=
 github.com/opencontainers/go-digest v1.0.0 h1:apOUWs51W5PlhuyGyz9FCeeBIOUDA/6nW8Oi/yOhh5U=
 github.com/opencontainers/go-digest v1.0.0/go.mod h1:0JzlMkj0TRzQZfJkVvzbP0HBR3IKzErnv2BNG4W4MAM=
 github.com/opencontainers/image-spec v1.1.1 h1:y0fUlFfIZhPF1W537XOLg0/fcx6zcHCJwooC2xJA040=
@@ -978,8 +978,8 @@ golang.org/x/sys v0.8.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.10.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.11.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.16.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
-golang.org/x/sys v0.39.0 h1:CvCKL8MeisomCi6qNZ+wbb0DN9E5AATixKsvNtMoMFk=
-golang.org/x/sys v0.39.0/go.mod h1:OgkHotnGiDImocRcuBABYBEXf8A9a87e/uXjp9XT3ks=
+golang.org/x/sys v0.40.0 h1:DBZZqJ2Rkml6QMQsZywtnjnnGvHza6BTfYFWY9kjEWQ=
+golang.org/x/sys v0.40.0/go.mod h1:OgkHotnGiDImocRcuBABYBEXf8A9a87e/uXjp9XT3ks=
 golang.org/x/telemetry v0.0.0-20251111182119-bc8e575c7b54 h1:E2/AqCUMZGgd73TQkxUMcMla25GB9i/5HOdLr+uH7Vo=
 golang.org/x/telemetry v0.0.0-20251111182119-bc8e575c7b54/go.mod h1:hKdjCMrbv9skySur+Nek8Hd0uJ0GuxJIoIX2payrIdQ=
 golang.org/x/term v0.0.0-20201126162022-7de9c90e9dd1/go.mod h1:bj7SfCRtBDWHUb9snDiAeCFNEtKQo2Wmx5Cou7ajbmo=

pkg/functions/parse.go

@@ -1492,7 +1492,7 @@ func ParseFunctionCall(llmresult string, functionConfig FunctionsConfig) []FuncC
 	results := []FuncCallResults{}
 	llmResults := []string{}
 
-	returnResult := func(results []string) (result []FuncCallResults, e error) {
+	extractJSON := func(results []string) (result []FuncCallResults, e error) {
 		// As we have to change the result before processing, we can't stream the answer token-by-token (yet?)
 		result = make([]FuncCallResults, 0)
 
@@ -1593,7 +1593,7 @@ func ParseFunctionCall(llmresult string, functionConfig FunctionsConfig) []FuncC
 		if len(llmResults) == 0 {
 			llmResults = append(llmResults, llmresult)
 		}
-		results, _ = returnResult(llmResults)
+		results, _ = extractJSON(llmResults)
 	}
 
 	// Determine which XML format to use (if any)
@@ -1632,8 +1632,16 @@ func ParseFunctionCall(llmresult string, functionConfig FunctionsConfig) []FuncC
 		// But skip if JSONRegexMatch or ResponseRegex was used (they already extracted the content)
 		xmlResults, err := ParseXML(llmresult, xmlFormat)
 		if err == nil && len(xmlResults) > 0 {
-			xlog.Debug("Found additional XML tool calls alongside JSON", "xml_count", len(xmlResults))
-			results = append(results, xmlResults...)
+			// Check if JSON is inside XML tags, if so, skip it
+			for _, result := range xmlResults {
+				jsonResults, _ := extractJSON([]string{result.Name})
+				if len(jsonResults) > 0 {
+					xlog.Debug("Found valid JSON inside XML tags, skipping XML parsing", "json_count", len(jsonResults))
+				} else {
+					xlog.Debug("Found additional XML tool calls alongside JSON", "xml_count", len(xmlResults))
+					results = append(results, xmlResults...)
+				}
+			}
 		}
 	}
 

pkg/functions/parse_test.go

@@ -820,6 +820,23 @@ Final text`
 			Expect(results[0].Name).To(Equal("first"))
 			Expect(results[1].Name).To(Equal("second"))
 		})
+
+		It("should not duplicate parse JSON inside tool_call tags", func() {
+			// This test reproduces a bug where JSON inside <tool_call> tags
+			// gets parsed twice: once as JSON (correctly) and once as XML (incorrectly)
+			// The XML parser should not run when JSON parsing already found valid results
+			input := `<tool_call>
+{"name": "get_current_weather", "arguments": {"location": "Beijing", "unit": "celsius"}}
+</tool_call>`
+
+			results := ParseFunctionCall(input, functionConfig)
+			// Should only have 1 result, not 2 (one correct + one malformed)
+			Expect(results).To(HaveLen(1), "Should not create duplicate entries when JSON is inside XML tags")
+			Expect(results[0].Name).To(Equal("get_current_weather"))
+			Expect(results[0].Arguments).To(Equal(`{"location":"Beijing","unit":"celsius"}`))
+			// Verify the name is not the entire JSON object (which would indicate malformed XML parsing)
+			Expect(results[0].Name).NotTo(ContainSubstring(`{"name"`), "Function name should not contain JSON object")
+		})
 	})
 
 	Context("Iterative Parser (ChatMsgParser)", func() {

pkg/functions/reasoning.go

@@ -1,114 +0,0 @@
-package functions
-
-import (
-	"strings"
-)
-
-// ExtractReasoning extracts reasoning content from thinking tags and returns
-// both the extracted reasoning and the cleaned content (with tags removed).
-// It handles <thinking>...</thinking> and <think>...</think> tags.
-// Multiple reasoning blocks are concatenated with newlines.
-func ExtractReasoning(content string) (reasoning string, cleanedContent string) {
-	if content == "" {
-		return "", content
-	}
-
-	var reasoningParts []string
-	var cleanedParts []string
-	remaining := content
-
-	// Define tag pairs to look for
-	tagPairs := []struct {
-		start string
-		end   string
-	}{
-		{"<thinking>", "</thinking>"},
-		{"<think>", "</think>"},
-	}
-
-	// Track the last position we've processed
-	lastPos := 0
-
-	for {
-		// Find the earliest tag start
-		earliestStart := -1
-		earliestEnd := -1
-		isUnclosed := false
-		var matchedTag struct {
-			start string
-			end   string
-		}
-
-		for _, tagPair := range tagPairs {
-			startIdx := strings.Index(remaining[lastPos:], tagPair.start)
-			if startIdx == -1 {
-				continue
-			}
-			startIdx += lastPos
-
-			// Find the corresponding end tag
-			endIdx := strings.Index(remaining[startIdx+len(tagPair.start):], tagPair.end)
-			if endIdx == -1 {
-				// Unclosed tag - extract what we have
-				if earliestStart == -1 || startIdx < earliestStart {
-					earliestStart = startIdx
-					earliestEnd = len(remaining)
-					isUnclosed = true
-					matchedTag = tagPair
-				}
-				continue
-			}
-			endIdx += startIdx + len(tagPair.start)
-
-			// Found a complete tag pair
-			if earliestStart == -1 || startIdx < earliestStart {
-				earliestStart = startIdx
-				earliestEnd = endIdx + len(tagPair.end)
-				isUnclosed = false
-				matchedTag = tagPair
-			}
-		}
-
-		if earliestStart == -1 {
-			// No more tags found, add remaining content
-			if lastPos < len(remaining) {
-				cleanedParts = append(cleanedParts, remaining[lastPos:])
-			}
-			break
-		}
-
-		// Add content before the tag
-		if earliestStart > lastPos {
-			cleanedParts = append(cleanedParts, remaining[lastPos:earliestStart])
-		}
-
-		// Extract reasoning content
-		reasoningStart := earliestStart + len(matchedTag.start)
-		// For unclosed tags, earliestEnd is already at the end of the string
-		// For closed tags, earliestEnd points to after the closing tag, so we subtract the end tag length
-		var reasoningEnd int
-		if isUnclosed {
-			// Unclosed tag - extract everything to the end
-			reasoningEnd = len(remaining)
-		} else {
-			// Closed tag - exclude the end tag
-			reasoningEnd = earliestEnd - len(matchedTag.end)
-		}
-		if reasoningEnd > reasoningStart {
-			reasoningContent := strings.TrimSpace(remaining[reasoningStart:reasoningEnd])
-			if reasoningContent != "" {
-				reasoningParts = append(reasoningParts, reasoningContent)
-			}
-		}
-
-		// Move past this tag
-		lastPos = earliestEnd
-	}
-
-	// Combine reasoning parts
-	reasoning = strings.Join(reasoningParts, "\n\n")
-	// Combine cleaned content parts
-	cleanedContent = strings.Join(cleanedParts, "")
-
-	return reasoning, cleanedContent
-}

pkg/functions/reasoning_test.go

@@ -1,261 +0,0 @@
-package functions_test
-
-import (
-	"strings"
-
-	. "github.com/mudler/LocalAI/pkg/functions"
-	. "github.com/onsi/ginkgo/v2"
-	. "github.com/onsi/gomega"
-)
-
-var _ = Describe("ExtractReasoning", func() {
-	Context("when content has no reasoning tags", func() {
-		It("should return empty reasoning and original content", func() {
-			content := "This is regular content without any tags."
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(BeEmpty())
-			Expect(cleaned).To(Equal(content))
-		})
-
-		It("should handle empty string", func() {
-			content := ""
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(BeEmpty())
-			Expect(cleaned).To(BeEmpty())
-		})
-
-		It("should handle content with only whitespace", func() {
-			content := "   \n\t  "
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(BeEmpty())
-			Expect(cleaned).To(Equal(content))
-		})
-	})
-
-	Context("when content has <thinking> tags", func() {
-		It("should extract reasoning from single thinking block", func() {
-			content := "Some text <thinking>This is my reasoning</thinking> More text"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("This is my reasoning"))
-			Expect(cleaned).To(Equal("Some text  More text"))
-		})
-
-		It("should extract reasoning and preserve surrounding content", func() {
-			content := "Before <thinking>Reasoning here</thinking> After"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("Reasoning here"))
-			Expect(cleaned).To(Equal("Before  After"))
-		})
-
-		It("should handle thinking block at the start", func() {
-			content := "<thinking>Start reasoning</thinking> Regular content"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("Start reasoning"))
-			Expect(cleaned).To(Equal(" Regular content"))
-		})
-
-		It("should handle thinking block at the end", func() {
-			content := "Regular content <thinking>End reasoning</thinking>"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("End reasoning"))
-			Expect(cleaned).To(Equal("Regular content "))
-		})
-
-		It("should handle only thinking block", func() {
-			content := "<thinking>Only reasoning</thinking>"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("Only reasoning"))
-			Expect(cleaned).To(BeEmpty())
-		})
-
-		It("should trim whitespace from reasoning content", func() {
-			content := "Text <thinking>  \n  Reasoning with spaces  \n  </thinking> More"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("Reasoning with spaces"))
-			Expect(cleaned).To(Equal("Text  More"))
-		})
-	})
-
-	Context("when content has <think> tags", func() {
-		It("should extract reasoning from redacted_reasoning block", func() {
-			content := "Text <think>Redacted reasoning</think> More"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("Redacted reasoning"))
-			Expect(cleaned).To(Equal("Text  More"))
-		})
-
-		It("should handle redacted_reasoning with multiline content", func() {
-			content := "Before <think>Line 1\nLine 2\nLine 3</think> After"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("Line 1\nLine 2\nLine 3"))
-			Expect(cleaned).To(Equal("Before  After"))
-		})
-
-		It("should handle redacted_reasoning with complex content", func() {
-			content := "Start <think>Complex reasoning\nwith\nmultiple\nlines</think> End"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("Complex reasoning\nwith\nmultiple\nlines"))
-			Expect(cleaned).To(Equal("Start  End"))
-		})
-	})
-
-	Context("when content has multiple reasoning blocks", func() {
-		It("should concatenate multiple thinking blocks with newlines", func() {
-			content := "Text <thinking>First</thinking> Middle <thinking>Second</thinking> End"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("First\n\nSecond"))
-			Expect(cleaned).To(Equal("Text  Middle  End"))
-		})
-
-		It("should handle multiple different tag types", func() {
-			content := "A <thinking>One</thinking> B <think>Two</think> C <think>Three</think> D"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(ContainSubstring("One"))
-			Expect(reasoning).To(ContainSubstring("Two"))
-			Expect(reasoning).To(ContainSubstring("Three"))
-			Expect(cleaned).To(Equal("A  B  C  D"))
-		})
-
-		It("should handle nested tags correctly (extracts first match)", func() {
-			content := "Text <thinking>Outer <think>Inner</think></thinking> More"
-			reasoning, cleaned := ExtractReasoning(content)
-			// Should extract the outer thinking block
-			Expect(reasoning).To(ContainSubstring("Outer"))
-			Expect(reasoning).To(ContainSubstring("Inner"))
-			Expect(cleaned).To(Equal("Text  More"))
-		})
-	})
-
-	Context("when content has unclosed reasoning tags", func() {
-		It("should extract unclosed thinking block", func() {
-			content := "Text <thinking>Unclosed reasoning"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("Unclosed reasoning"))
-			Expect(cleaned).To(Equal("Text "))
-		})
-
-		It("should extract unclosed think block", func() {
-			content := "Before <think>Incomplete"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("Incomplete"))
-			Expect(cleaned).To(Equal("Before "))
-		})
-
-		It("should extract unclosed redacted_reasoning block", func() {
-			content := "Start <think>Partial reasoning content"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("Partial reasoning content"))
-			Expect(cleaned).To(Equal("Start "))
-		})
-
-		It("should handle unclosed tag at the end", func() {
-			content := "Regular content <thinking>Unclosed at end"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("Unclosed at end"))
-			Expect(cleaned).To(Equal("Regular content "))
-		})
-	})
-
-	Context("when content has empty reasoning blocks", func() {
-		It("should ignore empty thinking block", func() {
-			content := "Text <thinking></thinking> More"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(BeEmpty())
-			Expect(cleaned).To(Equal("Text  More"))
-		})
-
-		It("should ignore thinking block with only whitespace", func() {
-			content := "Text <thinking>   \n\t  </thinking> More"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(BeEmpty())
-			Expect(cleaned).To(Equal("Text  More"))
-		})
-	})
-
-	Context("when content has reasoning tags with special characters", func() {
-		It("should handle reasoning with newlines", func() {
-			content := "Before <thinking>Line 1\nLine 2\nLine 3</thinking> After"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("Line 1\nLine 2\nLine 3"))
-			Expect(cleaned).To(Equal("Before  After"))
-		})
-
-		It("should handle reasoning with code blocks", func() {
-			content := "Text <thinking>Reasoning with ```code``` blocks</thinking> More"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("Reasoning with ```code``` blocks"))
-			Expect(cleaned).To(Equal("Text  More"))
-		})
-
-		It("should handle reasoning with JSON", func() {
-			content := "Before <think>{\"key\": \"value\"}</think> After"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("{\"key\": \"value\"}"))
-			Expect(cleaned).To(Equal("Before  After"))
-		})
-
-		It("should handle reasoning with HTML-like content", func() {
-			content := "Text <thinking>Reasoning with <tags> inside</thinking> More"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("Reasoning with <tags> inside"))
-			Expect(cleaned).To(Equal("Text  More"))
-		})
-	})
-
-	Context("when content has reasoning mixed with regular content", func() {
-		It("should preserve content order correctly", func() {
-			content := "Start <thinking>Reasoning</thinking> Middle <think>More reasoning</think> End"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(ContainSubstring("Reasoning"))
-			Expect(reasoning).To(ContainSubstring("More reasoning"))
-			Expect(cleaned).To(Equal("Start  Middle  End"))
-		})
-
-		It("should handle reasoning in the middle of a sentence", func() {
-			content := "This is a <thinking>reasoning</thinking> sentence."
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("reasoning"))
-			Expect(cleaned).To(Equal("This is a  sentence."))
-		})
-	})
-
-	Context("edge cases", func() {
-		It("should handle content with only opening tag", func() {
-			content := "<thinking>"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(BeEmpty())
-			Expect(cleaned).To(Equal(""))
-		})
-
-		It("should handle content with only closing tag", func() {
-			content := "</thinking>"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(BeEmpty())
-			Expect(cleaned).To(Equal("</thinking>"))
-		})
-
-		It("should handle mismatched tags", func() {
-			content := "<thinking>Content</think>"
-			reasoning, cleaned := ExtractReasoning(content)
-			// Should extract unclosed thinking block
-			Expect(reasoning).To(ContainSubstring("Content"))
-			Expect(cleaned).To(Equal(""))
-		})
-
-		It("should handle very long reasoning content", func() {
-			longReasoning := strings.Repeat("This is reasoning content. ", 100)
-			content := "Text <thinking>" + longReasoning + "</thinking> More"
-			reasoning, cleaned := ExtractReasoning(content)
-			// TrimSpace is applied, so we need to account for that
-			Expect(reasoning).To(Equal(strings.TrimSpace(longReasoning)))
-			Expect(cleaned).To(Equal("Text  More"))
-		})
-
-		It("should handle reasoning with unicode characters", func() {
-			content := "Text <thinking>Reasoning with 中文 and emoji 🧠</thinking> More"
-			reasoning, cleaned := ExtractReasoning(content)
-			Expect(reasoning).To(Equal("Reasoning with 中文 and emoji 🧠"))
-			Expect(cleaned).To(Equal("Text  More"))
-		})
-	})
-})

pkg/reasoning/config.go

@@ -0,0 +1,8 @@
+package reasoning
+
+type ReasoningConfig struct {
+	// ThinkingForcedOpen indicates that the model outputs reasoning without an opening tag.
+	// When true, all content from the start is treated as reasoning until a closing tag is found.
+	// This is useful for models like GLM-4 that output reasoning without <think> but end with </think>.
+	ThinkingForcedOpen bool `yaml:"thinking_forced_open,omitempty" json:"thinking_forced_open,omitempty"`
+}

pkg/reasoning/options.go

@@ -0,0 +1,18 @@
+package reasoning
+
+// options holds the configuration for reasoning extraction
+type options struct {
+	thinkingForcedOpen bool
+}
+
+// Option is a functional option for configuring reasoning extraction
+type Option func(*options)
+
+// WithThinkingForcedOpen configures the extractor to treat all content from the start
+// as reasoning until a closing tag is found. This is useful for models like GLM-4
+// that output reasoning without <think> but end with </think>.
+func WithThinkingForcedOpen() Option {
+	return func(o *options) {
+		o.thinkingForcedOpen = true
+	}
+}

pkg/reasoning/reasoning.go

@@ -0,0 +1,256 @@
+package reasoning
+
+import (
+	"strings"
+)
+
+// Common thinking/reasoning opening tags used by various models.
+// These match the tags detected by llama.cpp in common/chat.cpp
+var thinkingOpenTags = []string{
+	// DeepSeek R1, V3.1, Nemotron V2, MiniMax M2, Hermes 2 Pro, Granite, Exaone MOE
+	"<think>\n",
+	"<think>",
+	// Generic thinking tags
+	"<thinking>\n",
+	"<thinking>",
+	// Apertus
+	"<|inner_prefix|>",
+	// Command R7B
+	"<|START_THINKING|>",
+	// Seed
+	"<seed:think>",
+	// Magistral (not in llama.cpp but common)
+	"[THINK]\n",
+	"[THINK]",
+}
+
+// DetectThinkingForcedOpen checks if a prompt ends with a thinking opening tag.
+// This is used to automatically detect when the model template has already added
+// the opening thinking tag, meaning the model will output reasoning content directly.
+// Returns true if the prompt ends with a known thinking opening tag.
+func DetectThinkingForcedOpen(prompt string) bool {
+	for _, tag := range thinkingOpenTags {
+		if strings.HasSuffix(prompt, tag) {
+			return true
+		}
+	}
+	return false
+}
+
+// Extract extracts reasoning content from thinking tags and returns
+// both the extracted reasoning and the cleaned content (with tags removed).
+// It handles <thinking>...</thinking> and <think>...</think> tags.
+// Multiple reasoning blocks are concatenated with newlines.
+// It also handles the case where only a closing tag is present (no opening tag),
+// in which case everything before the closing tag is treated as reasoning.
+//
+// Use WithThinkingForcedOpen() option when all content from the start should be
+// treated as reasoning until a closing tag is found.
+func Extract(content string, opts ...Option) (reasoning string, cleanedContent string) {
+	if content == "" {
+		return "", content
+	}
+
+	cfg := &options{}
+	for _, opt := range opts {
+		opt(cfg)
+	}
+
+	if cfg.thinkingForcedOpen {
+		return extractForcedOpen(content)
+	}
+
+	return extractFromTags(content)
+}
+
+// extractForcedOpen handles the case where reasoning starts without an opening tag.
+// All content from the start is treated as reasoning until a closing tag is found.
+func extractForcedOpen(content string) (reasoning string, cleanedContent string) {
+	// Look for the earliest closing tag
+	// These match the closing tags used by llama.cpp for various models
+	closingTags := []string{
+		"</thinking>",
+		"</think>",
+		"<|END_THINKING|>", // Command R7B
+		"<|inner_suffix|>", // Apertus
+		"</seed:think>",    // Seed
+		"[/THINK]",         // Magistral
+	}
+
+	earliestCloseIdx := -1
+	var matchedCloseTag string
+
+	for _, closeTag := range closingTags {
+		idx := strings.Index(content, closeTag)
+		if idx != -1 && (earliestCloseIdx == -1 || idx < earliestCloseIdx) {
+			earliestCloseIdx = idx
+			matchedCloseTag = closeTag
+		}
+	}
+
+	if earliestCloseIdx == -1 {
+		// No closing tag found - all content is reasoning (still streaming)
+		return strings.TrimSpace(content), ""
+	}
+
+	// Found closing tag - everything before is reasoning, everything after is content
+	reasoning = strings.TrimSpace(content[:earliestCloseIdx])
+	cleanedContent = content[earliestCloseIdx+len(matchedCloseTag):]
+
+	// Continue processing the rest for any additional reasoning blocks
+	if cleanedContent != "" {
+		additionalReasoning, finalContent := extractFromTags(cleanedContent)
+		if additionalReasoning != "" {
+			if reasoning != "" {
+				reasoning = reasoning + "\n\n" + additionalReasoning
+			} else {
+				reasoning = additionalReasoning
+			}
+		}
+		cleanedContent = finalContent
+	}
+
+	return reasoning, cleanedContent
+}
+
+// extractFromTags extracts reasoning content from thinking tags.
+// This is the core implementation that handles standard tag-based extraction.
+func extractFromTags(content string) (reasoning string, cleanedContent string) {
+	if content == "" {
+		return "", content
+	}
+
+	var reasoningParts []string
+	var cleanedParts []string
+	remaining := content
+
+	// Define tag pairs to look for
+	// These match the tags used by llama.cpp for various models
+	tagPairs := []struct {
+		start string
+		end   string
+	}{
+		{"<thinking>", "</thinking>"},
+		{"<think>", "</think>"},
+		{"<|START_THINKING|>", "<|END_THINKING|>"}, // Command R7B
+		{"<|inner_prefix|>", "<|inner_suffix|>"},   // Apertus
+		{"<seed:think>", "</seed:think>"},          // Seed
+		{"[THINK]", "[/THINK]"},                    // Magistral
+	}
+
+	// Track the last position we've processed
+	lastPos := 0
+
+	for {
+		// Find the earliest tag start
+		earliestStart := -1
+		earliestEnd := -1
+		isUnclosed := false
+		isClosingOnly := false
+		var matchedTag struct {
+			start string
+			end   string
+		}
+
+		for _, tagPair := range tagPairs {
+			startIdx := strings.Index(remaining[lastPos:], tagPair.start)
+			endIdx := strings.Index(remaining[lastPos:], tagPair.end)
+
+			// Check for closing-only tag (closing tag appears before or without opening tag)
+			if endIdx != -1 && (startIdx == -1 || endIdx < startIdx) {
+				// Found a closing tag without a preceding opening tag
+				closingTagPos := endIdx + lastPos
+				if earliestStart == -1 || closingTagPos < earliestStart || (isClosingOnly && closingTagPos < earliestEnd) {
+					earliestStart = lastPos
+					earliestEnd = closingTagPos + len(tagPair.end)
+					isClosingOnly = true
+					isUnclosed = false
+					matchedTag = tagPair
+				}
+				continue
+			}
+
+			if startIdx == -1 {
+				continue
+			}
+			startIdx += lastPos
+
+			// Find the corresponding end tag after the start tag
+			endIdxAfterStart := strings.Index(remaining[startIdx+len(tagPair.start):], tagPair.end)
+			if endIdxAfterStart == -1 {
+				// Unclosed tag - extract what we have
+				if earliestStart == -1 || startIdx < earliestStart {
+					earliestStart = startIdx
+					earliestEnd = len(remaining)
+					isUnclosed = true
+					isClosingOnly = false
+					matchedTag = tagPair
+				}
+				continue
+			}
+			endIdxAfterStart += startIdx + len(tagPair.start)
+
+			// Found a complete tag pair
+			if earliestStart == -1 || startIdx < earliestStart {
+				earliestStart = startIdx
+				earliestEnd = endIdxAfterStart + len(tagPair.end)
+				isUnclosed = false
+				isClosingOnly = false
+				matchedTag = tagPair
+			}
+		}
+
+		if earliestStart == -1 {
+			// No more tags found, add remaining content
+			if lastPos < len(remaining) {
+				cleanedParts = append(cleanedParts, remaining[lastPos:])
+			}
+			break
+		}
+
+		if isClosingOnly {
+			// Closing tag without opening tag - content before closing tag is reasoning
+			reasoningContent := strings.TrimSpace(remaining[lastPos : earliestEnd-len(matchedTag.end)])
+			if reasoningContent != "" {
+				reasoningParts = append(reasoningParts, reasoningContent)
+			}
+			// Move past the closing tag
+			lastPos = earliestEnd
+			continue
+		}
+
+		// Add content before the tag
+		if earliestStart > lastPos {
+			cleanedParts = append(cleanedParts, remaining[lastPos:earliestStart])
+		}
+
+		// Extract reasoning content
+		reasoningStart := earliestStart + len(matchedTag.start)
+		// For unclosed tags, earliestEnd is already at the end of the string
+		// For closed tags, earliestEnd points to after the closing tag, so we subtract the end tag length
+		var reasoningEnd int
+		if isUnclosed {
+			// Unclosed tag - extract everything to the end
+			reasoningEnd = len(remaining)
+		} else {
+			// Closed tag - exclude the end tag
+			reasoningEnd = earliestEnd - len(matchedTag.end)
+		}
+		if reasoningEnd > reasoningStart {
+			reasoningContent := strings.TrimSpace(remaining[reasoningStart:reasoningEnd])
+			if reasoningContent != "" {
+				reasoningParts = append(reasoningParts, reasoningContent)
+			}
+		}
+
+		// Move past this tag
+		lastPos = earliestEnd
+	}
+
+	// Combine reasoning parts
+	reasoning = strings.Join(reasoningParts, "\n\n")
+	// Combine cleaned content parts
+	cleanedContent = strings.Join(cleanedParts, "")
+
+	return reasoning, cleanedContent
+}

pkg/reasoning/reasoning_suite_test.go

@@ -0,0 +1,13 @@
+package reasoning_test
+
+import (
+	"testing"
+
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+)
+
+func TestReasoning(t *testing.T) {
+	RegisterFailHandler(Fail)
+	RunSpecs(t, "Reasoning Suite")
+}

pkg/reasoning/reasoning_test.go

@@ -0,0 +1,499 @@
+package reasoning_test
+
+import (
+	"strings"
+
+	. "github.com/mudler/LocalAI/pkg/reasoning"
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+)
+
+var _ = Describe("DetectThinkingForcedOpen", func() {
+	It("should detect <think> at end of prompt", func() {
+		Expect(DetectThinkingForcedOpen("Some prompt<think>")).To(BeTrue())
+		Expect(DetectThinkingForcedOpen("Some prompt<think>\n")).To(BeTrue())
+	})
+
+	It("should detect <thinking> at end of prompt", func() {
+		Expect(DetectThinkingForcedOpen("Some prompt<thinking>")).To(BeTrue())
+		Expect(DetectThinkingForcedOpen("Some prompt<thinking>\n")).To(BeTrue())
+	})
+
+	It("should detect model-specific tags", func() {
+		Expect(DetectThinkingForcedOpen("Some prompt<|inner_prefix|>")).To(BeTrue())
+		Expect(DetectThinkingForcedOpen("Some prompt<|START_THINKING|>")).To(BeTrue())
+		Expect(DetectThinkingForcedOpen("Some prompt<seed:think>")).To(BeTrue())
+		Expect(DetectThinkingForcedOpen("Some prompt[THINK]")).To(BeTrue())
+		Expect(DetectThinkingForcedOpen("Some prompt[THINK]\n")).To(BeTrue())
+	})
+
+	It("should not detect if tag is in the middle", func() {
+		Expect(DetectThinkingForcedOpen("Some <think> prompt")).To(BeFalse())
+		Expect(DetectThinkingForcedOpen("<think>reasoning</think>")).To(BeFalse())
+	})
+
+	It("should not detect if no thinking tag", func() {
+		Expect(DetectThinkingForcedOpen("Some regular prompt")).To(BeFalse())
+		Expect(DetectThinkingForcedOpen("")).To(BeFalse())
+	})
+})
+
+var _ = Describe("Extract", func() {
+	Context("when content has no reasoning tags", func() {
+		It("should return empty reasoning and original content", func() {
+			content := "This is regular content without any tags."
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(BeEmpty())
+			Expect(cleaned).To(Equal(content))
+		})
+
+		It("should handle empty string", func() {
+			content := ""
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(BeEmpty())
+			Expect(cleaned).To(BeEmpty())
+		})
+
+		It("should handle content with only whitespace", func() {
+			content := "   \n\t  "
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(BeEmpty())
+			Expect(cleaned).To(Equal(content))
+		})
+	})
+
+	Context("when content has <thinking> tags", func() {
+		It("should extract reasoning from single thinking block", func() {
+			content := "Some text <thinking>This is my reasoning</thinking> More text"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("This is my reasoning"))
+			Expect(cleaned).To(Equal("Some text  More text"))
+		})
+
+		It("should extract reasoning and preserve surrounding content", func() {
+			content := "Before <thinking>Reasoning here</thinking> After"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Reasoning here"))
+			Expect(cleaned).To(Equal("Before  After"))
+		})
+
+		It("should handle thinking block at the start", func() {
+			content := "<thinking>Start reasoning</thinking> Regular content"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Start reasoning"))
+			Expect(cleaned).To(Equal(" Regular content"))
+		})
+
+		It("should handle thinking block at the end", func() {
+			content := "Regular content <thinking>End reasoning</thinking>"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("End reasoning"))
+			Expect(cleaned).To(Equal("Regular content "))
+		})
+
+		It("should handle only thinking block", func() {
+			content := "<thinking>Only reasoning</thinking>"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Only reasoning"))
+			Expect(cleaned).To(BeEmpty())
+		})
+
+		It("should trim whitespace from reasoning content", func() {
+			content := "Text <thinking>  \n  Reasoning with spaces  \n  </thinking> More"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Reasoning with spaces"))
+			Expect(cleaned).To(Equal("Text  More"))
+		})
+	})
+
+	Context("when content has <think> tags", func() {
+		It("should extract reasoning from redacted_reasoning block", func() {
+			content := "Text <think>Redacted reasoning</think> More"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Redacted reasoning"))
+			Expect(cleaned).To(Equal("Text  More"))
+		})
+
+		It("should handle redacted_reasoning with multiline content", func() {
+			content := "Before <think>Line 1\nLine 2\nLine 3</think> After"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Line 1\nLine 2\nLine 3"))
+			Expect(cleaned).To(Equal("Before  After"))
+		})
+
+		It("should handle redacted_reasoning with complex content", func() {
+			content := "Start <think>Complex reasoning\nwith\nmultiple\nlines</think> End"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Complex reasoning\nwith\nmultiple\nlines"))
+			Expect(cleaned).To(Equal("Start  End"))
+		})
+	})
+
+	Context("when content has multiple reasoning blocks", func() {
+		It("should concatenate multiple thinking blocks with newlines", func() {
+			content := "Text <thinking>First</thinking> Middle <thinking>Second</thinking> End"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("First\n\nSecond"))
+			Expect(cleaned).To(Equal("Text  Middle  End"))
+		})
+
+		It("should handle multiple different tag types", func() {
+			content := "A <thinking>One</thinking> B <think>Two</think> C <think>Three</think> D"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(ContainSubstring("One"))
+			Expect(reasoning).To(ContainSubstring("Two"))
+			Expect(reasoning).To(ContainSubstring("Three"))
+			Expect(cleaned).To(Equal("A  B  C  D"))
+		})
+
+		It("should handle nested tags correctly (extracts first match)", func() {
+			content := "Text <thinking>Outer <think>Inner</think></thinking> More"
+			reasoning, cleaned := Extract(content)
+			// Should extract the outer thinking block
+			Expect(reasoning).To(ContainSubstring("Outer"))
+			Expect(reasoning).To(ContainSubstring("Inner"))
+			Expect(cleaned).To(Equal("Text  More"))
+		})
+	})
+
+	Context("when content has unclosed reasoning tags", func() {
+		It("should extract unclosed thinking block", func() {
+			content := "Text <thinking>Unclosed reasoning"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Unclosed reasoning"))
+			Expect(cleaned).To(Equal("Text "))
+		})
+
+		It("should extract unclosed think block", func() {
+			content := "Before <think>Incomplete"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Incomplete"))
+			Expect(cleaned).To(Equal("Before "))
+		})
+
+		It("should extract unclosed redacted_reasoning block", func() {
+			content := "Start <think>Partial reasoning content"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Partial reasoning content"))
+			Expect(cleaned).To(Equal("Start "))
+		})
+
+		It("should handle unclosed tag at the end", func() {
+			content := "Regular content <thinking>Unclosed at end"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Unclosed at end"))
+			Expect(cleaned).To(Equal("Regular content "))
+		})
+	})
+
+	Context("when content has empty reasoning blocks", func() {
+		It("should ignore empty thinking block", func() {
+			content := "Text <thinking></thinking> More"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(BeEmpty())
+			Expect(cleaned).To(Equal("Text  More"))
+		})
+
+		It("should ignore thinking block with only whitespace", func() {
+			content := "Text <thinking>   \n\t  </thinking> More"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(BeEmpty())
+			Expect(cleaned).To(Equal("Text  More"))
+		})
+	})
+
+	Context("when content has reasoning tags with special characters", func() {
+		It("should handle reasoning with newlines", func() {
+			content := "Before <thinking>Line 1\nLine 2\nLine 3</thinking> After"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Line 1\nLine 2\nLine 3"))
+			Expect(cleaned).To(Equal("Before  After"))
+		})
+
+		It("should handle reasoning with code blocks", func() {
+			content := "Text <thinking>Reasoning with ```code``` blocks</thinking> More"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Reasoning with ```code``` blocks"))
+			Expect(cleaned).To(Equal("Text  More"))
+		})
+
+		It("should handle reasoning with JSON", func() {
+			content := "Before <think>{\"key\": \"value\"}</think> After"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("{\"key\": \"value\"}"))
+			Expect(cleaned).To(Equal("Before  After"))
+		})
+
+		It("should handle reasoning with HTML-like content", func() {
+			content := "Text <thinking>Reasoning with <tags> inside</thinking> More"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Reasoning with <tags> inside"))
+			Expect(cleaned).To(Equal("Text  More"))
+		})
+	})
+
+	Context("when content has reasoning mixed with regular content", func() {
+		It("should preserve content order correctly", func() {
+			content := "Start <thinking>Reasoning</thinking> Middle <think>More reasoning</think> End"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(ContainSubstring("Reasoning"))
+			Expect(reasoning).To(ContainSubstring("More reasoning"))
+			Expect(cleaned).To(Equal("Start  Middle  End"))
+		})
+
+		It("should handle reasoning in the middle of a sentence", func() {
+			content := "This is a <thinking>reasoning</thinking> sentence."
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("reasoning"))
+			Expect(cleaned).To(Equal("This is a  sentence."))
+		})
+	})
+
+	Context("edge cases without WithThinkingForcedOpen", func() {
+		It("should handle content with only opening tag", func() {
+			content := "<thinking>"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(BeEmpty())
+			Expect(cleaned).To(Equal(""))
+		})
+
+		It("should handle content with only closing tag (no content before)", func() {
+			content := "</thinking>"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(BeEmpty())
+			Expect(cleaned).To(BeEmpty())
+		})
+
+		It("should extract reasoning when only closing tag is present", func() {
+			// GLM-4 style: reasoning content followed by closing tag without opening tag
+			content := "This is reasoning content</think>this is the actual response"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("This is reasoning content"))
+			Expect(cleaned).To(Equal("this is the actual response"))
+		})
+
+		It("should handle closing-only tag with multiline reasoning", func() {
+			content := "1. First point\n2. Second point\n3. Third point</think>Final answer"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("1. First point\n2. Second point\n3. Third point"))
+			Expect(cleaned).To(Equal("Final answer"))
+		})
+
+		It("should handle closing-only tag with complex reasoning (GLM-4 example)", func() {
+			content := "**Analyze the user's input:** The user says something.\n\n**Final Decision:** Output the text.</think>this is a test"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("**Analyze the user's input:** The user says something.\n\n**Final Decision:** Output the text."))
+			Expect(cleaned).To(Equal("this is a test"))
+		})
+
+		It("should handle closing-only thinking tag", func() {
+			content := "Some reasoning here</thinking>actual content"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Some reasoning here"))
+			Expect(cleaned).To(Equal("actual content"))
+		})
+
+		It("should handle mismatched tags", func() {
+			content := "<thinking>Content</think>"
+			reasoning, cleaned := Extract(content)
+			// Should extract unclosed thinking block
+			Expect(reasoning).To(ContainSubstring("Content"))
+			Expect(cleaned).To(Equal(""))
+		})
+
+		It("should handle very long reasoning content", func() {
+			longReasoning := strings.Repeat("This is reasoning content. ", 100)
+			content := "Text <thinking>" + longReasoning + "</thinking> More"
+			reasoning, cleaned := Extract(content)
+			// TrimSpace is applied, so we need to account for that
+			Expect(reasoning).To(Equal(strings.TrimSpace(longReasoning)))
+			Expect(cleaned).To(Equal("Text  More"))
+		})
+
+		It("should handle reasoning with unicode characters", func() {
+			content := "Text <thinking>Reasoning with 中文 and emoji 🧠</thinking> More"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Reasoning with 中文 and emoji 🧠"))
+			Expect(cleaned).To(Equal("Text  More"))
+		})
+	})
+
+	Context("with WithThinkingForcedOpen option", func() {
+		It("should treat all content as reasoning until closing tag", func() {
+			content := "This is reasoning</think>this is content"
+			reasoning, cleaned := Extract(content, WithThinkingForcedOpen())
+			Expect(reasoning).To(Equal("This is reasoning"))
+			Expect(cleaned).To(Equal("this is content"))
+		})
+
+		It("should treat all content as reasoning when no closing tag (streaming)", func() {
+			content := "This is reasoning content still streaming"
+			reasoning, cleaned := Extract(content, WithThinkingForcedOpen())
+			Expect(reasoning).To(Equal("This is reasoning content still streaming"))
+			Expect(cleaned).To(BeEmpty())
+		})
+
+		It("should handle GLM-4 style output", func() {
+			content := "**Analyze:** The user says something.\n\n**Final Decision:** Output the text.</think>this is a test"
+			reasoning, cleaned := Extract(content, WithThinkingForcedOpen())
+			Expect(reasoning).To(Equal("**Analyze:** The user says something.\n\n**Final Decision:** Output the text."))
+			Expect(cleaned).To(Equal("this is a test"))
+		})
+
+		It("should handle multiline reasoning with closing tag", func() {
+			content := "1. First point\n2. Second point\n3. Third point</think>Final answer"
+			reasoning, cleaned := Extract(content, WithThinkingForcedOpen())
+			Expect(reasoning).To(Equal("1. First point\n2. Second point\n3. Third point"))
+			Expect(cleaned).To(Equal("Final answer"))
+		})
+
+		It("should handle </thinking> closing tag", func() {
+			content := "Some reasoning here</thinking>actual content"
+			reasoning, cleaned := Extract(content, WithThinkingForcedOpen())
+			Expect(reasoning).To(Equal("Some reasoning here"))
+			Expect(cleaned).To(Equal("actual content"))
+		})
+
+		It("should handle additional reasoning blocks after initial forced open", func() {
+			content := "Initial reasoning</think>content<think>more reasoning</think>final content"
+			reasoning, cleaned := Extract(content, WithThinkingForcedOpen())
+			Expect(reasoning).To(Equal("Initial reasoning\n\nmore reasoning"))
+			Expect(cleaned).To(Equal("contentfinal content"))
+		})
+
+		It("should handle empty content", func() {
+			reasoning, cleaned := Extract("", WithThinkingForcedOpen())
+			Expect(reasoning).To(BeEmpty())
+			Expect(cleaned).To(BeEmpty())
+		})
+
+		It("should handle only closing tag", func() {
+			content := "</think>only content"
+			reasoning, cleaned := Extract(content, WithThinkingForcedOpen())
+			Expect(reasoning).To(BeEmpty())
+			Expect(cleaned).To(Equal("only content"))
+		})
+
+		It("should find earliest closing tag", func() {
+			// </think> comes before </thinking>
+			content := "Reasoning</think>content</thinking>more"
+			reasoning, cleaned := Extract(content, WithThinkingForcedOpen())
+			Expect(reasoning).To(Equal("Reasoning"))
+			Expect(cleaned).To(Equal("content</thinking>more"))
+		})
+
+		It("should handle Command R7B closing tag", func() {
+			content := "Reasoning content<|END_THINKING|>actual response"
+			reasoning, cleaned := Extract(content, WithThinkingForcedOpen())
+			Expect(reasoning).To(Equal("Reasoning content"))
+			Expect(cleaned).To(Equal("actual response"))
+		})
+
+		It("should handle Apertus closing tag", func() {
+			content := "Reasoning content<|inner_suffix|>actual response"
+			reasoning, cleaned := Extract(content, WithThinkingForcedOpen())
+			Expect(reasoning).To(Equal("Reasoning content"))
+			Expect(cleaned).To(Equal("actual response"))
+		})
+
+		It("should handle Seed closing tag", func() {
+			content := "Reasoning content</seed:think>actual response"
+			reasoning, cleaned := Extract(content, WithThinkingForcedOpen())
+			Expect(reasoning).To(Equal("Reasoning content"))
+			Expect(cleaned).To(Equal("actual response"))
+		})
+
+		It("should handle Magistral closing tag", func() {
+			content := "Reasoning content[/THINK]actual response"
+			reasoning, cleaned := Extract(content, WithThinkingForcedOpen())
+			Expect(reasoning).To(Equal("Reasoning content"))
+			Expect(cleaned).To(Equal("actual response"))
+		})
+	})
+
+	Context("with model-specific tag pairs", func() {
+		It("should extract Command R7B reasoning tags", func() {
+			content := "Before <|START_THINKING|>reasoning here<|END_THINKING|> After"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("reasoning here"))
+			Expect(cleaned).To(Equal("Before  After"))
+		})
+
+		It("should extract Apertus reasoning tags", func() {
+			content := "Before <|inner_prefix|>reasoning here<|inner_suffix|> After"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("reasoning here"))
+			Expect(cleaned).To(Equal("Before  After"))
+		})
+
+		It("should extract Seed reasoning tags", func() {
+			content := "Before <seed:think>reasoning here</seed:think> After"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("reasoning here"))
+			Expect(cleaned).To(Equal("Before  After"))
+		})
+
+		It("should extract Magistral reasoning tags", func() {
+			content := "Before [THINK]reasoning here[/THINK] After"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("reasoning here"))
+			Expect(cleaned).To(Equal("Before  After"))
+		})
+
+		It("should handle unclosed Command R7B tag", func() {
+			content := "Before <|START_THINKING|>reasoning still streaming"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("reasoning still streaming"))
+			Expect(cleaned).To(Equal("Before "))
+		})
+
+		It("should handle unclosed Apertus tag", func() {
+			content := "Before <|inner_prefix|>reasoning still streaming"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("reasoning still streaming"))
+			Expect(cleaned).To(Equal("Before "))
+		})
+
+		It("should handle unclosed Seed tag", func() {
+			content := "Before <seed:think>reasoning still streaming"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("reasoning still streaming"))
+			Expect(cleaned).To(Equal("Before "))
+		})
+
+		It("should handle unclosed Magistral tag", func() {
+			content := "Before [THINK]reasoning still streaming"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("reasoning still streaming"))
+			Expect(cleaned).To(Equal("Before "))
+		})
+
+		It("should handle closing-only Command R7B tag", func() {
+			content := "Reasoning content<|END_THINKING|>actual response"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Reasoning content"))
+			Expect(cleaned).To(Equal("actual response"))
+		})
+
+		It("should handle closing-only Apertus tag", func() {
+			content := "Reasoning content<|inner_suffix|>actual response"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Reasoning content"))
+			Expect(cleaned).To(Equal("actual response"))
+		})
+
+		It("should handle closing-only Seed tag", func() {
+			content := "Reasoning content</seed:think>actual response"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Reasoning content"))
+			Expect(cleaned).To(Equal("actual response"))
+		})
+
+		It("should handle closing-only Magistral tag", func() {
+			content := "Reasoning content[/THINK]actual response"
+			reasoning, cleaned := Extract(content)
+			Expect(reasoning).To(Equal("Reasoning content"))
+			Expect(cleaned).To(Equal("actual response"))
+		})
+	})
+})

scripts/build/package-gpu-libs.sh

@@ -148,10 +148,10 @@ package_cuda_libs() {
     done
 
     # Copy CUDA target directory for runtime compilation support
-    if [ -d "/usr/local/cuda/targets" ]; then
-        mkdir -p "$TARGET_LIB_DIR/../cuda"
-        cp -arfL /usr/local/cuda/targets "$TARGET_LIB_DIR/../cuda/" 2>/dev/null || true
-    fi
+    # if [ -d "/usr/local/cuda/targets" ]; then
+    #     mkdir -p "$TARGET_LIB_DIR/../cuda"
+    #     cp -arfL /usr/local/cuda/targets "$TARGET_LIB_DIR/../cuda/" 2>/dev/null || true
+    # fi
 
     echo "CUDA libraries packaged successfully"
 }

swagger/docs.go

@@ -1259,6 +1259,116 @@ const docTemplate = `{
                 }
             }
         },
+        "/v1/responses": {
+            "post": {
+                "summary": "Create a response using the Open Responses API",
+                "parameters": [
+                    {
+                        "description": "Request body",
+                        "name": "request",
+                        "in": "body",
+                        "required": true,
+                        "schema": {
+                            "$ref": "#/definitions/schema.OpenResponsesRequest"
+                        }
+                    }
+                ],
+                "responses": {
+                    "200": {
+                        "description": "Response",
+                        "schema": {
+                            "$ref": "#/definitions/schema.ORResponseResource"
+                        }
+                    }
+                }
+            }
+        },
+        "/v1/responses/{id}": {
+            "get": {
+                "description": "Retrieve a response by ID. Can be used for polling background responses or resuming streaming responses.",
+                "summary": "Get a response by ID",
+                "parameters": [
+                    {
+                        "type": "string",
+                        "description": "Response ID",
+                        "name": "id",
+                        "in": "path",
+                        "required": true
+                    },
+                    {
+                        "type": "string",
+                        "description": "Set to 'true' to resume streaming",
+                        "name": "stream",
+                        "in": "query"
+                    },
+                    {
+                        "type": "integer",
+                        "description": "Sequence number to resume from (for streaming)",
+                        "name": "starting_after",
+                        "in": "query"
+                    }
+                ],
+                "responses": {
+                    "200": {
+                        "description": "Response",
+                        "schema": {
+                            "$ref": "#/definitions/schema.ORResponseResource"
+                        }
+                    },
+                    "400": {
+                        "description": "Bad Request",
+                        "schema": {
+                            "type": "object",
+                            "additionalProperties": true
+                        }
+                    },
+                    "404": {
+                        "description": "Not Found",
+                        "schema": {
+                            "type": "object",
+                            "additionalProperties": true
+                        }
+                    }
+                }
+            }
+        },
+        "/v1/responses/{id}/cancel": {
+            "post": {
+                "description": "Cancel a background response if it's still in progress",
+                "summary": "Cancel a response",
+                "parameters": [
+                    {
+                        "type": "string",
+                        "description": "Response ID",
+                        "name": "id",
+                        "in": "path",
+                        "required": true
+                    }
+                ],
+                "responses": {
+                    "200": {
+                        "description": "Response",
+                        "schema": {
+                            "$ref": "#/definitions/schema.ORResponseResource"
+                        }
+                    },
+                    "400": {
+                        "description": "Bad Request",
+                        "schema": {
+                            "type": "object",
+                            "additionalProperties": true
+                        }
+                    },
+                    "404": {
+                        "description": "Not Found",
+                        "schema": {
+                            "type": "object",
+                            "additionalProperties": true
+                        }
+                    }
+                }
+            }
+        },
         "/v1/sound-generation": {
             "post": {
                 "summary": "Generates audio from the input text.",
@@ -2507,6 +2617,322 @@ const docTemplate = `{
                 }
             }
         },
+        "schema.ORError": {
+            "type": "object",
+            "properties": {
+                "code": {
+                    "type": "string"
+                },
+                "message": {
+                    "type": "string"
+                },
+                "param": {
+                    "type": "string"
+                },
+                "type": {
+                    "description": "invalid_request|not_found|server_error|model_error|too_many_requests",
+                    "type": "string"
+                }
+            }
+        },
+        "schema.ORFunctionTool": {
+            "type": "object",
+            "properties": {
+                "description": {
+                    "type": "string"
+                },
+                "name": {
+                    "type": "string"
+                },
+                "parameters": {
+                    "type": "object",
+                    "additionalProperties": true
+                },
+                "strict": {
+                    "description": "Always include in response",
+                    "type": "boolean"
+                },
+                "type": {
+                    "description": "always \"function\"",
+                    "type": "string"
+                }
+            }
+        },
+        "schema.ORIncompleteDetails": {
+            "type": "object",
+            "properties": {
+                "reason": {
+                    "type": "string"
+                }
+            }
+        },
+        "schema.ORInputTokensDetails": {
+            "type": "object",
+            "properties": {
+                "cached_tokens": {
+                    "description": "Always include, even if 0",
+                    "type": "integer"
+                }
+            }
+        },
+        "schema.ORItemField": {
+            "type": "object",
+            "properties": {
+                "arguments": {
+                    "type": "string"
+                },
+                "call_id": {
+                    "description": "Function call fields",
+                    "type": "string"
+                },
+                "content": {
+                    "description": "string or []ORContentPart for messages"
+                },
+                "id": {
+                    "description": "Present for all output items",
+                    "type": "string"
+                },
+                "name": {
+                    "type": "string"
+                },
+                "output": {
+                    "description": "Function call output fields"
+                },
+                "role": {
+                    "description": "Message fields",
+                    "type": "string"
+                },
+                "status": {
+                    "description": "in_progress|completed|incomplete",
+                    "type": "string"
+                },
+                "type": {
+                    "description": "message|function_call|function_call_output|reasoning|item_reference",
+                    "type": "string"
+                }
+            }
+        },
+        "schema.OROutputTokensDetails": {
+            "type": "object",
+            "properties": {
+                "reasoning_tokens": {
+                    "description": "Always include, even if 0",
+                    "type": "integer"
+                }
+            }
+        },
+        "schema.ORReasoning": {
+            "type": "object",
+            "properties": {
+                "effort": {
+                    "type": "string"
+                },
+                "summary": {
+                    "type": "string"
+                }
+            }
+        },
+        "schema.ORReasoningParam": {
+            "type": "object",
+            "properties": {
+                "effort": {
+                    "description": "\"none\"|\"low\"|\"medium\"|\"high\"|\"xhigh\"",
+                    "type": "string"
+                },
+                "summary": {
+                    "description": "\"auto\"|\"concise\"|\"detailed\"",
+                    "type": "string"
+                }
+            }
+        },
+        "schema.ORResponseResource": {
+            "type": "object",
+            "properties": {
+                "background": {
+                    "type": "boolean"
+                },
+                "completed_at": {
+                    "description": "Required: present as number or null",
+                    "type": "integer"
+                },
+                "created_at": {
+                    "type": "integer"
+                },
+                "error": {
+                    "description": "Always present, null if no error",
+                    "allOf": [
+                        {
+                            "$ref": "#/definitions/schema.ORError"
+                        }
+                    ]
+                },
+                "frequency_penalty": {
+                    "type": "number"
+                },
+                "id": {
+                    "type": "string"
+                },
+                "incomplete_details": {
+                    "description": "Always present, null if complete",
+                    "allOf": [
+                        {
+                            "$ref": "#/definitions/schema.ORIncompleteDetails"
+                        }
+                    ]
+                },
+                "instructions": {
+                    "type": "string"
+                },
+                "max_output_tokens": {
+                    "type": "integer"
+                },
+                "max_tool_calls": {
+                    "description": "nullable",
+                    "type": "integer"
+                },
+                "metadata": {
+                    "description": "Metadata and operational flags",
+                    "type": "object",
+                    "additionalProperties": {
+                        "type": "string"
+                    }
+                },
+                "model": {
+                    "type": "string"
+                },
+                "object": {
+                    "description": "always \"response\"",
+                    "type": "string"
+                },
+                "output": {
+                    "type": "array",
+                    "items": {
+                        "$ref": "#/definitions/schema.ORItemField"
+                    }
+                },
+                "parallel_tool_calls": {
+                    "type": "boolean"
+                },
+                "presence_penalty": {
+                    "type": "number"
+                },
+                "previous_response_id": {
+                    "type": "string"
+                },
+                "prompt_cache_key": {
+                    "description": "nullable",
+                    "type": "string"
+                },
+                "reasoning": {
+                    "description": "nullable",
+                    "allOf": [
+                        {
+                            "$ref": "#/definitions/schema.ORReasoning"
+                        }
+                    ]
+                },
+                "safety_identifier": {
+                    "description": "Safety and caching",
+                    "type": "string"
+                },
+                "service_tier": {
+                    "type": "string"
+                },
+                "status": {
+                    "description": "in_progress|completed|failed|incomplete",
+                    "type": "string"
+                },
+                "store": {
+                    "type": "boolean"
+                },
+                "temperature": {
+                    "description": "Sampling parameters (always required)",
+                    "type": "number"
+                },
+                "text": {
+                    "description": "Text format configuration",
+                    "allOf": [
+                        {
+                            "$ref": "#/definitions/schema.ORTextConfig"
+                        }
+                    ]
+                },
+                "tool_choice": {},
+                "tools": {
+                    "description": "Tool-related fields",
+                    "type": "array",
+                    "items": {
+                        "$ref": "#/definitions/schema.ORFunctionTool"
+                    }
+                },
+                "top_logprobs": {
+                    "description": "Default to 0",
+                    "type": "integer"
+                },
+                "top_p": {
+                    "type": "number"
+                },
+                "truncation": {
+                    "description": "Truncation and reasoning",
+                    "type": "string"
+                },
+                "usage": {
+                    "description": "Usage statistics",
+                    "allOf": [
+                        {
+                            "$ref": "#/definitions/schema.ORUsage"
+                        }
+                    ]
+                }
+            }
+        },
+        "schema.ORTextConfig": {
+            "type": "object",
+            "properties": {
+                "format": {
+                    "$ref": "#/definitions/schema.ORTextFormat"
+                }
+            }
+        },
+        "schema.ORTextFormat": {
+            "type": "object",
+            "properties": {
+                "type": {
+                    "description": "\"text\" or \"json_schema\"",
+                    "type": "string"
+                }
+            }
+        },
+        "schema.ORUsage": {
+            "type": "object",
+            "properties": {
+                "input_tokens": {
+                    "type": "integer"
+                },
+                "input_tokens_details": {
+                    "description": "Always present",
+                    "allOf": [
+                        {
+                            "$ref": "#/definitions/schema.ORInputTokensDetails"
+                        }
+                    ]
+                },
+                "output_tokens": {
+                    "type": "integer"
+                },
+                "output_tokens_details": {
+                    "description": "Always present",
+                    "allOf": [
+                        {
+                            "$ref": "#/definitions/schema.OROutputTokensDetails"
+                        }
+                    ]
+                },
+                "total_tokens": {
+                    "type": "integer"
+                }
+            }
+        },
         "schema.OpenAIModel": {
             "type": "object",
             "properties": {
@@ -2781,6 +3207,114 @@ const docTemplate = `{
                 }
             }
         },
+        "schema.OpenResponsesRequest": {
+            "type": "object",
+            "properties": {
+                "allowed_tools": {
+                    "description": "Restrict which tools can be invoked",
+                    "type": "array",
+                    "items": {
+                        "type": "string"
+                    }
+                },
+                "background": {
+                    "description": "Run request in background",
+                    "type": "boolean"
+                },
+                "frequency_penalty": {
+                    "description": "Frequency penalty (-2.0 to 2.0)",
+                    "type": "number"
+                },
+                "include": {
+                    "description": "What to include in response",
+                    "type": "array",
+                    "items": {
+                        "type": "string"
+                    }
+                },
+                "input": {
+                    "description": "string or []ORItemParam"
+                },
+                "instructions": {
+                    "type": "string"
+                },
+                "logit_bias": {
+                    "description": "OpenAI-compatible extensions (not in Open Responses spec)",
+                    "type": "object",
+                    "additionalProperties": {
+                        "type": "number",
+                        "format": "float64"
+                    }
+                },
+                "max_output_tokens": {
+                    "type": "integer"
+                },
+                "max_tool_calls": {
+                    "description": "Maximum number of tool calls",
+                    "type": "integer"
+                },
+                "metadata": {
+                    "type": "object",
+                    "additionalProperties": {
+                        "type": "string"
+                    }
+                },
+                "model": {
+                    "type": "string"
+                },
+                "parallel_tool_calls": {
+                    "description": "Allow parallel tool calls",
+                    "type": "boolean"
+                },
+                "presence_penalty": {
+                    "description": "Presence penalty (-2.0 to 2.0)",
+                    "type": "number"
+                },
+                "previous_response_id": {
+                    "type": "string"
+                },
+                "reasoning": {
+                    "$ref": "#/definitions/schema.ORReasoningParam"
+                },
+                "service_tier": {
+                    "description": "\"auto\"|\"default\"|priority hint",
+                    "type": "string"
+                },
+                "store": {
+                    "description": "Whether to store the response",
+                    "type": "boolean"
+                },
+                "stream": {
+                    "type": "boolean"
+                },
+                "temperature": {
+                    "type": "number"
+                },
+                "text_format": {
+                    "description": "Additional parameters from spec"
+                },
+                "tool_choice": {
+                    "description": "\"auto\"|\"required\"|\"none\"|{type:\"function\",name:\"...\"}"
+                },
+                "tools": {
+                    "type": "array",
+                    "items": {
+                        "$ref": "#/definitions/schema.ORFunctionTool"
+                    }
+                },
+                "top_logprobs": {
+                    "description": "Number of top logprobs to return",
+                    "type": "integer"
+                },
+                "top_p": {
+                    "type": "number"
+                },
+                "truncation": {
+                    "description": "\"auto\"|\"disabled\"",
+                    "type": "string"
+                }
+            }
+        },
         "schema.P2PNodesResponse": {
             "type": "object",
             "properties": {

swagger/swagger.json

@@ -1252,6 +1252,116 @@
                 }
             }
         },
+        "/v1/responses": {
+            "post": {
+                "summary": "Create a response using the Open Responses API",
+                "parameters": [
+                    {
+                        "description": "Request body",
+                        "name": "request",
+                        "in": "body",
+                        "required": true,
+                        "schema": {
+                            "$ref": "#/definitions/schema.OpenResponsesRequest"
+                        }
+                    }
+                ],
+                "responses": {
+                    "200": {
+                        "description": "Response",
+                        "schema": {
+                            "$ref": "#/definitions/schema.ORResponseResource"
+                        }
+                    }
+                }
+            }
+        },
+        "/v1/responses/{id}": {
+            "get": {
+                "description": "Retrieve a response by ID. Can be used for polling background responses or resuming streaming responses.",
+                "summary": "Get a response by ID",
+                "parameters": [
+                    {
+                        "type": "string",
+                        "description": "Response ID",
+                        "name": "id",
+                        "in": "path",
+                        "required": true
+                    },
+                    {
+                        "type": "string",
+                        "description": "Set to 'true' to resume streaming",
+                        "name": "stream",
+                        "in": "query"
+                    },
+                    {
+                        "type": "integer",
+                        "description": "Sequence number to resume from (for streaming)",
+                        "name": "starting_after",
+                        "in": "query"
+                    }
+                ],
+                "responses": {
+                    "200": {
+                        "description": "Response",
+                        "schema": {
+                            "$ref": "#/definitions/schema.ORResponseResource"
+                        }
+                    },
+                    "400": {
+                        "description": "Bad Request",
+                        "schema": {
+                            "type": "object",
+                            "additionalProperties": true
+                        }
+                    },
+                    "404": {
+                        "description": "Not Found",
+                        "schema": {
+                            "type": "object",
+                            "additionalProperties": true
+                        }
+                    }
+                }
+            }
+        },
+        "/v1/responses/{id}/cancel": {
+            "post": {
+                "description": "Cancel a background response if it's still in progress",
+                "summary": "Cancel a response",
+                "parameters": [
+                    {
+                        "type": "string",
+                        "description": "Response ID",
+                        "name": "id",
+                        "in": "path",
+                        "required": true
+                    }
+                ],
+                "responses": {
+                    "200": {
+                        "description": "Response",
+                        "schema": {
+                            "$ref": "#/definitions/schema.ORResponseResource"
+                        }
+                    },
+                    "400": {
+                        "description": "Bad Request",
+                        "schema": {
+                            "type": "object",
+                            "additionalProperties": true
+                        }
+                    },
+                    "404": {
+                        "description": "Not Found",
+                        "schema": {
+                            "type": "object",
+                            "additionalProperties": true
+                        }
+                    }
+                }
+            }
+        },
         "/v1/sound-generation": {
             "post": {
                 "summary": "Generates audio from the input text.",
@@ -2500,6 +2610,322 @@
                 }
             }
         },
+        "schema.ORError": {
+            "type": "object",
+            "properties": {
+                "code": {
+                    "type": "string"
+                },
+                "message": {
+                    "type": "string"
+                },
+                "param": {
+                    "type": "string"
+                },
+                "type": {
+                    "description": "invalid_request|not_found|server_error|model_error|too_many_requests",
+                    "type": "string"
+                }
+            }
+        },
+        "schema.ORFunctionTool": {
+            "type": "object",
+            "properties": {
+                "description": {
+                    "type": "string"
+                },
+                "name": {
+                    "type": "string"
+                },
+                "parameters": {
+                    "type": "object",
+                    "additionalProperties": true
+                },
+                "strict": {
+                    "description": "Always include in response",
+                    "type": "boolean"
+                },
+                "type": {
+                    "description": "always \"function\"",
+                    "type": "string"
+                }
+            }
+        },
+        "schema.ORIncompleteDetails": {
+            "type": "object",
+            "properties": {
+                "reason": {
+                    "type": "string"
+                }
+            }
+        },
+        "schema.ORInputTokensDetails": {
+            "type": "object",
+            "properties": {
+                "cached_tokens": {
+                    "description": "Always include, even if 0",
+                    "type": "integer"
+                }
+            }
+        },
+        "schema.ORItemField": {
+            "type": "object",
+            "properties": {
+                "arguments": {
+                    "type": "string"
+                },
+                "call_id": {
+                    "description": "Function call fields",
+                    "type": "string"
+                },
+                "content": {
+                    "description": "string or []ORContentPart for messages"
+                },
+                "id": {
+                    "description": "Present for all output items",
+                    "type": "string"
+                },
+                "name": {
+                    "type": "string"
+                },
+                "output": {
+                    "description": "Function call output fields"
+                },
+                "role": {
+                    "description": "Message fields",
+                    "type": "string"
+                },
+                "status": {
+                    "description": "in_progress|completed|incomplete",
+                    "type": "string"
+                },
+                "type": {
+                    "description": "message|function_call|function_call_output|reasoning|item_reference",
+                    "type": "string"
+                }
+            }
+        },
+        "schema.OROutputTokensDetails": {
+            "type": "object",
+            "properties": {
+                "reasoning_tokens": {
+                    "description": "Always include, even if 0",
+                    "type": "integer"
+                }
+            }
+        },
+        "schema.ORReasoning": {
+            "type": "object",
+            "properties": {
+                "effort": {
+                    "type": "string"
+                },
+                "summary": {
+                    "type": "string"
+                }
+            }
+        },
+        "schema.ORReasoningParam": {
+            "type": "object",
+            "properties": {
+                "effort": {
+                    "description": "\"none\"|\"low\"|\"medium\"|\"high\"|\"xhigh\"",
+                    "type": "string"
+                },
+                "summary": {
+                    "description": "\"auto\"|\"concise\"|\"detailed\"",
+                    "type": "string"
+                }
+            }
+        },
+        "schema.ORResponseResource": {
+            "type": "object",
+            "properties": {
+                "background": {
+                    "type": "boolean"
+                },
+                "completed_at": {
+                    "description": "Required: present as number or null",
+                    "type": "integer"
+                },
+                "created_at": {
+                    "type": "integer"
+                },
+                "error": {
+                    "description": "Always present, null if no error",
+                    "allOf": [
+                        {
+                            "$ref": "#/definitions/schema.ORError"
+                        }
+                    ]
+                },
+                "frequency_penalty": {
+                    "type": "number"
+                },
+                "id": {
+                    "type": "string"
+                },
+                "incomplete_details": {
+                    "description": "Always present, null if complete",
+                    "allOf": [
+                        {
+                            "$ref": "#/definitions/schema.ORIncompleteDetails"
+                        }
+                    ]
+                },
+                "instructions": {
+                    "type": "string"
+                },
+                "max_output_tokens": {
+                    "type": "integer"
+                },
+                "max_tool_calls": {
+                    "description": "nullable",
+                    "type": "integer"
+                },
+                "metadata": {
+                    "description": "Metadata and operational flags",
+                    "type": "object",
+                    "additionalProperties": {
+                        "type": "string"
+                    }
+                },
+                "model": {
+                    "type": "string"
+                },
+                "object": {
+                    "description": "always \"response\"",
+                    "type": "string"
+                },
+                "output": {
+                    "type": "array",
+                    "items": {
+                        "$ref": "#/definitions/schema.ORItemField"
+                    }
+                },
+                "parallel_tool_calls": {
+                    "type": "boolean"
+                },
+                "presence_penalty": {
+                    "type": "number"
+                },
+                "previous_response_id": {
+                    "type": "string"
+                },
+                "prompt_cache_key": {
+                    "description": "nullable",
+                    "type": "string"
+                },
+                "reasoning": {
+                    "description": "nullable",
+                    "allOf": [
+                        {
+                            "$ref": "#/definitions/schema.ORReasoning"
+                        }
+                    ]
+                },
+                "safety_identifier": {
+                    "description": "Safety and caching",
+                    "type": "string"
+                },
+                "service_tier": {
+                    "type": "string"
+                },
+                "status": {
+                    "description": "in_progress|completed|failed|incomplete",
+                    "type": "string"
+                },
+                "store": {
+                    "type": "boolean"
+                },
+                "temperature": {
+                    "description": "Sampling parameters (always required)",
+                    "type": "number"
+                },
+                "text": {
+                    "description": "Text format configuration",
+                    "allOf": [
+                        {
+                            "$ref": "#/definitions/schema.ORTextConfig"
+                        }
+                    ]
+                },
+                "tool_choice": {},
+                "tools": {
+                    "description": "Tool-related fields",
+                    "type": "array",
+                    "items": {
+                        "$ref": "#/definitions/schema.ORFunctionTool"
+                    }
+                },
+                "top_logprobs": {
+                    "description": "Default to 0",
+                    "type": "integer"
+                },
+                "top_p": {
+                    "type": "number"
+                },
+                "truncation": {
+                    "description": "Truncation and reasoning",
+                    "type": "string"
+                },
+                "usage": {
+                    "description": "Usage statistics",
+                    "allOf": [
+                        {
+                            "$ref": "#/definitions/schema.ORUsage"
+                        }
+                    ]
+                }
+            }
+        },
+        "schema.ORTextConfig": {
+            "type": "object",
+            "properties": {
+                "format": {
+                    "$ref": "#/definitions/schema.ORTextFormat"
+                }
+            }
+        },
+        "schema.ORTextFormat": {
+            "type": "object",
+            "properties": {
+                "type": {
+                    "description": "\"text\" or \"json_schema\"",
+                    "type": "string"
+                }
+            }
+        },
+        "schema.ORUsage": {
+            "type": "object",
+            "properties": {
+                "input_tokens": {
+                    "type": "integer"
+                },
+                "input_tokens_details": {
+                    "description": "Always present",
+                    "allOf": [
+                        {
+                            "$ref": "#/definitions/schema.ORInputTokensDetails"
+                        }
+                    ]
+                },
+                "output_tokens": {
+                    "type": "integer"
+                },
+                "output_tokens_details": {
+                    "description": "Always present",
+                    "allOf": [
+                        {
+                            "$ref": "#/definitions/schema.OROutputTokensDetails"
+                        }
+                    ]
+                },
+                "total_tokens": {
+                    "type": "integer"
+                }
+            }
+        },
         "schema.OpenAIModel": {
             "type": "object",
             "properties": {
@@ -2774,6 +3200,114 @@
                 }
             }
         },
+        "schema.OpenResponsesRequest": {
+            "type": "object",
+            "properties": {
+                "allowed_tools": {
+                    "description": "Restrict which tools can be invoked",
+                    "type": "array",
+                    "items": {
+                        "type": "string"
+                    }
+                },
+                "background": {
+                    "description": "Run request in background",
+                    "type": "boolean"
+                },
+                "frequency_penalty": {
+                    "description": "Frequency penalty (-2.0 to 2.0)",
+                    "type": "number"
+                },
+                "include": {
+                    "description": "What to include in response",
+                    "type": "array",
+                    "items": {
+                        "type": "string"
+                    }
+                },
+                "input": {
+                    "description": "string or []ORItemParam"
+                },
+                "instructions": {
+                    "type": "string"
+                },
+                "logit_bias": {
+                    "description": "OpenAI-compatible extensions (not in Open Responses spec)",
+                    "type": "object",
+                    "additionalProperties": {
+                        "type": "number",
+                        "format": "float64"
+                    }
+                },
+                "max_output_tokens": {
+                    "type": "integer"
+                },
+                "max_tool_calls": {
+                    "description": "Maximum number of tool calls",
+                    "type": "integer"
+                },
+                "metadata": {
+                    "type": "object",
+                    "additionalProperties": {
+                        "type": "string"
+                    }
+                },
+                "model": {
+                    "type": "string"
+                },
+                "parallel_tool_calls": {
+                    "description": "Allow parallel tool calls",
+                    "type": "boolean"
+                },
+                "presence_penalty": {
+                    "description": "Presence penalty (-2.0 to 2.0)",
+                    "type": "number"
+                },
+                "previous_response_id": {
+                    "type": "string"
+                },
+                "reasoning": {
+                    "$ref": "#/definitions/schema.ORReasoningParam"
+                },
+                "service_tier": {
+                    "description": "\"auto\"|\"default\"|priority hint",
+                    "type": "string"
+                },
+                "store": {
+                    "description": "Whether to store the response",
+                    "type": "boolean"
+                },
+                "stream": {
+                    "type": "boolean"
+                },
+                "temperature": {
+                    "type": "number"
+                },
+                "text_format": {
+                    "description": "Additional parameters from spec"
+                },
+                "tool_choice": {
+                    "description": "\"auto\"|\"required\"|\"none\"|{type:\"function\",name:\"...\"}"
+                },
+                "tools": {
+                    "type": "array",
+                    "items": {
+                        "$ref": "#/definitions/schema.ORFunctionTool"
+                    }
+                },
+                "top_logprobs": {
+                    "description": "Number of top logprobs to return",
+                    "type": "integer"
+                },
+                "top_p": {
+                    "type": "number"
+                },
+                "truncation": {
+                    "description": "\"auto\"|\"disabled\"",
+                    "type": "string"
+                }
+            }
+        },
         "schema.P2PNodesResponse": {
             "type": "object",
             "properties": {

swagger/swagger.yaml

@@ -742,6 +742,212 @@ definitions:
       tunnelAddress:
         type: string
     type: object
+  schema.ORError:
+    properties:
+      code:
+        type: string
+      message:
+        type: string
+      param:
+        type: string
+      type:
+        description: invalid_request|not_found|server_error|model_error|too_many_requests
+        type: string
+    type: object
+  schema.ORFunctionTool:
+    properties:
+      description:
+        type: string
+      name:
+        type: string
+      parameters:
+        additionalProperties: true
+        type: object
+      strict:
+        description: Always include in response
+        type: boolean
+      type:
+        description: always "function"
+        type: string
+    type: object
+  schema.ORIncompleteDetails:
+    properties:
+      reason:
+        type: string
+    type: object
+  schema.ORInputTokensDetails:
+    properties:
+      cached_tokens:
+        description: Always include, even if 0
+        type: integer
+    type: object
+  schema.ORItemField:
+    properties:
+      arguments:
+        type: string
+      call_id:
+        description: Function call fields
+        type: string
+      content:
+        description: string or []ORContentPart for messages
+      id:
+        description: Present for all output items
+        type: string
+      name:
+        type: string
+      output:
+        description: Function call output fields
+      role:
+        description: Message fields
+        type: string
+      status:
+        description: in_progress|completed|incomplete
+        type: string
+      type:
+        description: message|function_call|function_call_output|reasoning|item_reference
+        type: string
+    type: object
+  schema.OROutputTokensDetails:
+    properties:
+      reasoning_tokens:
+        description: Always include, even if 0
+        type: integer
+    type: object
+  schema.ORReasoning:
+    properties:
+      effort:
+        type: string
+      summary:
+        type: string
+    type: object
+  schema.ORReasoningParam:
+    properties:
+      effort:
+        description: '"none"|"low"|"medium"|"high"|"xhigh"'
+        type: string
+      summary:
+        description: '"auto"|"concise"|"detailed"'
+        type: string
+    type: object
+  schema.ORResponseResource:
+    properties:
+      background:
+        type: boolean
+      completed_at:
+        description: 'Required: present as number or null'
+        type: integer
+      created_at:
+        type: integer
+      error:
+        allOf:
+        - $ref: '#/definitions/schema.ORError'
+        description: Always present, null if no error
+      frequency_penalty:
+        type: number
+      id:
+        type: string
+      incomplete_details:
+        allOf:
+        - $ref: '#/definitions/schema.ORIncompleteDetails'
+        description: Always present, null if complete
+      instructions:
+        type: string
+      max_output_tokens:
+        type: integer
+      max_tool_calls:
+        description: nullable
+        type: integer
+      metadata:
+        additionalProperties:
+          type: string
+        description: Metadata and operational flags
+        type: object
+      model:
+        type: string
+      object:
+        description: always "response"
+        type: string
+      output:
+        items:
+          $ref: '#/definitions/schema.ORItemField'
+        type: array
+      parallel_tool_calls:
+        type: boolean
+      presence_penalty:
+        type: number
+      previous_response_id:
+        type: string
+      prompt_cache_key:
+        description: nullable
+        type: string
+      reasoning:
+        allOf:
+        - $ref: '#/definitions/schema.ORReasoning'
+        description: nullable
+      safety_identifier:
+        description: Safety and caching
+        type: string
+      service_tier:
+        type: string
+      status:
+        description: in_progress|completed|failed|incomplete
+        type: string
+      store:
+        type: boolean
+      temperature:
+        description: Sampling parameters (always required)
+        type: number
+      text:
+        allOf:
+        - $ref: '#/definitions/schema.ORTextConfig'
+        description: Text format configuration
+      tool_choice: {}
+      tools:
+        description: Tool-related fields
+        items:
+          $ref: '#/definitions/schema.ORFunctionTool'
+        type: array
+      top_logprobs:
+        description: Default to 0
+        type: integer
+      top_p:
+        type: number
+      truncation:
+        description: Truncation and reasoning
+        type: string
+      usage:
+        allOf:
+        - $ref: '#/definitions/schema.ORUsage'
+        description: Usage statistics
+    type: object
+  schema.ORTextConfig:
+    properties:
+      format:
+        $ref: '#/definitions/schema.ORTextFormat'
+    type: object
+  schema.ORTextFormat:
+    properties:
+      type:
+        description: '"text" or "json_schema"'
+        type: string
+    type: object
+  schema.ORUsage:
+    properties:
+      input_tokens:
+        type: integer
+      input_tokens_details:
+        allOf:
+        - $ref: '#/definitions/schema.ORInputTokensDetails'
+        description: Always present
+      output_tokens:
+        type: integer
+      output_tokens_details:
+        allOf:
+        - $ref: '#/definitions/schema.OROutputTokensDetails'
+        description: Always present
+      total_tokens:
+        type: integer
+    type: object
   schema.OpenAIModel:
     properties:
       id:
@@ -936,6 +1142,82 @@ definitions:
       total_tokens:
         type: integer
     type: object
+  schema.OpenResponsesRequest:
+    properties:
+      allowed_tools:
+        description: Restrict which tools can be invoked
+        items:
+          type: string
+        type: array
+      background:
+        description: Run request in background
+        type: boolean
+      frequency_penalty:
+        description: Frequency penalty (-2.0 to 2.0)
+        type: number
+      include:
+        description: What to include in response
+        items:
+          type: string
+        type: array
+      input:
+        description: string or []ORItemParam
+      instructions:
+        type: string
+      logit_bias:
+        additionalProperties:
+          format: float64
+          type: number
+        description: OpenAI-compatible extensions (not in Open Responses spec)
+        type: object
+      max_output_tokens:
+        type: integer
+      max_tool_calls:
+        description: Maximum number of tool calls
+        type: integer
+      metadata:
+        additionalProperties:
+          type: string
+        type: object
+      model:
+        type: string
+      parallel_tool_calls:
+        description: Allow parallel tool calls
+        type: boolean
+      presence_penalty:
+        description: Presence penalty (-2.0 to 2.0)
+        type: number
+      previous_response_id:
+        type: string
+      reasoning:
+        $ref: '#/definitions/schema.ORReasoningParam'
+      service_tier:
+        description: '"auto"|"default"|priority hint'
+        type: string
+      store:
+        description: Whether to store the response
+        type: boolean
+      stream:
+        type: boolean
+      temperature:
+        type: number
+      text_format:
+        description: Additional parameters from spec
+      tool_choice:
+        description: '"auto"|"required"|"none"|{type:"function",name:"..."}'
+      tools:
+        items:
+          $ref: '#/definitions/schema.ORFunctionTool'
+        type: array
+      top_logprobs:
+        description: Number of top logprobs to return
+        type: integer
+      top_p:
+        type: number
+      truncation:
+        description: '"auto"|"disabled"'
+        type: string
+    type: object
   schema.P2PNodesResponse:
     properties:
       federated_nodes:
@@ -1962,6 +2244,80 @@ paths:
           schema:
             $ref: '#/definitions/schema.JINARerankResponse'
       summary: Reranks a list of phrases by relevance to a given text query.
+  /v1/responses:
+    post:
+      parameters:
+      - description: Request body
+        in: body
+        name: request
+        required: true
+        schema:
+          $ref: '#/definitions/schema.OpenResponsesRequest'
+      responses:
+        "200":
+          description: Response
+          schema:
+            $ref: '#/definitions/schema.ORResponseResource'
+      summary: Create a response using the Open Responses API
+  /v1/responses/{id}:
+    get:
+      description: Retrieve a response by ID. Can be used for polling background responses
+        or resuming streaming responses.
+      parameters:
+      - description: Response ID
+        in: path
+        name: id
+        required: true
+        type: string
+      - description: Set to 'true' to resume streaming
+        in: query
+        name: stream
+        type: string
+      - description: Sequence number to resume from (for streaming)
+        in: query
+        name: starting_after
+        type: integer
+      responses:
+        "200":
+          description: Response
+          schema:
+            $ref: '#/definitions/schema.ORResponseResource'
+        "400":
+          description: Bad Request
+          schema:
+            additionalProperties: true
+            type: object
+        "404":
+          description: Not Found
+          schema:
+            additionalProperties: true
+            type: object
+      summary: Get a response by ID
+  /v1/responses/{id}/cancel:
+    post:
+      description: Cancel a background response if it's still in progress
+      parameters:
+      - description: Response ID
+        in: path
+        name: id
+        required: true
+        type: string
+      responses:
+        "200":
+          description: Response
+          schema:
+            $ref: '#/definitions/schema.ORResponseResource'
+        "400":
+          description: Bad Request
+          schema:
+            additionalProperties: true
+            type: object
+        "404":
+          description: Not Found
+          schema:
+            additionalProperties: true
+            type: object
+      summary: Cancel a response
   /v1/sound-generation:
     post:
       parameters: