Set model

Don't require download check
Fix sending images
2026-02-27 11:48:40 -05:00 · 2026-02-26 15:28:58 -07:00 · 2026-02-26 14:33:08 -07:00 · 2026-02-26 14:33:08 -07:00 · 2026-02-26 14:33:08 -07:00 · 2026-02-26 14:33:08 -07:00
56 changed files with 3719 additions and 430 deletions
--- a/docker/tensorrt/requirements-amd64.txt
+++ b/docker/tensorrt/requirements-amd64.txt
@@ -15,4 +15,4 @@ nvidia_nccl_cu12==2.23.4; platform_machine == 'x86_64'
 nvidia_nvjitlink_cu12==12.5.82; platform_machine == 'x86_64'
 onnx==1.16.*; platform_machine == 'x86_64'
 onnxruntime-gpu==1.22.*; platform_machine == 'x86_64'
-protobuf==7.34.0; platform_machine == 'x86_64'
+protobuf==3.20.3; platform_machine == 'x86_64'
--- a/docker/tensorrt/requirements-models-arm64.txt
+++ b/docker/tensorrt/requirements-models-arm64.txt
@@ -1,2 +1,2 @@
 onnx == 1.14.0; platform_machine == 'aarch64'
-protobuf == 7.34.0; platform_machine == 'aarch64'
+protobuf == 3.20.3; platform_machine == 'aarch64'
--- a/docs/docs/configuration/hardware_acceleration_enrichments.md
+++ b/docs/docs/configuration/hardware_acceleration_enrichments.md
@@ -12,20 +12,23 @@ Some of Frigate's enrichments can use a discrete GPU or integrated GPU for accel
 Object detection and enrichments (like Semantic Search, Face Recognition, and License Plate Recognition) are independent features. To use a GPU / NPU for object detection, see the [Object Detectors](/configuration/object_detectors.md) documentation. If you want to use your GPU for any supported enrichments, you must choose the appropriate Frigate Docker image for your GPU / NPU and configure the enrichment according to its specific documentation.

 - **AMD**
+
  - ROCm support in the `-rocm` Frigate image is automatically detected for enrichments, but only some enrichment models are available due to ROCm's focus on LLMs and limited stability with certain neural network models. Frigate disables models that perform poorly or are unstable to ensure reliable operation, so only compatible enrichments may be active.

 - **Intel**
+
  - OpenVINO will automatically be detected and used for enrichments in the default Frigate image.
  - **Note:** Intel NPUs have limited model support for enrichments. GPU is recommended for enrichments when available.

 - **Nvidia**
+
  - Nvidia GPUs will automatically be detected and used for enrichments in the `-tensorrt` Frigate image.
  - Jetson devices will automatically be detected and used for enrichments in the `-tensorrt-jp6` Frigate image.

 - **RockChip**
  - RockChip NPU will automatically be detected and used for semantic search v1 and face recognition in the `-rk` Frigate image.

-Utilizing a GPU for enrichments does not require you to use the same GPU for object detection. For example, you can run the `tensorrt` Docker image to run enrichments on an Nvidia GPU and still use other dedicated hardware like a Coral or Hailo for object detection. However, one combination that is not supported is the `tensorrt` image for object detection on an Nvidia GPU and Intel iGPU for enrichments.
+Utilizing a GPU for enrichments does not require you to use the same GPU for object detection. For example, you can run the `tensorrt` Docker image for enrichments and still use other dedicated hardware like a Coral or Hailo for object detection. However, one combination that is not supported is TensorRT for object detection and OpenVINO for enrichments.

 :::note

--- a/docs/docs/configuration/index.md
+++ b/docs/docs/configuration/index.md
@@ -29,12 +29,12 @@ cameras:

 When running Frigate through the HA Add-on, the Frigate `/config` directory is mapped to `/addon_configs/<addon_directory>` in the host, where `<addon_directory>` is specific to the variant of the Frigate Add-on you are running.

-| Add-on Variant             | Configuration directory                   |
-| -------------------------- | ----------------------------------------- |
-| Frigate                    | `/addon_configs/ccab4aaf_frigate`         |
-| Frigate (Full Access)      | `/addon_configs/ccab4aaf_frigate-fa`      |
-| Frigate Beta               | `/addon_configs/ccab4aaf_frigate-beta`    |
-| Frigate Beta (Full Access) | `/addon_configs/ccab4aaf_frigate-fa-beta` |
+| Add-on Variant             | Configuration directory                      |
+| -------------------------- | -------------------------------------------- |
+| Frigate                    | `/addon_configs/ccab4aaf_frigate`            |
+| Frigate (Full Access)      | `/addon_configs/ccab4aaf_frigate-fa`         |
+| Frigate Beta               | `/addon_configs/ccab4aaf_frigate-beta`       |
+| Frigate Beta (Full Access) | `/addon_configs/ccab4aaf_frigate-fa-beta`    |

 **Whenever you see `/config` in the documentation, it refers to this directory.**

@@ -109,16 +109,15 @@ detectors:

 record:
  enabled: True
-  motion:
+  retain:
    days: 7
+    mode: motion
  alerts:
    retain:
      days: 30
-      mode: motion
  detections:
    retain:
      days: 30
-      mode: motion

 snapshots:
  enabled: True
@@ -166,16 +165,15 @@ detectors:

 record:
  enabled: True
-  motion:
+  retain:
    days: 7
+    mode: motion
  alerts:
    retain:
      days: 30
-      mode: motion
  detections:
    retain:
      days: 30
-      mode: motion

 snapshots:
  enabled: True
@@ -233,16 +231,15 @@ model:

 record:
  enabled: True
-  motion:
+  retain:
    days: 7
+    mode: motion
  alerts:
    retain:
      days: 30
-      mode: motion
  detections:
    retain:
      days: 30
-      mode: motion

 snapshots:
  enabled: True
--- a/docs/docs/configuration/object_detectors.md
+++ b/docs/docs/configuration/object_detectors.md
@@ -34,7 +34,7 @@ Frigate supports multiple different detectors that work on different types of ha

 **Nvidia GPU**

- [ONNX](#onnx): Nvidia GPUs will automatically be detected and used as a detector in the `-tensorrt` Frigate image when a supported ONNX model is configured.
+- [ONNX](#onnx): TensorRT will automatically be detected and used as a detector in the `-tensorrt` Frigate image when a supported ONNX model is configured.

 **Nvidia Jetson** <CommunityBadge />

@@ -65,7 +65,7 @@ This does not affect using hardware for accelerating other tasks such as [semant

 # Officially Supported Detectors

-Frigate provides a number of builtin detector types. By default, Frigate will use a single CPU detector. Other detectors may require additional configuration as described below. When using multiple detectors they will run in dedicated processes, but pull from a common queue of detection requests from across all cameras.
+Frigate provides the following builtin detector types: `cpu`, `edgetpu`, `hailo8l`, `memryx`, `onnx`, `openvino`, `rknn`, and `tensorrt`. By default, Frigate will use a single CPU detector. Other detectors may require additional configuration as described below. When using multiple detectors they will run in dedicated processes, but pull from a common queue of detection requests from across all cameras.

 ## Edge TPU Detector

@@ -157,13 +157,7 @@ A TensorFlow Lite model is provided in the container at `/edgetpu_model.tflite`

 #### YOLOv9

-YOLOv9 models that are compiled for TensorFlow Lite and properly quantized are supported, but not included by default. [Instructions](#yolov9-for-google-coral-support) for downloading a model with support for the Google Coral.
-
-:::tip
-
-**Frigate+ Users:** Follow the [instructions](../integrations/plus#use-models) to set a model ID in your config file.
-
-:::
+YOLOv9 models that are compiled for TensorFlow Lite and properly quantized are supported, but not included by default. [Download the model](https://github.com/dbro/frigate-detector-edgetpu-yolo9/releases/download/v1.0/yolov9-s-relu6-best_320_int8_edgetpu.tflite), bind mount the file into the container, and provide the path with `model.path`. Note that the linked model requires a 17-label [labelmap file](https://raw.githubusercontent.com/dbro/frigate-detector-edgetpu-yolo9/refs/heads/main/labels-coco17.txt) that includes only 17 COCO classes.

 <details>
  <summary>YOLOv9 Setup & Config</summary>
@@ -660,9 +654,11 @@ ONNX is an open format for building machine learning models, Frigate supports ru
 If the correct build is used for your GPU then the GPU will be detected and used automatically.

 - **AMD**
+
  - ROCm will automatically be detected and used with the ONNX detector in the `-rocm` Frigate image.

 - **Intel**
+
  - OpenVINO will automatically be detected and used with the ONNX detector in the default Frigate image.

 - **Nvidia**
@@ -1560,11 +1556,7 @@ cd tensorrt_demos/yolo
 python3 yolo_to_onnx.py -m yolov7-320
 ```

-#### YOLOv9 for Google Coral Support
-
-[Download the model](https://github.com/dbro/frigate-detector-edgetpu-yolo9/releases/download/v1.0/yolov9-s-relu6-best_320_int8_edgetpu.tflite), bind mount the file into the container, and provide the path with `model.path`. Note that the linked model requires a 17-label [labelmap file](https://raw.githubusercontent.com/dbro/frigate-detector-edgetpu-yolo9/refs/heads/main/labels-coco17.txt) that includes only 17 COCO classes.
-
-#### YOLOv9 for other detectors
+#### YOLOv9

 YOLOv9 model can be exported as ONNX using the command below. You can copy and paste the whole thing to your terminal and execute, altering `MODEL_SIZE=t` and `IMG_SIZE=320` in the first line to the [model size](https://github.com/WongKinYiu/yolov9#performance) you would like to convert (available model sizes are `t`, `s`, `m`, `c`, and `e`, common image sizes are `320` and `640`).

--- a/docs/docs/configuration/semantic_search.md
+++ b/docs/docs/configuration/semantic_search.md
@@ -76,6 +76,40 @@ Switching between V1 and V2 requires reindexing your embeddings. The embeddings

 :::

+### GenAI Provider (llama.cpp)
+
+Frigate can use a GenAI provider for semantic search embeddings when that provider has the `embeddings` role. Currently, only **llama.cpp** supports multimodal embeddings (both text and images).
+
+To use llama.cpp for semantic search:
+
+1. Configure a GenAI provider in your config with `embeddings` in its `roles`.
+2. Set `semantic_search.model` to the GenAI config key (e.g. `default`).
+3. Start the llama.cpp server with `--embeddings` and `--mmproj` for image support:
+
+```yaml
+genai:
+  default:
+    provider: llamacpp
+    base_url: http://localhost:8080
+    model: your-model-name
+    roles:
+      - embeddings
+      - vision
+      - tools
+
+semantic_search:
+  enabled: True
+  model: default
+```
+
+The llama.cpp server must be started with `--embeddings` for the embeddings API, and `--mmproj <mmproj.gguf>` when using image embeddings. See the [llama.cpp server documentation](https://github.com/ggml-org/llama.cpp/blob/master/tools/server/README.md) for details.
+
+:::note
+
+Switching between Jina models and a GenAI provider requires reindexing. Embeddings from different backends are incompatible.
+
+:::
+
 ### GPU Acceleration

 The CLIP models are downloaded in ONNX format, and the `large` model can be accelerated using GPU hardware, when available. This depends on the Docker build that is used. You can also target a specific device in a multi-GPU installation.
--- a/docs/docs/frigate/hardware.md
+++ b/docs/docs/frigate/hardware.md
@@ -41,8 +41,8 @@ If the EQ13 is out of stock, the link below may take you to a suggested alternat
 | Name                                                                                                          | Capabilities                                                               | Notes                                               |
 | ------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------- | --------------------------------------------------- |
 | Beelink EQ13 (<a href="https://amzn.to/4jn2qVr" target="_blank" rel="nofollow noopener sponsored">Amazon</a>) | Can run object detection on several 1080p cameras with low-medium activity | Dual gigabit NICs for easy isolated camera network. |
-| Intel 1120p ([Amazon](https://www.amazon.com/Beelink-i3-1220P-Computer-Display-Gigabit/dp/B0DDCKT9YP))        | Can handle a large number of 1080p cameras with high activity              |                                                     |
-| Intel 125H ([Amazon](https://www.amazon.com/MINISFORUM-Pro-125H-Barebone-Computer-HDMI2-1/dp/B0FH21FSZM))     | Can handle a significant number of 1080p cameras with high activity        | Includes NPU for more efficient detection in 0.17+  |
+| Intel 1120p ([Amazon](https://www.amazon.com/Beelink-i3-1220P-Computer-Display-Gigabit/dp/B0DDCKT9YP)         | Can handle a large number of 1080p cameras with high activity              |                                                     |
+| Intel 125H ([Amazon](https://www.amazon.com/MINISFORUM-Pro-125H-Barebone-Computer-HDMI2-1/dp/B0FH21FSZM)      | Can handle a significant number of 1080p cameras with high activity        | Includes NPU for more efficient detection in 0.17+  |

 ## Detectors

@@ -86,7 +86,7 @@ Frigate supports multiple different detectors that work on different types of ha

 **Nvidia**

- [Nvidia GPU](#nvidia-gpus): Nvidia GPUs can provide efficient object detection.
+- [TensortRT](#tensorrt---nvidia-gpu): TensorRT can run on Nvidia GPUs to provide efficient object detection.
  - [Supports majority of model architectures via ONNX](../../configuration/object_detectors#onnx-supported-models)
  - Runs well with any size models including large

@@ -172,7 +172,7 @@ Inference speeds vary greatly depending on the CPU or GPU used, some known examp
 | Intel Arc A380 | ~ 6 ms                     |                                                   | 320: ~ 10 ms 640: ~ 22 ms | 336: 20 ms 448: 27 ms  |                                    |
 | Intel Arc A750 | ~ 4 ms                     |                                                   | 320: ~ 8 ms               |                        |                                    |

-### Nvidia GPUs
+### TensorRT - Nvidia GPU

 Frigate is able to utilize an Nvidia GPU which supports the 12.x series of CUDA libraries.

@@ -182,6 +182,8 @@ Frigate is able to utilize an Nvidia GPU which supports the 12.x series of CUDA

 Make sure your host system has the [nvidia-container-runtime](https://docs.docker.com/config/containers/resource_constraints/#access-an-nvidia-gpu) installed to pass through the GPU to the container and the host system has a compatible driver installed for your GPU.

+There are improved capabilities in newer GPU architectures that TensorRT can benefit from, such as INT8 operations and Tensor cores. The features compatible with your hardware will be optimized when the model is converted to a trt file. Currently the script provided for generating the model provides a switch to enable/disable FP16 operations. If you wish to use newer features such as INT8 optimization, more work is required.
+
 #### Compatibility References:

 [NVIDIA TensorRT Support Matrix](https://docs.nvidia.com/deeplearning/tensorrt-rtx/latest/getting-started/support-matrix.html)
@@ -190,7 +192,7 @@ Make sure your host system has the [nvidia-container-runtime](https://docs.docke

 [NVIDIA GPU Compute Capability](https://developer.nvidia.com/cuda-gpus)

-Inference is done with the `onnx` detector type. Speeds will vary greatly depending on the GPU and the model used.
+Inference speeds will vary greatly depending on the GPU and the model used.
 `tiny (t)` variants are faster than the equivalent non-tiny model, some known examples are below:

 ✅ - Accelerated with CUDA Graphs
--- a/docs/docs/frigate/installation.md
+++ b/docs/docs/frigate/installation.md
@@ -56,7 +56,7 @@ services:
    volumes:
      - /path/to/your/config:/config
      - /path/to/your/storage:/media/frigate
-      - type: tmpfs # 1GB In-memory filesystem for recording segment storage
+      - type: tmpfs # Recommended: 1GB of memory
        target: /tmp/cache
        tmpfs:
          size: 1000000000
@@ -123,7 +123,7 @@ On Raspberry Pi OS **Trixie**, the Hailo driver is no longer shipped with the ke
   :::note

   If you are **not** using a Raspberry Pi with **Bookworm OS**, skip this step and proceed directly to step 2.
-
+   
   If you are using Raspberry Pi with **Trixie OS**, also skip this step and proceed directly to step 2.

   :::
@@ -133,13 +133,13 @@ On Raspberry Pi OS **Trixie**, the Hailo driver is no longer shipped with the ke
   ```bash
   lsmod | grep hailo
   ```
-
+   
   If it shows `hailo_pci`, unload it:

   ```bash
   sudo modprobe -r hailo_pci
   ```
-
+   
   Then locate the built-in kernel driver and rename it so it cannot be loaded.
   Renaming allows the original driver to be restored later if needed.
   First, locate the currently installed kernel module:
@@ -149,29 +149,28 @@ On Raspberry Pi OS **Trixie**, the Hailo driver is no longer shipped with the ke
   ```

   Example output:
-
+   
   ```
   /lib/modules/6.6.31+rpt-rpi-2712/kernel/drivers/media/pci/hailo/hailo_pci.ko.xz
   ```
-
   Save the module path to a variable:
-
+   
   ```bash
   BUILTIN=$(modinfo -n hailo_pci)
   ```

   And rename the module by appending .bak:
-
+    
   ```bash
   sudo mv "$BUILTIN" "${BUILTIN}.bak"
   ```
-
+   
   Now refresh the kernel module map so the system recognizes the change:
-
+   
   ```bash
   sudo depmod -a
   ```
-
+   
   Reboot your Raspberry Pi:

   ```bash
@@ -207,6 +206,7 @@ On Raspberry Pi OS **Trixie**, the Hailo driver is no longer shipped with the ke
   ```

   The script will:
+
   - Install necessary build dependencies
   - Clone and build the Hailo driver from the official repository
   - Install the driver
@@ -236,18 +236,18 @@ On Raspberry Pi OS **Trixie**, the Hailo driver is no longer shipped with the ke
   ```

   Verify the driver version:
-
+   
   ```bash
   cat /sys/module/hailo_pci/version
   ```
-
+   
   Verify that the firmware was installed correctly:
-
+   
   ```bash
   ls -l /lib/firmware/hailo/hailo8_fw.bin
   ```

-   **Optional: Fix PCIe descriptor page size error**
+  **Optional: Fix PCIe descriptor page size error**

   If you encounter the following error:

@@ -462,7 +462,7 @@ services:
      - /etc/localtime:/etc/localtime:ro
      - /path/to/your/config:/config
      - /path/to/your/storage:/media/frigate
-      - type: tmpfs # 1GB In-memory filesystem for recording segment storage
+      - type: tmpfs # Recommended: 1GB of memory
        target: /tmp/cache
        tmpfs:
          size: 1000000000
@@ -502,12 +502,12 @@ The official docker image tags for the current stable version are:

 - `stable` - Standard Frigate build for amd64 & RPi Optimized Frigate build for arm64. This build includes support for Hailo devices as well.
 - `stable-standard-arm64` - Standard Frigate build for arm64
- `stable-tensorrt` - Frigate build specific for amd64 devices running an Nvidia GPU
+- `stable-tensorrt` - Frigate build specific for amd64 devices running an nvidia GPU
 - `stable-rocm` - Frigate build for [AMD GPUs](../configuration/object_detectors.md#amdrocm-gpu-detector)

 The community supported docker image tags for the current stable version are:

- `stable-tensorrt-jp6` - Frigate build optimized for Nvidia Jetson devices running Jetpack 6
+- `stable-tensorrt-jp6` - Frigate build optimized for nvidia Jetson devices running Jetpack 6
 - `stable-rk` - Frigate build for SBCs with Rockchip SoC

 ## Home Assistant Add-on
@@ -521,7 +521,7 @@ There are important limitations in HA OS to be aware of:
 - Separate local storage for media is not yet supported by Home Assistant
 - AMD GPUs are not supported because HA OS does not include the mesa driver.
 - Intel NPUs are not supported because HA OS does not include the NPU firmware.
- Nvidia GPUs are not supported because addons do not support the Nvidia runtime.
+- Nvidia GPUs are not supported because addons do not support the nvidia runtime.

 :::

@@ -694,18 +694,17 @@ Log into QNAP, open Container Station. Frigate docker container should be listed

 :::warning

-macOS uses port 5000 for its Airplay Receiver service. If you want to expose port 5000 in Frigate for local app and API access the port will need to be mapped to another port on the host e.g. 5001
+macOS uses port 5000 for its Airplay Receiver service.  If you want to expose port 5000 in Frigate for local app and API access the port will need to be mapped to another port on the host e.g. 5001

 Failure to remap port 5000 on the host will result in the WebUI and all API endpoints on port 5000 being unreachable, even if port 5000 is exposed correctly in Docker.

 :::

-Docker containers on macOS can be orchestrated by either [Docker Desktop](https://docs.docker.com/desktop/setup/install/mac-install/) or [OrbStack](https://orbstack.dev) (native swift app). The difference in inference speeds is negligable, however CPU, power consumption and container start times will be lower on OrbStack because it is a native Swift application.
+Docker containers on macOS can be orchestrated by either [Docker Desktop](https://docs.docker.com/desktop/setup/install/mac-install/) or [OrbStack](https://orbstack.dev) (native swift app). The difference in inference speeds is negligable, however CPU, power consumption and container start times will be lower on OrbStack because it is a native Swift application. 

 To allow Frigate to use the Apple Silicon Neural Engine / Processing Unit (NPU) the host must be running [Apple Silicon Detector](../configuration/object_detectors.md#apple-silicon-detector) on the host (outside Docker)

 #### Docker Compose example
-
 ```yaml
 services:
  frigate:
@@ -720,7 +719,7 @@ services:
    ports:
      - "8971:8971"
      # If exposing on macOS map to a diffent host port like 5001 or any orher port with no conflicts
-      # - "5001:5000" # Internal unauthenticated access. Expose carefully.
+      # - "5001:5000" # Internal unauthenticated access. Expose carefully. 
      - "8554:8554" # RTSP feeds
    extra_hosts:
      # This is very important
--- a/docs/docs/frigate/updating.md
+++ b/docs/docs/frigate/updating.md
@@ -20,6 +20,7 @@ Keeping Frigate up to date ensures you benefit from the latest features, perform
 If you’re running Frigate via Docker (recommended method), follow these steps:

 1. **Stop the Container**:
+
   - If using Docker Compose:
     ```bash
     docker compose down frigate
@@ -30,8 +31,9 @@ If you’re running Frigate via Docker (recommended method), follow these steps:
     ```

 2. **Update and Pull the Latest Image**:
+
   - If using Docker Compose:
-     - Edit your `docker-compose.yml` file to specify the desired version tag (e.g., `0.17.0` instead of `0.16.4`). For example:
+     - Edit your `docker-compose.yml` file to specify the desired version tag (e.g., `0.17.0` instead of `0.16.3`). For example:
       ```yaml
       services:
         frigate:
@@ -49,6 +51,7 @@ If you’re running Frigate via Docker (recommended method), follow these steps:
       ```

 3. **Start the Container**:
+
   - If using Docker Compose:
     ```bash
     docker compose up -d
@@ -72,15 +75,18 @@ If you’re running Frigate via Docker (recommended method), follow these steps:
 For users running Frigate as a Home Assistant Addon:

 1. **Check for Updates**:
+
   - Navigate to **Settings > Add-ons** in Home Assistant.
   - Find your installed Frigate addon (e.g., "Frigate NVR" or "Frigate NVR (Full Access)").
   - If an update is available, you’ll see an "Update" button.

 2. **Update the Addon**:
+
   - Click the "Update" button next to the Frigate addon.
   - Wait for the process to complete. Home Assistant will handle downloading and installing the new version.

 3. **Restart the Addon**:
+
   - After updating, go to the addon’s page and click "Restart" to apply the changes.

 4. **Verify the Update**:
@@ -99,8 +105,8 @@ If an update causes issues:
 1. Stop Frigate.
 2. Restore your backed-up config file and database.
 3. Revert to the previous image version:
-   - For Docker: Specify an older tag (e.g., `ghcr.io/blakeblackshear/frigate:0.16.4`) in your `docker run` command.
-   - For Docker Compose: Edit your `docker-compose.yml`, specify the older version tag (e.g., `ghcr.io/blakeblackshear/frigate:0.16.4`), and re-run `docker compose up -d`.
+   - For Docker: Specify an older tag (e.g., `ghcr.io/blakeblackshear/frigate:0.16.3`) in your `docker run` command.
+   - For Docker Compose: Edit your `docker-compose.yml`, specify the older version tag (e.g., `ghcr.io/blakeblackshear/frigate:0.16.3`), and re-run `docker compose up -d`.
   - For Home Assistant: Reinstall the previous addon version manually via the repository if needed and restart the addon.
 4. Verify the old version is running again.

--- a/docs/docs/guides/getting_started.md
+++ b/docs/docs/guides/getting_started.md
@@ -119,7 +119,7 @@ services:
    volumes:
      - ./config:/config
      - ./storage:/media/frigate
-      - type: tmpfs # 1GB In-memory filesystem for recording segment storage
+      - type: tmpfs # Optional: 1GB of memory, reduces SSD/SD Card wear
        target: /tmp/cache
        tmpfs:
          size: 1000000000
--- a/docs/docs/integrations/plus.md
+++ b/docs/docs/integrations/plus.md
@@ -54,8 +54,6 @@ Once you have [requested your first model](../plus/first_model.md) and gotten yo
 You can either choose the new model from the Frigate+ pane in the Settings page of the Frigate UI, or manually set the model at the root level in your config:

 ```yaml
-detectors: ...
-
 model:
  path: plus://<your_model_id>
 ```
--- a/docs/docs/plus/first_model.md
+++ b/docs/docs/plus/first_model.md
@@ -24,8 +24,6 @@ You will receive an email notification when your Frigate+ model is ready.
 Models available in Frigate+ can be used with a special model path. No other information needs to be configured because it fetches the remaining config from Frigate+ automatically.

 ```yaml
-detectors: ...
-
 model:
  path: plus://<your_model_id>
 ```
--- a/docs/docs/plus/index.md
+++ b/docs/docs/plus/index.md
@@ -15,15 +15,15 @@ There are three model types offered in Frigate+, `mobiledet`, `yolonas`, and `yo

 Not all model types are supported by all detectors, so it's important to choose a model type to match your detector as shown in the table under [supported detector types](#supported-detector-types). You can test model types for compatibility and speed on your hardware by using the base models.

-| Model Type  | Description                                                                                                                                                    |
-| ----------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| `mobiledet` | Based on the same architecture as the default model included with Frigate. Runs on Google Coral devices and CPUs.                                              |
-| `yolonas`   | A newer architecture that offers slightly higher accuracy and improved detection of small objects. Runs on Intel, NVidia GPUs, and AMD GPUs.                   |
-| `yolov9`    | A leading SOTA (state of the art) object detection model with similar performance to yolonas, but on a wider range of hardware options. Runs on most hardware. |
+| Model Type  | Description                                                                                                                                                                                                                    |
+| ----------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
+| `mobiledet` | Based on the same architecture as the default model included with Frigate. Runs on Google Coral devices and CPUs.                                                                                                              |
+| `yolonas`   | A newer architecture that offers slightly higher accuracy and improved detection of small objects. Runs on Intel, NVidia GPUs, and AMD GPUs.                                                                                   |
+| `yolov9`    | A leading SOTA (state of the art) object detection model with similar performance to yolonas, but on a wider range of hardware options. Runs on Intel, NVidia GPUs, AMD GPUs, Hailo, MemryX, Apple Silicon, and Rockchip NPUs. |

 ### YOLOv9 Details

-YOLOv9 models are available in `s`, `t`, `edgetpu` variants. When requesting a `yolov9` model, you will be prompted to choose a variant. If you want the model to be compatible with a Google Coral, you will need to choose the `edgetpu` variant. If you are unsure what variant to choose, you should perform some tests with the base models to find the performance level that suits you. The `s` size is most similar to the current `yolonas` models in terms of inference times and accuracy, and a good place to start is the `320x320` resolution model for `yolov9s`.
+YOLOv9 models are available in `s` and `t` sizes. When requesting a `yolov9` model, you will be prompted to choose a size. If you are unsure what size to choose, you should perform some tests with the base models to find the performance level that suits you. The `s` size is most similar to the current `yolonas` models in terms of inference times and accuracy, and a good place to start is the `320x320` resolution model for `yolov9s`.

 :::info

@@ -37,21 +37,23 @@ If you have a Hailo device, you will need to specify the hardware you have when

 #### Rockchip (RKNN) Support

-Rockchip models are automatically converted as of 0.17. For 0.16, YOLOv9 onnx models will need to be manually converted. First, you will need to configure Frigate to use the model id for your YOLOv9 onnx model so it downloads the model to your `model_cache` directory. From there, you can follow the [documentation](/configuration/object_detectors.md#converting-your-own-onnx-model-to-rknn-format) to convert it.
+For 0.16, YOLOv9 onnx models will need to be manually converted. First, you will need to configure Frigate to use the model id for your YOLOv9 onnx model so it downloads the model to your `model_cache` directory. From there, you can follow the [documentation](/configuration/object_detectors.md#converting-your-own-onnx-model-to-rknn-format) to convert it. Automatic conversion is available in 0.17 and later.

 ## Supported detector types

-Currently, Frigate+ models support CPU (`cpu`), Google Coral (`edgetpu`), OpenVino (`openvino`), ONNX (`onnx`), Hailo (`hailo8l`), and Rockchip (`rknn`) detectors.
+Currently, Frigate+ models support CPU (`cpu`), Google Coral (`edgetpu`), OpenVino (`openvino`), ONNX (`onnx`), Hailo (`hailo8l`), and Rockchip\* (`rknn`) detectors.

 | Hardware                                                                         | Recommended Detector Type | Recommended Model Type |
 | -------------------------------------------------------------------------------- | ------------------------- | ---------------------- |
 | [CPU](/configuration/object_detectors.md#cpu-detector-not-recommended)           | `cpu`                     | `mobiledet`            |
-| [Coral (all form factors)](/configuration/object_detectors.md#edge-tpu-detector) | `edgetpu`                 | `yolov9`               |
+| [Coral (all form factors)](/configuration/object_detectors.md#edge-tpu-detector) | `edgetpu`                 | `mobiledet`            |
 | [Intel](/configuration/object_detectors.md#openvino-detector)                    | `openvino`                | `yolov9`               |
 | [NVidia GPU](/configuration/object_detectors#onnx)                               | `onnx`                    | `yolov9`               |
 | [AMD ROCm GPU](/configuration/object_detectors#amdrocm-gpu-detector)             | `onnx`                    | `yolov9`               |
 | [Hailo8/Hailo8L/Hailo8R](/configuration/object_detectors#hailo-8)                | `hailo8l`                 | `yolov9`               |
-| [Rockchip NPU](/configuration/object_detectors#rockchip-platform)                | `rknn`                    | `yolov9`               |
+| [Rockchip NPU](/configuration/object_detectors#rockchip-platform)\*              | `rknn`                    | `yolov9`               |
+
+_\* Requires manual conversion in 0.16. Automatic conversion available in 0.17 and later._

 ## Improving your model

@@ -79,7 +81,7 @@ Candidate labels are also available for annotation. These labels don't have enou

 Where possible, these labels are mapped to existing labels during training. For example, any `baby` labels are mapped to `person` until support for new labels is added.

-The candidate labels are: `baby`, `bpost`, `badger`, `possum`, `rodent`, `chicken`, `groundhog`, `boar`, `hedgehog`, `tractor`, `golf cart`, `garbage truck`, `bus`, `sports ball`, `la_poste`, `lawnmower`, `heron`, `rickshaw`, `wombat`, `auspost`, `aramex`, `bobcat`, `mustelid`, `transoflex`, `airplane`, `drone`, `mountain_lion`, `crocodile`, `turkey`, `baby_stroller`, `monkey`, `coyote`, `porcupine`, `parcelforce`, `sheep`, `snake`, `helicopter`, `lizard`, `duck`, `hermes`, `cargus`, `fan_courier`, `sameday`
+The candidate labels are: `baby`, `bpost`, `badger`, `possum`, `rodent`, `chicken`, `groundhog`, `boar`, `hedgehog`, `tractor`, `golf cart`, `garbage truck`, `bus`, `sports ball`

 Candidate labels are not available for automatic suggestions.

--- a/frigate/api/app.py
+++ b/frigate/api/app.py
@@ -38,6 +38,7 @@ from frigate.config.camera.updater import (
    CameraConfigUpdateTopic,
 )
 from frigate.ffmpeg_presets import FFMPEG_HWACCEL_VAAPI, _gpu_selector
+from frigate.genai import GenAIClientManager
 from frigate.jobs.media_sync import (
    get_current_media_sync_job,
    get_media_sync_job_by_id,
@@ -432,6 +433,7 @@ def config_set(request: Request, body: AppConfigSetBody):
    if body.requires_restart == 0 or body.update_topic:
        old_config: FrigateConfig = request.app.frigate_config
        request.app.frigate_config = config
+        request.app.genai_manager = GenAIClientManager(config)

        if body.update_topic:
            if body.update_topic.startswith("config/cameras/"):
--- a/frigate/api/chat.py
+++ b/frigate/api/chat.py
@@ -3,12 +3,13 @@
 import base64
 import json
 import logging
-from datetime import datetime, timezone
-from typing import Any, Dict, List, Optional
+import time
+from datetime import datetime
+from typing import Any, Dict, Generator, List, Optional

 import cv2
 from fastapi import APIRouter, Body, Depends, Request
-from fastapi.responses import JSONResponse
+from fastapi.responses import JSONResponse, StreamingResponse
 from pydantic import BaseModel

 from frigate.api.auth import (
@@ -20,16 +21,60 @@ from frigate.api.defs.request.chat_body import ChatCompletionRequest
 from frigate.api.defs.response.chat_response import (
    ChatCompletionResponse,
    ChatMessageResponse,
+    ToolCall,
 )
 from frigate.api.defs.tags import Tags
 from frigate.api.event import events
-from frigate.genai import get_genai_client
+from frigate.genai.utils import build_assistant_message_for_conversation

 logger = logging.getLogger(__name__)

 router = APIRouter(tags=[Tags.chat])


+def _chunk_content(content: str, chunk_size: int = 80) -> Generator[str, None, None]:
+    """Yield content in word-aware chunks for streaming."""
+    if not content:
+        return
+    words = content.split(" ")
+    current: List[str] = []
+    current_len = 0
+    for w in words:
+        current.append(w)
+        current_len += len(w) + 1
+        if current_len >= chunk_size:
+            yield " ".join(current) + " "
+            current = []
+            current_len = 0
+    if current:
+        yield " ".join(current)
+
+
+def _format_events_with_local_time(
+    events_list: List[Dict[str, Any]],
+) -> List[Dict[str, Any]]:
+    """Add human-readable local start/end times to each event for the LLM."""
+    result = []
+    for evt in events_list:
+        if not isinstance(evt, dict):
+            result.append(evt)
+            continue
+        copy_evt = dict(evt)
+        try:
+            start_ts = evt.get("start_time")
+            end_ts = evt.get("end_time")
+            if start_ts is not None:
+                dt_start = datetime.fromtimestamp(start_ts)
+                copy_evt["start_time_local"] = dt_start.strftime("%Y-%m-%d %I:%M:%S %p")
+            if end_ts is not None:
+                dt_end = datetime.fromtimestamp(end_ts)
+                copy_evt["end_time_local"] = dt_end.strftime("%Y-%m-%d %I:%M:%S %p")
+        except (TypeError, ValueError, OSError):
+            pass
+        result.append(copy_evt)
+    return result
+
+
 class ToolExecuteRequest(BaseModel):
    """Request model for tool execution."""

@@ -53,19 +98,25 @@ def get_tool_definitions() -> List[Dict[str, Any]]:
                    "Search for detected objects in Frigate by camera, object label, time range, "
                    "zones, and other filters. Use this to answer questions about when "
                    "objects were detected, what objects appeared, or to find specific object detections. "
-                    "An 'object' in Frigate represents a tracked detection (e.g., a person, package, car)."
+                    "An 'object' in Frigate represents a tracked detection (e.g., a person, package, car). "
+                    "When the user asks about a specific name (person, delivery company, animal, etc.), "
+                    "filter by sub_label only and do not set label."
                ),
                "parameters": {
                    "type": "object",
                    "properties": {
                        "camera": {
                            "type": "string",
-                            "description": "Camera name to filter by (optional). Use 'all' for all cameras.",
+                            "description": "Camera name to filter by (optional).",
                        },
                        "label": {
                            "type": "string",
                            "description": "Object label to filter by (e.g., 'person', 'package', 'car').",
                        },
+                        "sub_label": {
+                            "type": "string",
+                            "description": "Name of a person, delivery company, animal, etc. When filtering by a specific name, use only sub_label; do not set label.",
+                        },
                        "after": {
                            "type": "string",
                            "description": "Start time in ISO 8601 format (e.g., '2024-01-01T00:00:00Z').",
@@ -81,8 +132,8 @@ def get_tool_definitions() -> List[Dict[str, Any]]:
                        },
                        "limit": {
                            "type": "integer",
-                            "description": "Maximum number of objects to return (default: 10).",
-                            "default": 10,
+                            "description": "Maximum number of objects to return (default: 25).",
+                            "default": 25,
                        },
                    },
                },
@@ -120,14 +171,13 @@ def get_tool_definitions() -> List[Dict[str, Any]]:
    summary="Get available tools",
    description="Returns OpenAI-compatible tool definitions for function calling.",
 )
-def get_tools(request: Request) -> JSONResponse:
+def get_tools() -> JSONResponse:
    """Get list of available tools for LLM function calling."""
    tools = get_tool_definitions()
    return JSONResponse(content={"tools": tools})


 async def _execute_search_objects(
-    request: Request,
    arguments: Dict[str, Any],
    allowed_cameras: List[str],
 ) -> JSONResponse:
@@ -137,23 +187,26 @@ async def _execute_search_objects(
    This searches for detected objects (events) in Frigate using the same
    logic as the events API endpoint.
    """
-    # Parse ISO 8601 timestamps to Unix timestamps if provided
+    # Parse after/before as server local time; convert to Unix timestamp
    after = arguments.get("after")
    before = arguments.get("before")

+    def _parse_as_local_timestamp(s: str):
+        s = s.replace("Z", "").strip()[:19]
+        dt = datetime.strptime(s, "%Y-%m-%dT%H:%M:%S")
+        return time.mktime(dt.timetuple())
+
    if after:
        try:
-            after_dt = datetime.fromisoformat(after.replace("Z", "+00:00"))
-            after = after_dt.timestamp()
-        except (ValueError, AttributeError):
+            after = _parse_as_local_timestamp(after)
+        except (ValueError, AttributeError, TypeError):
            logger.warning(f"Invalid 'after' timestamp format: {after}")
            after = None

    if before:
        try:
-            before_dt = datetime.fromisoformat(before.replace("Z", "+00:00"))
-            before = before_dt.timestamp()
-        except (ValueError, AttributeError):
+            before = _parse_as_local_timestamp(before)
+        except (ValueError, AttributeError, TypeError):
            logger.warning(f"Invalid 'before' timestamp format: {before}")
            before = None

@@ -166,15 +219,14 @@ async def _execute_search_objects(

    # Build query parameters compatible with EventsQueryParams
    query_params = EventsQueryParams(
-        camera=arguments.get("camera", "all"),
        cameras=arguments.get("camera", "all"),
-        label=arguments.get("label", "all"),
        labels=arguments.get("label", "all"),
+        sub_labels=arguments.get("sub_label", "all").lower(),
        zones=zones,
        zone=zones,
        after=after,
        before=before,
-        limit=arguments.get("limit", 10),
+        limit=arguments.get("limit", 25),
    )

    try:
@@ -190,7 +242,7 @@ async def _execute_search_objects(
        return JSONResponse(
            content={
                "success": False,
-                "message": "Error searching objects",
+                "message": f"Error searching objects: {str(e)}",
            },
            status_code=500,
        )
@@ -203,7 +255,6 @@ async def _execute_search_objects(
    description="Execute a tool function call from an LLM.",
 )
 async def execute_tool(
-    request: Request,
    body: ToolExecuteRequest = Body(...),
    allowed_cameras: List[str] = Depends(get_allowed_cameras_for_filter),
 ) -> JSONResponse:
@@ -219,7 +270,7 @@ async def execute_tool(
    logger.debug(f"Executing tool: {tool_name} with arguments: {arguments}")

    if tool_name == "search_objects":
-        return await _execute_search_objects(request, arguments, allowed_cameras)
+        return await _execute_search_objects(arguments, allowed_cameras)

    return JSONResponse(
        content={
@@ -279,7 +330,7 @@ async def _execute_get_live_context(
    except Exception as e:
        logger.error(f"Error executing get_live_context: {e}", exc_info=True)
        return {
-            "error": "Error getting live context",
+            "error": f"Error getting live context: {str(e)}",
        }


@@ -335,7 +386,7 @@ async def _execute_tool_internal(
    This is used by the chat completion endpoint to execute tools.
    """
    if tool_name == "search_objects":
-        response = await _execute_search_objects(request, arguments, allowed_cameras)
+        response = await _execute_search_objects(arguments, allowed_cameras)
        try:
            if hasattr(response, "body"):
                body_str = response.body.decode("utf-8")
@@ -350,15 +401,109 @@ async def _execute_tool_internal(
    elif tool_name == "get_live_context":
        camera = arguments.get("camera")
        if not camera:
+            logger.error(
+                "Tool get_live_context failed: camera parameter is required. "
+                "Arguments: %s",
+                json.dumps(arguments),
+            )
            return {"error": "Camera parameter is required"}
        return await _execute_get_live_context(request, camera, allowed_cameras)
    else:
+        logger.error(
+            "Tool call failed: unknown tool %r. Expected one of: search_objects, get_live_context. "
+            "Arguments received: %s",
+            tool_name,
+            json.dumps(arguments),
+        )
        return {"error": f"Unknown tool: {tool_name}"}


+async def _execute_pending_tools(
+    pending_tool_calls: List[Dict[str, Any]],
+    request: Request,
+    allowed_cameras: List[str],
+) -> tuple[List[ToolCall], List[Dict[str, Any]]]:
+    """
+    Execute a list of tool calls; return (ToolCall list for API response, tool result dicts for conversation).
+    """
+    tool_calls_out: List[ToolCall] = []
+    tool_results: List[Dict[str, Any]] = []
+    for tool_call in pending_tool_calls:
+        tool_name = tool_call["name"]
+        tool_args = tool_call.get("arguments") or {}
+        tool_call_id = tool_call["id"]
+        logger.debug(
+            f"Executing tool: {tool_name} (id: {tool_call_id}) with arguments: {json.dumps(tool_args, indent=2)}"
+        )
+        try:
+            tool_result = await _execute_tool_internal(
+                tool_name, tool_args, request, allowed_cameras
+            )
+            if isinstance(tool_result, dict) and tool_result.get("error"):
+                logger.error(
+                    "Tool call %s (id: %s) returned error: %s. Arguments: %s",
+                    tool_name,
+                    tool_call_id,
+                    tool_result.get("error"),
+                    json.dumps(tool_args),
+                )
+            if tool_name == "search_objects" and isinstance(tool_result, list):
+                tool_result = _format_events_with_local_time(tool_result)
+                _keys = {
+                    "id",
+                    "camera",
+                    "label",
+                    "zones",
+                    "start_time_local",
+                    "end_time_local",
+                    "sub_label",
+                    "event_count",
+                }
+                tool_result = [
+                    {k: evt[k] for k in _keys if k in evt}
+                    for evt in tool_result
+                    if isinstance(evt, dict)
+                ]
+            result_content = (
+                json.dumps(tool_result)
+                if isinstance(tool_result, (dict, list))
+                else (tool_result if isinstance(tool_result, str) else str(tool_result))
+            )
+            tool_calls_out.append(
+                ToolCall(name=tool_name, arguments=tool_args, response=result_content)
+            )
+            tool_results.append(
+                {
+                    "role": "tool",
+                    "tool_call_id": tool_call_id,
+                    "content": result_content,
+                }
+            )
+        except Exception as e:
+            logger.error(
+                "Error executing tool %s (id: %s): %s. Arguments: %s",
+                tool_name,
+                tool_call_id,
+                e,
+                json.dumps(tool_args),
+                exc_info=True,
+            )
+            error_content = json.dumps({"error": f"Tool execution failed: {str(e)}"})
+            tool_calls_out.append(
+                ToolCall(name=tool_name, arguments=tool_args, response=error_content)
+            )
+            tool_results.append(
+                {
+                    "role": "tool",
+                    "tool_call_id": tool_call_id,
+                    "content": error_content,
+                }
+            )
+    return (tool_calls_out, tool_results)
+
+
@router.post(
    "/chat/completion",
-    response_model=ChatCompletionResponse,
    dependencies=[Depends(allow_any_authenticated())],
    summary="Chat completion with tool calling",
    description=(
@@ -370,7 +515,7 @@ async def chat_completion(
    request: Request,
    body: ChatCompletionRequest = Body(...),
    allowed_cameras: List[str] = Depends(get_allowed_cameras_for_filter),
-) -> JSONResponse:
+):
    """
    Chat completion endpoint with tool calling support.

@@ -383,7 +528,7 @@ async def chat_completion(
    6. Repeats until final answer
    7. Returns response to user
    """
-    genai_client = get_genai_client(request.app.frigate_config)
+    genai_client = request.app.genai_manager.tool_client
    if not genai_client:
        return JSONResponse(
            content={
@@ -395,9 +540,9 @@ async def chat_completion(
    tools = get_tool_definitions()
    conversation = []

-    current_datetime = datetime.now(timezone.utc)
+    current_datetime = datetime.now()
    current_date_str = current_datetime.strftime("%Y-%m-%d")
-    current_time_str = current_datetime.strftime("%H:%M:%S %Z")
+    current_time_str = current_datetime.strftime("%I:%M:%S %p")

    cameras_info = []
    config = request.app.frigate_config
@@ -430,9 +575,12 @@ async def chat_completion(

    system_prompt = f"""You are a helpful assistant for Frigate, a security camera NVR system. You help users answer questions about their cameras, detected objects, and events.

-Current date and time: {current_date_str} at {current_time_str} (UTC)
+Current server local date and time: {current_date_str} at {current_time_str}

-When users ask questions about "today", "yesterday", "this week", etc., use the current date above as reference.
+Do not start your response with phrases like "I will check...", "Let me see...", or "Let me look...". Answer directly.
+
+Always present times to the user in the server's local timezone. When tool results include start_time_local and end_time_local, use those exact strings when listing or describing detection times—do not convert or invent timestamps. Do not use UTC or ISO format with Z for the user-facing answer unless the tool result only provides Unix timestamps without local time fields.
+When users ask about "today", "yesterday", "this week", etc., use the current date above as reference.
 When searching for objects or events, use ISO 8601 format for dates (e.g., {current_date_str}T00:00:00Z for the start of today).
 Always be accurate with time calculations based on the current date provided.{cameras_section}{live_image_note}"""

@@ -472,6 +620,7 @@ Always be accurate with time calculations based on the current date provided.{ca
        conversation.append(msg_dict)

    tool_iterations = 0
+    tool_calls: List[ToolCall] = []
    max_iterations = body.max_tool_iterations

    logger.debug(
@@ -479,6 +628,81 @@ Always be accurate with time calculations based on the current date provided.{ca
        f"{len(tools)} tool(s) available, max_iterations={max_iterations}"
    )

+    # True LLM streaming when client supports it and stream requested
+    if body.stream and hasattr(genai_client, "chat_with_tools_stream"):
+        stream_tool_calls: List[ToolCall] = []
+        stream_iterations = 0
+
+        async def stream_body_llm():
+            nonlocal conversation, stream_tool_calls, stream_iterations
+            while stream_iterations < max_iterations:
+                logger.debug(
+                    f"Streaming LLM (iteration {stream_iterations + 1}/{max_iterations}) "
+                    f"with {len(conversation)} message(s)"
+                )
+                async for event in genai_client.chat_with_tools_stream(
+                    messages=conversation,
+                    tools=tools if tools else None,
+                    tool_choice="auto",
+                ):
+                    kind, value = event
+                    if kind == "content_delta":
+                        yield (
+                            json.dumps({"type": "content", "delta": value}).encode(
+                                "utf-8"
+                            )
+                            + b"\n"
+                        )
+                    elif kind == "message":
+                        msg = value
+                        if msg.get("finish_reason") == "error":
+                            yield (
+                                json.dumps(
+                                    {
+                                        "type": "error",
+                                        "error": "An error occurred while processing your request.",
+                                    }
+                                ).encode("utf-8")
+                                + b"\n"
+                            )
+                            return
+                        pending = msg.get("tool_calls")
+                        if pending:
+                            stream_iterations += 1
+                            conversation.append(
+                                build_assistant_message_for_conversation(
+                                    msg.get("content"), pending
+                                )
+                            )
+                            executed_calls, tool_results = await _execute_pending_tools(
+                                pending, request, allowed_cameras
+                            )
+                            stream_tool_calls.extend(executed_calls)
+                            conversation.extend(tool_results)
+                            yield (
+                                json.dumps(
+                                    {
+                                        "type": "tool_calls",
+                                        "tool_calls": [
+                                            tc.model_dump() for tc in stream_tool_calls
+                                        ],
+                                    }
+                                ).encode("utf-8")
+                                + b"\n"
+                            )
+                            break
+                        else:
+                            yield (json.dumps({"type": "done"}).encode("utf-8") + b"\n")
+                            return
+            else:
+                yield json.dumps({"type": "done"}).encode("utf-8") + b"\n"
+
+        return StreamingResponse(
+            stream_body_llm(),
+            media_type="application/x-ndjson",
+            headers={"X-Accel-Buffering": "no"},
+        )
+
    try:
        while tool_iterations < max_iterations:
            logger.debug(
@@ -500,117 +724,71 @@ Always be accurate with time calculations based on the current date provided.{ca
                    status_code=500,
                )

-            assistant_message = {
-                "role": "assistant",
-                "content": response.get("content"),
-            }
-            if response.get("tool_calls"):
-                assistant_message["tool_calls"] = [
-                    {
-                        "id": tc["id"],
-                        "type": "function",
-                        "function": {
-                            "name": tc["name"],
-                            "arguments": json.dumps(tc["arguments"]),
-                        },
-                    }
-                    for tc in response["tool_calls"]
-                ]
-            conversation.append(assistant_message)
+            conversation.append(
+                build_assistant_message_for_conversation(
+                    response.get("content"), response.get("tool_calls")
+                )
+            )

-            tool_calls = response.get("tool_calls")
-            if not tool_calls:
+            pending_tool_calls = response.get("tool_calls")
+            if not pending_tool_calls:
                logger.debug(
                    f"Chat completion finished with final answer (iterations: {tool_iterations})"
                )
+                final_content = response.get("content") or ""
+
+                if body.stream:
+
+                    async def stream_body() -> Any:
+                        if tool_calls:
+                            yield (
+                                json.dumps(
+                                    {
+                                        "type": "tool_calls",
+                                        "tool_calls": [
+                                            tc.model_dump() for tc in tool_calls
+                                        ],
+                                    }
+                                ).encode("utf-8")
+                                + b"\n"
+                            )
+                        # Stream content in word-sized chunks for smooth UX
+                        for part in _chunk_content(final_content):
+                            yield (
+                                json.dumps({"type": "content", "delta": part}).encode(
+                                    "utf-8"
+                                )
+                                + b"\n"
+                            )
+                        yield json.dumps({"type": "done"}).encode("utf-8") + b"\n"
+
+                    return StreamingResponse(
+                        stream_body(),
+                        media_type="application/x-ndjson",
+                    )
+
                return JSONResponse(
                    content=ChatCompletionResponse(
                        message=ChatMessageResponse(
                            role="assistant",
-                            content=response.get("content"),
+                            content=final_content,
                            tool_calls=None,
                        ),
                        finish_reason=response.get("finish_reason", "stop"),
                        tool_iterations=tool_iterations,
+                        tool_calls=tool_calls,
                    ).model_dump(),
                )

-            # Execute tools
            tool_iterations += 1
            logger.debug(
                f"Tool calls detected (iteration {tool_iterations}/{max_iterations}): "
-                f"{len(tool_calls)} tool(s) to execute"
+                f"{len(pending_tool_calls)} tool(s) to execute"
            )
-            tool_results = []
-
-            for tool_call in tool_calls:
-                tool_name = tool_call["name"]
-                tool_args = tool_call["arguments"]
-                tool_call_id = tool_call["id"]
-
-                logger.debug(
-                    f"Executing tool: {tool_name} (id: {tool_call_id}) with arguments: {json.dumps(tool_args, indent=2)}"
-                )
-
-                try:
-                    tool_result = await _execute_tool_internal(
-                        tool_name, tool_args, request, allowed_cameras
-                    )
-
-                    if isinstance(tool_result, dict):
-                        result_content = json.dumps(tool_result)
-                        result_summary = tool_result
-                        if isinstance(tool_result, dict) and isinstance(
-                            tool_result.get("content"), list
-                        ):
-                            result_count = len(tool_result.get("content", []))
-                            result_summary = {
-                                "count": result_count,
-                                "sample": tool_result.get("content", [])[:2]
-                                if result_count > 0
-                                else [],
-                            }
-                        logger.debug(
-                            f"Tool {tool_name} (id: {tool_call_id}) completed successfully. "
-                            f"Result: {json.dumps(result_summary, indent=2)}"
-                        )
-                    elif isinstance(tool_result, str):
-                        result_content = tool_result
-                        logger.debug(
-                            f"Tool {tool_name} (id: {tool_call_id}) completed successfully. "
-                            f"Result length: {len(result_content)} characters"
-                        )
-                    else:
-                        result_content = str(tool_result)
-                        logger.debug(
-                            f"Tool {tool_name} (id: {tool_call_id}) completed successfully. "
-                            f"Result type: {type(tool_result).__name__}"
-                        )
-
-                    tool_results.append(
-                        {
-                            "role": "tool",
-                            "tool_call_id": tool_call_id,
-                            "content": result_content,
-                        }
-                    )
-                except Exception as e:
-                    logger.error(
-                        f"Error executing tool {tool_name} (id: {tool_call_id}): {e}",
-                        exc_info=True,
-                    )
-                    error_content = json.dumps({"error": "Tool execution failed"})
-                    tool_results.append(
-                        {
-                            "role": "tool",
-                            "tool_call_id": tool_call_id,
-                            "content": error_content,
-                        }
-                    )
-                    logger.debug(
-                        f"Tool {tool_name} (id: {tool_call_id}) failed. Error result added to conversation."
-                    )
-
+            executed_calls, tool_results = await _execute_pending_tools(
+                pending_tool_calls, request, allowed_cameras
+            )
+            tool_calls.extend(executed_calls)
            conversation.extend(tool_results)
            logger.debug(
                f"Added {len(tool_results)} tool result(s) to conversation. "
@@ -629,6 +807,7 @@ Always be accurate with time calculations based on the current date provided.{ca
                ),
                finish_reason="length",
                tool_iterations=tool_iterations,
+                tool_calls=tool_calls,
            ).model_dump(),
        )

--- a/frigate/api/defs/request/chat_body.py
+++ b/frigate/api/defs/request/chat_body.py
@@ -39,3 +39,7 @@ class ChatCompletionRequest(BaseModel):
            "user message as multimodal content. Use with get_live_context for detection info."
        ),
    )
+    stream: bool = Field(
+        default=False,
+        description="If true, stream the final assistant response in the body as newline-delimited JSON.",
+    )
--- a/frigate/api/defs/response/chat_response.py
+++ b/frigate/api/defs/response/chat_response.py
@@ -5,8 +5,8 @@ from typing import Any, Optional
 from pydantic import BaseModel, Field


-class ToolCall(BaseModel):
-    """A tool call from the LLM."""
+class ToolCallInvocation(BaseModel):
+    """A tool call requested by the LLM (before execution)."""

    id: str = Field(description="Unique identifier for this tool call")
    name: str = Field(description="Tool name to call")
@@ -20,11 +20,24 @@ class ChatMessageResponse(BaseModel):
    content: Optional[str] = Field(
        default=None, description="Message content (None if tool calls present)"
    )
-    tool_calls: Optional[list[ToolCall]] = Field(
+    tool_calls: Optional[list[ToolCallInvocation]] = Field(
        default=None, description="Tool calls if LLM wants to call tools"
    )


+class ToolCall(BaseModel):
+    """A tool that was executed during the completion, with its response."""
+
+    name: str = Field(description="Tool name that was called")
+    arguments: dict[str, Any] = Field(
+        default_factory=dict, description="Arguments passed to the tool"
+    )
+    response: str = Field(
+        default="",
+        description="The response or result returned from the tool execution",
+    )
+
+
 class ChatCompletionResponse(BaseModel):
    """Response from chat completion."""

@@ -35,3 +48,7 @@ class ChatCompletionResponse(BaseModel):
    tool_iterations: int = Field(
        default=0, description="Number of tool call iterations performed"
    )
+    tool_calls: list[ToolCall] = Field(
+        default_factory=list,
+        description="List of tool calls that were executed during this completion",
+    )
--- a/frigate/api/fastapi_app.py
+++ b/frigate/api/fastapi_app.py
@@ -33,6 +33,7 @@ from frigate.comms.event_metadata_updater import (
 from frigate.config import FrigateConfig
 from frigate.config.camera.updater import CameraConfigUpdatePublisher
 from frigate.embeddings import EmbeddingsContext
+from frigate.genai import GenAIClientManager
 from frigate.ptz.onvif import OnvifController
 from frigate.stats.emitter import StatsEmitter
 from frigate.storage import StorageMaintainer
@@ -134,6 +135,7 @@ def create_fastapi_app(
    app.include_router(record.router)
    # App Properties
    app.frigate_config = frigate_config
+    app.genai_manager = GenAIClientManager(frigate_config)
    app.embeddings = embeddings
    app.detected_frames_processor = detected_frames_processor
    app.storage_maintainer = storage_maintainer
--- a/frigate/api/review.py
+++ b/frigate/api/review.py
@@ -33,7 +33,6 @@ from frigate.api.defs.response.review_response import (
    ReviewSummaryResponse,
 )
 from frigate.api.defs.tags import Tags
-from frigate.config import FrigateConfig
 from frigate.embeddings import EmbeddingsContext
 from frigate.models import Recordings, ReviewSegment, UserReviewStatus
 from frigate.review.types import SeverityEnum
@@ -747,9 +746,7 @@ async def set_not_reviewed(
    description="Use GenAI to summarize review items over a period of time.",
 )
 def generate_review_summary(request: Request, start_ts: float, end_ts: float):
-    config: FrigateConfig = request.app.frigate_config
-
-    if not config.genai.provider:
+    if not request.app.genai_manager.vision_client:
        return JSONResponse(
            content=(
                {
--- a/frigate/config/camera/genai.py
+++ b/frigate/config/camera/genai.py
@@ -6,7 +6,7 @@ from pydantic import Field
 from ..base import FrigateBaseModel
 from ..env import EnvString

-__all__ = ["GenAIConfig", "GenAIProviderEnum"]
+__all__ = ["GenAIConfig", "GenAIProviderEnum", "GenAIRoleEnum"]


 class GenAIProviderEnum(str, Enum):
@@ -17,6 +17,12 @@ class GenAIProviderEnum(str, Enum):
    llamacpp = "llamacpp"


+class GenAIRoleEnum(str, Enum):
+    tools = "tools"
+    vision = "vision"
+    embeddings = "embeddings"
+
+
 class GenAIConfig(FrigateBaseModel):
    """Primary GenAI Config to define GenAI Provider."""

@@ -24,6 +30,14 @@ class GenAIConfig(FrigateBaseModel):
    base_url: Optional[str] = Field(default=None, title="Provider base url.")
    model: str = Field(default="gpt-4o", title="GenAI model.")
    provider: GenAIProviderEnum | None = Field(default=None, title="GenAI provider.")
+    roles: list[GenAIRoleEnum] = Field(
+        default_factory=lambda: [
+            GenAIRoleEnum.embeddings,
+            GenAIRoleEnum.vision,
+            GenAIRoleEnum.tools,
+        ],
+        title="GenAI roles (tools, vision, embeddings); one provider per role.",
+    )
    provider_options: dict[str, Any] = Field(
        default={}, title="GenAI Provider extra options."
    )
--- a/frigate/config/classification.py
+++ b/frigate/config/classification.py
@@ -1,5 +1,5 @@
 from enum import Enum
-from typing import Dict, List, Optional
+from typing import Dict, List, Optional, Union

 from pydantic import ConfigDict, Field

@@ -128,9 +128,10 @@ class SemanticSearchConfig(FrigateBaseModel):
    reindex: Optional[bool] = Field(
        default=False, title="Reindex all tracked objects on startup."
    )
-    model: Optional[SemanticSearchModelEnum] = Field(
+    model: Optional[Union[SemanticSearchModelEnum, str]] = Field(
        default=SemanticSearchModelEnum.jinav1,
-        title="The CLIP model to use for semantic search.",
+        title="The CLIP model or GenAI provider name for semantic search.",
+        description="Use 'jinav1', 'jinav2' for ONNX models, or a GenAI config key (e.g. 'default') when that provider has the embeddings role.",
    )
    model_size: str = Field(
        default="small", title="The size of the embeddings model used."
--- a/frigate/config/config.py
+++ b/frigate/config/config.py
@@ -45,7 +45,7 @@ from .camera.audio import AudioConfig
 from .camera.birdseye import BirdseyeConfig
 from .camera.detect import DetectConfig
 from .camera.ffmpeg import FfmpegConfig
-from .camera.genai import GenAIConfig
+from .camera.genai import GenAIConfig, GenAIRoleEnum
 from .camera.motion import MotionConfig
 from .camera.notification import NotificationConfig
 from .camera.objects import FilterConfig, ObjectConfig
@@ -347,9 +347,9 @@ class FrigateConfig(FrigateBaseModel):
        default_factory=ModelConfig, title="Detection model configuration."
    )

-    # GenAI config
-    genai: GenAIConfig = Field(
-        default_factory=GenAIConfig, title="Generative AI configuration."
+    # GenAI config (named provider configs: name -> GenAIConfig)
+    genai: Dict[str, GenAIConfig] = Field(
+        default_factory=dict, title="Generative AI configuration (named providers)."
    )

    # Camera config
@@ -431,6 +431,34 @@ class FrigateConfig(FrigateBaseModel):
        # set notifications state
        self.notifications.enabled_in_config = self.notifications.enabled

+        # validate genai: each role (tools, vision, embeddings) at most once
+        role_to_name: dict[GenAIRoleEnum, str] = {}
+        for name, genai_cfg in self.genai.items():
+            for role in genai_cfg.roles:
+                if role in role_to_name:
+                    raise ValueError(
+                        f"GenAI role '{role.value}' is assigned to both "
+                        f"'{role_to_name[role]}' and '{name}'; each role must have "
+                        "exactly one provider."
+                    )
+                role_to_name[role] = name
+
+        # validate semantic_search.model when it is a GenAI provider name
+        if self.semantic_search.enabled and isinstance(
+            self.semantic_search.model, str
+        ):
+            if self.semantic_search.model not in self.genai:
+                raise ValueError(
+                    f"semantic_search.model '{self.semantic_search.model}' is not a "
+                    "valid GenAI config key. Must match a key in genai config."
+                )
+            genai_cfg = self.genai[self.semantic_search.model]
+            if GenAIRoleEnum.embeddings not in genai_cfg.roles:
+                raise ValueError(
+                    f"GenAI provider '{self.semantic_search.model}' must have "
+                    "'embeddings' in its roles for semantic search."
+                )
+
        # set default min_score for object attributes
        for attribute in self.model.all_attributes:
            if not self.objects.filters.get(attribute):
--- a/frigate/data_processing/real_time/bird.py
+++ b/frigate/data_processing/real_time/bird.py
@@ -22,7 +22,7 @@ from .api import RealTimeProcessorApi
 try:
    from tflite_runtime.interpreter import Interpreter
 except ModuleNotFoundError:
-    from ai_edge_litert.interpreter import Interpreter
+    from tensorflow.lite.python.interpreter import Interpreter

 logger = logging.getLogger(__name__)

--- a/frigate/data_processing/real_time/custom_classification.py
+++ b/frigate/data_processing/real_time/custom_classification.py
@@ -32,7 +32,7 @@ from .api import RealTimeProcessorApi
 try:
    from tflite_runtime.interpreter import Interpreter
 except ModuleNotFoundError:
-    from ai_edge_litert.interpreter import Interpreter
+    from tensorflow.lite.python.interpreter import Interpreter

 logger = logging.getLogger(__name__)

@@ -76,7 +76,7 @@ class CustomStateClassificationProcessor(RealTimeProcessorApi):
        try:
            from tflite_runtime.interpreter import Interpreter
        except ModuleNotFoundError:
-            from ai_edge_litert.interpreter import Interpreter
+            from tensorflow.lite.python.interpreter import Interpreter

        model_path = os.path.join(self.model_dir, "model.tflite")
        labelmap_path = os.path.join(self.model_dir, "labelmap.txt")
--- a/frigate/detectors/detector_utils.py
+++ b/frigate/detectors/detector_utils.py
@@ -6,7 +6,7 @@ import numpy as np
 try:
    from tflite_runtime.interpreter import Interpreter, load_delegate
 except ModuleNotFoundError:
-    from ai_edge_litert.interpreter import Interpreter, load_delegate
+    from tensorflow.lite.python.interpreter import Interpreter, load_delegate


 logger = logging.getLogger(__name__)
--- a/frigate/detectors/plugins/cpu_tfl.py
+++ b/frigate/detectors/plugins/cpu_tfl.py
@@ -12,7 +12,7 @@ from ..detector_utils import tflite_detect_raw, tflite_init
 try:
    from tflite_runtime.interpreter import Interpreter
 except ModuleNotFoundError:
-    from ai_edge_litert.interpreter import Interpreter
+    from tensorflow.lite.python.interpreter import Interpreter


 logger = logging.getLogger(__name__)
--- a/frigate/detectors/plugins/edgetpu_tfl.py
+++ b/frigate/detectors/plugins/edgetpu_tfl.py
@@ -13,7 +13,7 @@ from frigate.detectors.detector_config import BaseDetectorConfig, ModelTypeEnum
 try:
    from tflite_runtime.interpreter import Interpreter, load_delegate
 except ModuleNotFoundError:
-    from ai_edge_litert.interpreter import Interpreter, load_delegate
+    from tensorflow.lite.python.interpreter import Interpreter, load_delegate

 logger = logging.getLogger(__name__)

--- a/frigate/embeddings/embeddings.py
+++ b/frigate/embeddings/embeddings.py
@@ -28,6 +28,7 @@ from frigate.types import ModelStatusTypesEnum
 from frigate.util.builtin import EventsPerSecond, InferenceSpeed, serialize
 from frigate.util.file import get_event_thumbnail_bytes

+from .genai_embedding import GenAIEmbedding
 from .onnx.jina_v1_embedding import JinaV1ImageEmbedding, JinaV1TextEmbedding
 from .onnx.jina_v2_embedding import JinaV2Embedding

@@ -73,11 +74,13 @@ class Embeddings:
        config: FrigateConfig,
        db: SqliteVecQueueDatabase,
        metrics: DataProcessorMetrics,
+        genai_manager=None,
    ) -> None:
        self.config = config
        self.db = db
        self.metrics = metrics
        self.requestor = InterProcessRequestor()
+        self.genai_manager = genai_manager

        self.image_inference_speed = InferenceSpeed(self.metrics.image_embeddings_speed)
        self.image_eps = EventsPerSecond()
@@ -104,7 +107,27 @@ class Embeddings:
                },
            )

-        if self.config.semantic_search.model == SemanticSearchModelEnum.jinav2:
+        model_cfg = self.config.semantic_search.model
+        is_genai_model = isinstance(model_cfg, str)
+
+        if is_genai_model:
+            embeddings_client = (
+                genai_manager.embeddings_client if genai_manager else None
+            )
+            if not embeddings_client:
+                raise ValueError(
+                    f"semantic_search.model is '{model_cfg}' (GenAI provider) but "
+                    "no embeddings client is configured. Ensure the GenAI provider "
+                    "has 'embeddings' in its roles."
+                )
+            self.embedding = GenAIEmbedding(embeddings_client)
+            self.text_embedding = lambda input_data: self.embedding(
+                input_data, embedding_type="text"
+            )
+            self.vision_embedding = lambda input_data: self.embedding(
+                input_data, embedding_type="vision"
+            )
+        elif model_cfg == SemanticSearchModelEnum.jinav2:
            # Single JinaV2Embedding instance for both text and vision
            self.embedding = JinaV2Embedding(
                model_size=self.config.semantic_search.model_size,
@@ -118,7 +141,8 @@ class Embeddings:
            self.vision_embedding = lambda input_data: self.embedding(
                input_data, embedding_type="vision"
            )
-        else:  # Default to jinav1
+        else:
+            # Default to jinav1
            self.text_embedding = JinaV1TextEmbedding(
                model_size=config.semantic_search.model_size,
                requestor=self.requestor,
@@ -136,8 +160,11 @@ class Embeddings:
        self.metrics.text_embeddings_eps.value = self.text_eps.eps()

    def get_model_definitions(self):
-        # Version-specific models
-        if self.config.semantic_search.model == SemanticSearchModelEnum.jinav2:
+        model_cfg = self.config.semantic_search.model
+        if isinstance(model_cfg, str):
+            # GenAI provider: no ONNX models to download
+            models = []
+        elif model_cfg == SemanticSearchModelEnum.jinav2:
            models = [
                "jinaai/jina-clip-v2-tokenizer",
                "jinaai/jina-clip-v2-model_fp16.onnx"
@@ -224,6 +251,14 @@ class Embeddings:

        embeddings = self.vision_embedding(valid_thumbs)

+        if len(embeddings) != len(valid_ids):
+            logger.warning(
+                "Batch embed returned %d embeddings for %d thumbnails; skipping batch",
+                len(embeddings),
+                len(valid_ids),
+            )
+            return []
+
        if upsert:
            items = []
            for i in range(len(valid_ids)):
@@ -246,9 +281,15 @@ class Embeddings:

    def embed_description(
        self, event_id: str, description: str, upsert: bool = True
-    ) -> np.ndarray:
+    ) -> np.ndarray | None:
        start = datetime.datetime.now().timestamp()
-        embedding = self.text_embedding([description])[0]
+        embeddings = self.text_embedding([description])
+        if not embeddings:
+            logger.warning(
+                "Failed to generate description embedding for event %s", event_id
+            )
+            return None
+        embedding = embeddings[0]

        if upsert:
            self.db.execute_sql(
@@ -271,8 +312,32 @@ class Embeddings:
        # upsert embeddings one by one to avoid token limit
        embeddings = []

-        for desc in event_descriptions.values():
-            embeddings.append(self.text_embedding([desc])[0])
+        for eid, desc in event_descriptions.items():
+            result = self.text_embedding([desc])
+            if not result:
+                logger.warning(
+                    "Failed to generate description embedding for event %s", eid
+                )
+                continue
+            embeddings.append(result[0])
+
+        if not embeddings:
+            logger.warning("No description embeddings generated in batch")
+            return np.array([])
+
+        # Build ids list for only successful embeddings - we need to track which succeeded
+        ids = list(event_descriptions.keys())
+        if len(embeddings) != len(ids):
+            # Rebuild ids/embeddings for only successful ones (match by order)
+            ids = []
+            embeddings_filtered = []
+            for eid, desc in event_descriptions.items():
+                result = self.text_embedding([desc])
+                if result:
+                    ids.append(eid)
+                    embeddings_filtered.append(result[0])
+            ids = ids
+            embeddings = embeddings_filtered

        if upsert:
            ids = list(event_descriptions.keys())
@@ -314,7 +379,10 @@ class Embeddings:

        batch_size = (
            4
-            if self.config.semantic_search.model == SemanticSearchModelEnum.jinav2
+            if (
+                isinstance(self.config.semantic_search.model, str)
+                or self.config.semantic_search.model == SemanticSearchModelEnum.jinav2
+            )
            else 32
        )
        current_page = 1
@@ -601,6 +669,8 @@ class Embeddings:
        if trigger.type == "description":
            logger.debug(f"Generating embedding for trigger description {trigger_name}")
            embedding = self.embed_description(None, trigger.data, upsert=False)
+            if embedding is None:
+                return b""
            return embedding.astype(np.float32).tobytes()

        elif trigger.type == "thumbnail":
@@ -636,6 +706,8 @@ class Embeddings:
                embedding = self.embed_thumbnail(
                    str(trigger.data), thumbnail, upsert=False
                )
+                if embedding is None:
+                    return b""
                return embedding.astype(np.float32).tobytes()

        else:
--- a/frigate/embeddings/genai_embedding.py
+++ b/frigate/embeddings/genai_embedding.py
@@ -0,0 +1,85 @@
+"""GenAI-backed embeddings for semantic search."""
+
+import io
+import logging
+from typing import TYPE_CHECKING
+
+import numpy as np
+from PIL import Image
+
+if TYPE_CHECKING:
+    from frigate.genai import GenAIClient
+
+logger = logging.getLogger(__name__)
+
+EMBEDDING_DIM = 768
+
+
+class GenAIEmbedding:
+    """Embedding adapter that delegates to a GenAI provider's embed API.
+
+    Provides the same interface as JinaV2Embedding for semantic search:
+    __call__(inputs, embedding_type) -> list[np.ndarray]. Output embeddings are
+    normalized to 768 dimensions for Frigate's sqlite-vec schema.
+    """
+
+    def __init__(self, client: "GenAIClient") -> None:
+        self.client = client
+
+    def __call__(
+        self,
+        inputs: list[str] | list[bytes] | list[Image.Image],
+        embedding_type: str = "text",
+    ) -> list[np.ndarray]:
+        """Generate embeddings for text or images.
+
+        Args:
+            inputs: List of strings (text) or bytes/PIL images (vision).
+            embedding_type: "text" or "vision".
+
+        Returns:
+            List of 768-dim numpy float32 arrays.
+        """
+        if not inputs:
+            return []
+
+        if embedding_type == "text":
+            texts = [str(x) for x in inputs]
+            embeddings = self.client.embed(texts=texts)
+        elif embedding_type == "vision":
+            images: list[bytes] = []
+            for inp in inputs:
+                if isinstance(inp, bytes):
+                    images.append(inp)
+                elif isinstance(inp, Image.Image):
+                    buf = io.BytesIO()
+                    inp.convert("RGB").save(buf, format="JPEG")
+                    images.append(buf.getvalue())
+                else:
+                    logger.warning(
+                        "GenAIEmbedding: skipping unsupported vision input type %s",
+                        type(inp).__name__,
+                    )
+            if not images:
+                return []
+            embeddings = self.client.embed(images=images)
+        else:
+            raise ValueError(
+                f"Invalid embedding_type '{embedding_type}'. Must be 'text' or 'vision'."
+            )
+
+        result = []
+        for emb in embeddings:
+            arr = np.asarray(emb, dtype=np.float32).flatten()
+            if arr.size != EMBEDDING_DIM:
+                if arr.size > EMBEDDING_DIM:
+                    arr = arr[:EMBEDDING_DIM]
+                else:
+                    arr = np.pad(
+                        arr,
+                        (0, EMBEDDING_DIM - arr.size),
+                        mode="constant",
+                        constant_values=0,
+                    )
+            result.append(arr)
+        return result
--- a/frigate/embeddings/maintainer.py
+++ b/frigate/embeddings/maintainer.py
@@ -59,7 +59,7 @@ from frigate.data_processing.real_time.license_plate import (
 from frigate.data_processing.types import DataProcessorMetrics, PostProcessDataEnum
 from frigate.db.sqlitevecq import SqliteVecQueueDatabase
 from frigate.events.types import EventTypeEnum, RegenerateDescriptionEnum
-from frigate.genai import get_genai_client
+from frigate.genai import GenAIClientManager
 from frigate.models import Event, Recordings, ReviewSegment, Trigger
 from frigate.util.builtin import serialize
 from frigate.util.file import get_event_thumbnail_bytes
@@ -116,8 +116,10 @@ class EmbeddingMaintainer(threading.Thread):
        models = [Event, Recordings, ReviewSegment, Trigger]
        db.bind(models)

+        self.genai_manager = GenAIClientManager(config)
+
        if config.semantic_search.enabled:
-            self.embeddings = Embeddings(config, db, metrics)
+            self.embeddings = Embeddings(config, db, metrics, self.genai_manager)

            # Check if we need to re-index events
            if config.semantic_search.reindex:
@@ -144,7 +146,6 @@ class EmbeddingMaintainer(threading.Thread):
        self.frame_manager = SharedMemoryFrameManager()

        self.detected_license_plates: dict[str, dict[str, Any]] = {}
-        self.genai_client = get_genai_client(config)

        # model runners to share between realtime and post processors
        if self.config.lpr.enabled:
@@ -203,12 +204,15 @@ class EmbeddingMaintainer(threading.Thread):
        # post processors
        self.post_processors: list[PostProcessorApi] = []

-        if self.genai_client is not None and any(
+        if self.genai_manager.vision_client is not None and any(
            c.review.genai.enabled_in_config for c in self.config.cameras.values()
        ):
            self.post_processors.append(
                ReviewDescriptionProcessor(
-                    self.config, self.requestor, self.metrics, self.genai_client
+                    self.config,
+                    self.requestor,
+                    self.metrics,
+                    self.genai_manager.vision_client,
                )
            )

@@ -246,7 +250,7 @@ class EmbeddingMaintainer(threading.Thread):
            )
            self.post_processors.append(semantic_trigger_processor)

-        if self.genai_client is not None and any(
+        if self.genai_manager.vision_client is not None and any(
            c.objects.genai.enabled_in_config for c in self.config.cameras.values()
        ):
            self.post_processors.append(
@@ -255,7 +259,7 @@ class EmbeddingMaintainer(threading.Thread):
                    self.embeddings,
                    self.requestor,
                    self.metrics,
-                    self.genai_client,
+                    self.genai_manager.vision_client,
                    semantic_trigger_processor,
                )
            )
--- a/frigate/embeddings/onnx/face_embedding.py
+++ b/frigate/embeddings/onnx/face_embedding.py
@@ -17,7 +17,7 @@ from .base_embedding import BaseEmbedding
 try:
    from tflite_runtime.interpreter import Interpreter
 except ModuleNotFoundError:
-    from ai_edge_litert.interpreter import Interpreter
+    from tensorflow.lite.python.interpreter import Interpreter

 logger = logging.getLogger(__name__)

--- a/frigate/events/audio.py
+++ b/frigate/events/audio.py
@@ -43,7 +43,7 @@ from frigate.video import start_or_restart_ffmpeg, stop_ffmpeg
 try:
    from tflite_runtime.interpreter import Interpreter
 except ModuleNotFoundError:
-    from ai_edge_litert.interpreter import Interpreter
+    from tensorflow.lite.python.interpreter import Interpreter


 logger = logging.getLogger(__name__)
--- a/frigate/genai/init.py
+++ b/frigate/genai/init.py
@@ -7,15 +7,27 @@ import os
 import re
 from typing import Any, Optional

+import numpy as np
 from playhouse.shortcuts import model_to_dict

 from frigate.config import CameraConfig, FrigateConfig, GenAIConfig, GenAIProviderEnum
 from frigate.const import CLIPS_DIR
 from frigate.data_processing.post.types import ReviewMetadata
+from frigate.genai.manager import GenAIClientManager
 from frigate.models import Event

 logger = logging.getLogger(__name__)

+__all__ = [
+    "GenAIClient",
+    "GenAIClientManager",
+    "GenAIConfig",
+    "GenAIProviderEnum",
+    "PROVIDERS",
+    "load_providers",
+    "register_genai_provider",
+]
+
 PROVIDERS = {}


@@ -293,6 +305,25 @@ Guidelines:
        """Get the context window size for this provider in tokens."""
        return 4096

+    def embed(
+        self,
+        texts: list[str] | None = None,
+        images: list[bytes] | None = None,
+    ) -> list[np.ndarray]:
+        """Generate embeddings for text and/or images.
+
+        Returns list of numpy arrays (one per input). Expected dimension is 768
+        for Frigate semantic search compatibility.
+
+        Providers that support embeddings should override this method.
+        """
+        logger.warning(
+            "%s does not support embeddings. "
+            "This method should be overridden by the provider implementation.",
+            self.__class__.__name__,
+        )
+        return []
+
    def chat_with_tools(
        self,
        messages: list[dict[str, Any]],
@@ -352,19 +383,6 @@ Guidelines:
        }


-def get_genai_client(config: FrigateConfig) -> Optional[GenAIClient]:
-    """Get the GenAI client."""
-    if not config.genai.provider:
-        return None
-
-    load_providers()
-    provider = PROVIDERS.get(config.genai.provider)
-    if provider:
-        return provider(config.genai)
-
-    return None
-
-
 def load_providers():
    package_dir = os.path.dirname(__file__)
    for filename in os.listdir(package_dir):
--- a/frigate/genai/llama_cpp.py
+++ b/frigate/genai/llama_cpp.py
@@ -1,18 +1,37 @@
 """llama.cpp Provider for Frigate AI."""

 import base64
+import io
 import json
 import logging
 from typing import Any, Optional

+import httpx
+import numpy as np
 import requests
+from PIL import Image

 from frigate.config import GenAIProviderEnum
 from frigate.genai import GenAIClient, register_genai_provider
+from frigate.genai.utils import parse_tool_calls_from_message

 logger = logging.getLogger(__name__)


+def _to_jpeg(img_bytes: bytes) -> bytes | None:
+    """Convert image bytes to JPEG. llama.cpp/STB does not support WebP."""
+    try:
+        img = Image.open(io.BytesIO(img_bytes))
+        if img.mode != "RGB":
+            img = img.convert("RGB")
+        buf = io.BytesIO()
+        img.save(buf, format="JPEG", quality=85)
+        return buf.getvalue()
+    except Exception as e:
+        logger.warning("Failed to convert image to JPEG: %s", e)
+        return None
+
+
@register_genai_provider(GenAIProviderEnum.llamacpp)
 class LlamaCppClient(GenAIClient):
    """Generative AI client for Frigate using llama.cpp server."""
@@ -67,6 +86,7 @@ class LlamaCppClient(GenAIClient):

            # Build request payload with llama.cpp native options
            payload = {
+                "model": self.genai_config.model,
                "messages": [
                    {
                        "role": "user",
@@ -99,7 +119,179 @@ class LlamaCppClient(GenAIClient):

    def get_context_size(self) -> int:
        """Get the context window size for llama.cpp."""
-        return self.genai_config.provider_options.get("context_size", 4096)
+        return self.provider_options.get("context_size", 4096)
+
+    def _build_payload(
+        self,
+        messages: list[dict[str, Any]],
+        tools: Optional[list[dict[str, Any]]],
+        tool_choice: Optional[str],
+        stream: bool = False,
+    ) -> dict[str, Any]:
+        """Build request payload for chat completions (sync or stream)."""
+        openai_tool_choice = None
+        if tool_choice:
+            if tool_choice == "none":
+                openai_tool_choice = "none"
+            elif tool_choice == "auto":
+                openai_tool_choice = "auto"
+            elif tool_choice == "required":
+                openai_tool_choice = "required"
+
+        payload: dict[str, Any] = {
+            "messages": messages,
+            "model": self.genai_config.model,
+        }
+        if stream:
+            payload["stream"] = True
+        if tools:
+            payload["tools"] = tools
+            if openai_tool_choice is not None:
+                payload["tool_choice"] = openai_tool_choice
+        provider_opts = {
+            k: v for k, v in self.provider_options.items() if k != "context_size"
+        }
+        payload.update(provider_opts)
+        return payload
+
+    def _message_from_choice(self, choice: dict[str, Any]) -> dict[str, Any]:
+        """Parse OpenAI-style choice into {content, tool_calls, finish_reason}."""
+        message = choice.get("message", {})
+        content = message.get("content")
+        content = content.strip() if content else None
+        tool_calls = parse_tool_calls_from_message(message)
+        finish_reason = choice.get("finish_reason") or (
+            "tool_calls" if tool_calls else "stop" if content else "error"
+        )
+        return {
+            "content": content,
+            "tool_calls": tool_calls,
+            "finish_reason": finish_reason,
+        }
+
+    @staticmethod
+    def _streamed_tool_calls_to_list(
+        tool_calls_by_index: dict[int, dict[str, Any]],
+    ) -> Optional[list[dict[str, Any]]]:
+        """Convert streamed tool_calls index map to list of {id, name, arguments}."""
+        if not tool_calls_by_index:
+            return None
+        result = []
+        for idx in sorted(tool_calls_by_index.keys()):
+            t = tool_calls_by_index[idx]
+            args_str = t.get("arguments") or "{}"
+            try:
+                arguments = json.loads(args_str)
+            except json.JSONDecodeError:
+                arguments = {}
+            result.append(
+                {
+                    "id": t.get("id", ""),
+                    "name": t.get("name", ""),
+                    "arguments": arguments,
+                }
+            )
+        return result if result else None
+
+    def embed(
+        self,
+        texts: list[str] | None = None,
+        images: list[bytes] | None = None,
+    ) -> list[np.ndarray]:
+        """Generate embeddings via llama.cpp /embeddings endpoint.
+
+        Supports batch requests. Uses content format with prompt_string and
+        multimodal_data for images (PR #15108). Server must be started with
+        --embeddings and --mmproj for multimodal support.
+        """
+        if self.provider is None:
+            logger.warning(
+                "llama.cpp provider has not been initialized. Check your llama.cpp configuration."
+            )
+            return []
+
+        texts = texts or []
+        images = images or []
+        if not texts and not images:
+            return []
+
+        EMBEDDING_DIM = 768
+
+        content = []
+        for text in texts:
+            content.append({"prompt_string": text})
+        for img in images:
+            # llama.cpp uses STB which does not support WebP; convert to JPEG
+            jpeg_bytes = _to_jpeg(img)
+            to_encode = jpeg_bytes if jpeg_bytes is not None else img
+            encoded = base64.b64encode(to_encode).decode("utf-8")
+            # prompt_string must contain <__media__> placeholder for image tokenization
+            content.append(
+                {
+                    "prompt_string": "<__media__>\n",
+                    "multimodal_data": [encoded],
+                }
+            )
+
+        try:
+            response = requests.post(
+                f"{self.provider}/embeddings",
+                json={"model": self.genai_config.model, "content": content},
+                timeout=self.timeout,
+            )
+            response.raise_for_status()
+            result = response.json()
+
+            items = result.get("data", result) if isinstance(result, dict) else result
+            if not isinstance(items, list):
+                logger.warning("llama.cpp embeddings returned unexpected format")
+                return []
+
+            embeddings = []
+            for item in items:
+                emb = item.get("embedding") if isinstance(item, dict) else None
+                if emb is None:
+                    logger.warning("llama.cpp embeddings item missing embedding field")
+                    continue
+                arr = np.array(emb, dtype=np.float32)
+                orig_dim = arr.size
+                if orig_dim != EMBEDDING_DIM:
+                    if orig_dim > EMBEDDING_DIM:
+                        arr = arr[:EMBEDDING_DIM]
+                        logger.debug(
+                            "Truncated llama.cpp embedding from %d to %d dimensions",
+                            orig_dim,
+                            EMBEDDING_DIM,
+                        )
+                    else:
+                        arr = np.pad(
+                            arr,
+                            (0, EMBEDDING_DIM - orig_dim),
+                            mode="constant",
+                            constant_values=0,
+                        )
+                        logger.debug(
+                            "Padded llama.cpp embedding from %d to %d dimensions",
+                            orig_dim,
+                            EMBEDDING_DIM,
+                        )
+                embeddings.append(arr)
+            return embeddings
+        except requests.exceptions.Timeout:
+            logger.warning("llama.cpp embeddings request timed out")
+            return []
+        except requests.exceptions.RequestException as e:
+            error_detail = str(e)
+            if hasattr(e, "response") and e.response is not None:
+                try:
+                    error_detail = f"{str(e)} - Response: {e.response.text[:500]}"
+                except Exception:
+                    pass
+            logger.warning("llama.cpp embeddings error: %s", error_detail)
+            return []
+        except Exception as e:
+            logger.warning("Unexpected error in llama.cpp embeddings: %s", str(e))
+            return []

    def chat_with_tools(
        self,
@@ -122,31 +314,8 @@ class LlamaCppClient(GenAIClient):
                "tool_calls": None,
                "finish_reason": "error",
            }
-
        try:
-            openai_tool_choice = None
-            if tool_choice:
-                if tool_choice == "none":
-                    openai_tool_choice = "none"
-                elif tool_choice == "auto":
-                    openai_tool_choice = "auto"
-                elif tool_choice == "required":
-                    openai_tool_choice = "required"
-
-            payload = {
-                "messages": messages,
-            }
-
-            if tools:
-                payload["tools"] = tools
-                if openai_tool_choice is not None:
-                    payload["tool_choice"] = openai_tool_choice
-
-            provider_opts = {
-                k: v for k, v in self.provider_options.items() if k != "context_size"
-            }
-            payload.update(provider_opts)
-
+            payload = self._build_payload(messages, tools, tool_choice, stream=False)
            response = requests.post(
                f"{self.provider}/v1/chat/completions",
                json=payload,
@@ -154,60 +323,13 @@ class LlamaCppClient(GenAIClient):
            )
            response.raise_for_status()
            result = response.json()
-
            if result is None or "choices" not in result or len(result["choices"]) == 0:
                return {
                    "content": None,
                    "tool_calls": None,
                    "finish_reason": "error",
                }
-
-            choice = result["choices"][0]
-            message = choice.get("message", {})
-
-            content = message.get("content")
-            if content:
-                content = content.strip()
-            else:
-                content = None
-
-            tool_calls = None
-            if "tool_calls" in message and message["tool_calls"]:
-                tool_calls = []
-                for tool_call in message["tool_calls"]:
-                    try:
-                        function_data = tool_call.get("function", {})
-                        arguments_str = function_data.get("arguments", "{}")
-                        arguments = json.loads(arguments_str)
-                    except (json.JSONDecodeError, KeyError, TypeError) as e:
-                        logger.warning(
-                            f"Failed to parse tool call arguments: {e}, "
-                            f"tool: {function_data.get('name', 'unknown')}"
-                        )
-                        arguments = {}
-
-                    tool_calls.append(
-                        {
-                            "id": tool_call.get("id", ""),
-                            "name": function_data.get("name", ""),
-                            "arguments": arguments,
-                        }
-                    )
-
-            finish_reason = "error"
-            if "finish_reason" in choice and choice["finish_reason"]:
-                finish_reason = choice["finish_reason"]
-            elif tool_calls:
-                finish_reason = "tool_calls"
-            elif content:
-                finish_reason = "stop"
-
-            return {
-                "content": content,
-                "tool_calls": tool_calls,
-                "finish_reason": finish_reason,
-            }
-
+            return self._message_from_choice(result["choices"][0])
        except requests.exceptions.Timeout as e:
            logger.warning("llama.cpp request timed out: %s", str(e))
            return {
@@ -219,8 +341,7 @@ class LlamaCppClient(GenAIClient):
            error_detail = str(e)
            if hasattr(e, "response") and e.response is not None:
                try:
-                    error_body = e.response.text
-                    error_detail = f"{str(e)} - Response: {error_body[:500]}"
+                    error_detail = f"{str(e)} - Response: {e.response.text[:500]}"
                except Exception:
                    pass
            logger.warning("llama.cpp returned an error: %s", error_detail)
@@ -236,3 +357,111 @@ class LlamaCppClient(GenAIClient):
                "tool_calls": None,
                "finish_reason": "error",
            }
+
+    async def chat_with_tools_stream(
+        self,
+        messages: list[dict[str, Any]],
+        tools: Optional[list[dict[str, Any]]] = None,
+        tool_choice: Optional[str] = "auto",
+    ):
+        """Stream chat with tools via OpenAI-compatible streaming API."""
+        if self.provider is None:
+            logger.warning(
+                "llama.cpp provider has not been initialized. Check your llama.cpp configuration."
+            )
+            yield (
+                "message",
+                {
+                    "content": None,
+                    "tool_calls": None,
+                    "finish_reason": "error",
+                },
+            )
+            return
+        try:
+            payload = self._build_payload(messages, tools, tool_choice, stream=True)
+            content_parts: list[str] = []
+            tool_calls_by_index: dict[int, dict[str, Any]] = {}
+            finish_reason = "stop"
+
+            async with httpx.AsyncClient(timeout=float(self.timeout)) as client:
+                async with client.stream(
+                    "POST",
+                    f"{self.provider}/v1/chat/completions",
+                    json=payload,
+                ) as response:
+                    response.raise_for_status()
+                    async for line in response.aiter_lines():
+                        if not line.startswith("data: "):
+                            continue
+                        data_str = line[6:].strip()
+                        if data_str == "[DONE]":
+                            break
+                        try:
+                            data = json.loads(data_str)
+                        except json.JSONDecodeError:
+                            continue
+                        choices = data.get("choices") or []
+                        if not choices:
+                            continue
+                        delta = choices[0].get("delta", {})
+                        if choices[0].get("finish_reason"):
+                            finish_reason = choices[0]["finish_reason"]
+                        if delta.get("content"):
+                            content_parts.append(delta["content"])
+                            yield ("content_delta", delta["content"])
+                        for tc in delta.get("tool_calls") or []:
+                            idx = tc.get("index", 0)
+                            fn = tc.get("function") or {}
+                            if idx not in tool_calls_by_index:
+                                tool_calls_by_index[idx] = {
+                                    "id": tc.get("id", ""),
+                                    "name": tc.get("name") or fn.get("name", ""),
+                                    "arguments": "",
+                                }
+                            t = tool_calls_by_index[idx]
+                            if tc.get("id"):
+                                t["id"] = tc["id"]
+                            name = tc.get("name") or fn.get("name")
+                            if name:
+                                t["name"] = name
+                            arg = tc.get("arguments") or fn.get("arguments")
+                            if arg is not None:
+                                t["arguments"] += (
+                                    arg if isinstance(arg, str) else json.dumps(arg)
+                                )
+
+            full_content = "".join(content_parts).strip() or None
+            tool_calls_list = self._streamed_tool_calls_to_list(tool_calls_by_index)
+            if tool_calls_list:
+                finish_reason = "tool_calls"
+            yield (
+                "message",
+                {
+                    "content": full_content,
+                    "tool_calls": tool_calls_list,
+                    "finish_reason": finish_reason,
+                },
+            )
+        except httpx.HTTPStatusError as e:
+            logger.warning("llama.cpp streaming HTTP error: %s", e)
+            yield (
+                "message",
+                {
+                    "content": None,
+                    "tool_calls": None,
+                    "finish_reason": "error",
+                },
+            )
+        except Exception as e:
+            logger.warning(
+                "Unexpected error in llama.cpp chat_with_tools_stream: %s", str(e)
+            )
+            yield (
+                "message",
+                {
+                    "content": None,
+                    "tool_calls": None,
+                    "finish_reason": "error",
+                },
+            )
--- a/frigate/genai/manager.py
+++ b/frigate/genai/manager.py
@@ -0,0 +1,89 @@
+"""GenAI client manager for Frigate.
+
+Manages GenAI provider clients from Frigate config. Configuration is read only
+in _update_config(); no other code should read config.genai. Exposes clients
+by role: tool_client, vision_client, embeddings_client.
+"""
+
+import logging
+from typing import TYPE_CHECKING, Optional
+
+from frigate.config import FrigateConfig
+from frigate.config.camera.genai import GenAIRoleEnum
+
+if TYPE_CHECKING:
+    from frigate.genai import GenAIClient
+
+logger = logging.getLogger(__name__)
+
+
+class GenAIClientManager:
+    """Manages GenAI provider clients from Frigate config."""
+
+    def __init__(self, config: FrigateConfig) -> None:
+        self._config = config
+        self._tool_client: Optional[GenAIClient] = None
+        self._vision_client: Optional[GenAIClient] = None
+        self._embeddings_client: Optional[GenAIClient] = None
+        self._update_config()
+
+    def _update_config(self) -> None:
+        """Build role clients from current Frigate config.genai.
+
+        Called from __init__ and can be called again when config is reloaded.
+        Each role (tools, vision, embeddings) gets the client for the provider
+        that has that role in its roles list.
+        """
+        from frigate.genai import PROVIDERS, load_providers
+
+        self._tool_client = None
+        self._vision_client = None
+        self._embeddings_client = None
+
+        if not self._config.genai:
+            return
+
+        load_providers()
+
+        for _name, genai_cfg in self._config.genai.items():
+            if not genai_cfg.provider:
+                continue
+            provider_cls = PROVIDERS.get(genai_cfg.provider)
+            if not provider_cls:
+                logger.warning(
+                    "Unknown GenAI provider %s in config, skipping.",
+                    genai_cfg.provider,
+                )
+                continue
+            try:
+                client = provider_cls(genai_cfg)
+            except Exception as e:
+                logger.exception(
+                    "Failed to create GenAI client for provider %s: %s",
+                    genai_cfg.provider,
+                    e,
+                )
+                continue
+
+            for role in genai_cfg.roles:
+                if role == GenAIRoleEnum.tools:
+                    self._tool_client = client
+                elif role == GenAIRoleEnum.vision:
+                    self._vision_client = client
+                elif role == GenAIRoleEnum.embeddings:
+                    self._embeddings_client = client
+
+    @property
+    def tool_client(self) -> "Optional[GenAIClient]":
+        """Client configured for the tools role (e.g. chat with function calling)."""
+        return self._tool_client
+
+    @property
+    def vision_client(self) -> "Optional[GenAIClient]":
+        """Client configured for the vision role (e.g. review descriptions, object descriptions)."""
+        return self._vision_client
+
+    @property
+    def embeddings_client(self) -> "Optional[GenAIClient]":
+        """Client configured for the embeddings role."""
+        return self._embeddings_client
--- a/frigate/genai/ollama.py
+++ b/frigate/genai/ollama.py
@@ -1,15 +1,16 @@
 """Ollama Provider for Frigate AI."""

-import json
 import logging
 from typing import Any, Optional

 from httpx import RemoteProtocolError, TimeoutException
+from ollama import AsyncClient as OllamaAsyncClient
 from ollama import Client as ApiClient
 from ollama import ResponseError

 from frigate.config import GenAIProviderEnum
 from frigate.genai import GenAIClient, register_genai_provider
+from frigate.genai.utils import parse_tool_calls_from_message

 logger = logging.getLogger(__name__)

@@ -88,6 +89,73 @@ class OllamaClient(GenAIClient):
            "num_ctx", 4096
        )

+    def _build_request_params(
+        self,
+        messages: list[dict[str, Any]],
+        tools: Optional[list[dict[str, Any]]],
+        tool_choice: Optional[str],
+        stream: bool = False,
+    ) -> dict[str, Any]:
+        """Build request_messages and params for chat (sync or stream)."""
+        request_messages = []
+        for msg in messages:
+            msg_dict = {
+                "role": msg.get("role"),
+                "content": msg.get("content", ""),
+            }
+            if msg.get("tool_call_id"):
+                msg_dict["tool_call_id"] = msg["tool_call_id"]
+            if msg.get("name"):
+                msg_dict["name"] = msg["name"]
+            if msg.get("tool_calls"):
+                msg_dict["tool_calls"] = msg["tool_calls"]
+            request_messages.append(msg_dict)
+
+        request_params: dict[str, Any] = {
+            "model": self.genai_config.model,
+            "messages": request_messages,
+            **self.provider_options,
+        }
+        if stream:
+            request_params["stream"] = True
+        if tools:
+            request_params["tools"] = tools
+            if tool_choice:
+                request_params["tool_choice"] = (
+                    "none"
+                    if tool_choice == "none"
+                    else "required"
+                    if tool_choice == "required"
+                    else "auto"
+                )
+        return request_params
+
+    def _message_from_response(self, response: dict[str, Any]) -> dict[str, Any]:
+        """Parse Ollama chat response into {content, tool_calls, finish_reason}."""
+        if not response or "message" not in response:
+            return {
+                "content": None,
+                "tool_calls": None,
+                "finish_reason": "error",
+            }
+        message = response["message"]
+        content = message.get("content", "").strip() if message.get("content") else None
+        tool_calls = parse_tool_calls_from_message(message)
+        finish_reason = "error"
+        if response.get("done"):
+            finish_reason = (
+                "tool_calls" if tool_calls else "stop" if content else "error"
+            )
+        elif tool_calls:
+            finish_reason = "tool_calls"
+        elif content:
+            finish_reason = "stop"
+        return {
+            "content": content,
+            "tool_calls": tool_calls,
+            "finish_reason": finish_reason,
+        }
+
    def chat_with_tools(
        self,
        messages: list[dict[str, Any]],
@@ -103,93 +171,12 @@ class OllamaClient(GenAIClient):
                "tool_calls": None,
                "finish_reason": "error",
            }
-
        try:
-            request_messages = []
-            for msg in messages:
-                msg_dict = {
-                    "role": msg.get("role"),
-                    "content": msg.get("content", ""),
-                }
-                if msg.get("tool_call_id"):
-                    msg_dict["tool_call_id"] = msg["tool_call_id"]
-                if msg.get("name"):
-                    msg_dict["name"] = msg["name"]
-                if msg.get("tool_calls"):
-                    msg_dict["tool_calls"] = msg["tool_calls"]
-                request_messages.append(msg_dict)
-
-            request_params = {
-                "model": self.genai_config.model,
-                "messages": request_messages,
-            }
-
-            if tools:
-                request_params["tools"] = tools
-                if tool_choice:
-                    if tool_choice == "none":
-                        request_params["tool_choice"] = "none"
-                    elif tool_choice == "required":
-                        request_params["tool_choice"] = "required"
-                    elif tool_choice == "auto":
-                        request_params["tool_choice"] = "auto"
-
-            request_params.update(self.provider_options)
-
-            response = self.provider.chat(**request_params)
-
-            if not response or "message" not in response:
-                return {
-                    "content": None,
-                    "tool_calls": None,
-                    "finish_reason": "error",
-                }
-
-            message = response["message"]
-            content = (
-                message.get("content", "").strip() if message.get("content") else None
+            request_params = self._build_request_params(
+                messages, tools, tool_choice, stream=False
            )
-
-            tool_calls = None
-            if "tool_calls" in message and message["tool_calls"]:
-                tool_calls = []
-                for tool_call in message["tool_calls"]:
-                    try:
-                        function_data = tool_call.get("function", {})
-                        arguments_str = function_data.get("arguments", "{}")
-                        arguments = json.loads(arguments_str)
-                    except (json.JSONDecodeError, KeyError, TypeError) as e:
-                        logger.warning(
-                            f"Failed to parse tool call arguments: {e}, "
-                            f"tool: {function_data.get('name', 'unknown')}"
-                        )
-                        arguments = {}
-
-                    tool_calls.append(
-                        {
-                            "id": tool_call.get("id", ""),
-                            "name": function_data.get("name", ""),
-                            "arguments": arguments,
-                        }
-                    )
-
-            finish_reason = "error"
-            if "done" in response and response["done"]:
-                if tool_calls:
-                    finish_reason = "tool_calls"
-                elif content:
-                    finish_reason = "stop"
-            elif tool_calls:
-                finish_reason = "tool_calls"
-            elif content:
-                finish_reason = "stop"
-
-            return {
-                "content": content,
-                "tool_calls": tool_calls,
-                "finish_reason": finish_reason,
-            }
-
+            response = self.provider.chat(**request_params)
+            return self._message_from_response(response)
        except (TimeoutException, ResponseError, ConnectionError) as e:
            logger.warning("Ollama returned an error: %s", str(e))
            return {
@@ -204,3 +191,89 @@ class OllamaClient(GenAIClient):
                "tool_calls": None,
                "finish_reason": "error",
            }
+
+    async def chat_with_tools_stream(
+        self,
+        messages: list[dict[str, Any]],
+        tools: Optional[list[dict[str, Any]]] = None,
+        tool_choice: Optional[str] = "auto",
+    ):
+        """Stream chat with tools; yields content deltas then final message."""
+        if self.provider is None:
+            logger.warning(
+                "Ollama provider has not been initialized. Check your Ollama configuration."
+            )
+            yield (
+                "message",
+                {
+                    "content": None,
+                    "tool_calls": None,
+                    "finish_reason": "error",
+                },
+            )
+            return
+        try:
+            request_params = self._build_request_params(
+                messages, tools, tool_choice, stream=True
+            )
+            async_client = OllamaAsyncClient(
+                host=self.genai_config.base_url,
+                timeout=self.timeout,
+            )
+            content_parts: list[str] = []
+            final_message: dict[str, Any] | None = None
+            try:
+                stream = await async_client.chat(**request_params)
+                async for chunk in stream:
+                    if not chunk or "message" not in chunk:
+                        continue
+                    msg = chunk.get("message", {})
+                    delta = msg.get("content") or ""
+                    if delta:
+                        content_parts.append(delta)
+                        yield ("content_delta", delta)
+                    if chunk.get("done"):
+                        full_content = "".join(content_parts).strip() or None
+                        tool_calls = parse_tool_calls_from_message(msg)
+                        final_message = {
+                            "content": full_content,
+                            "tool_calls": tool_calls,
+                            "finish_reason": "tool_calls" if tool_calls else "stop",
+                        }
+                        break
+            finally:
+                await async_client.close()
+
+            if final_message is not None:
+                yield ("message", final_message)
+            else:
+                yield (
+                    "message",
+                    {
+                        "content": "".join(content_parts).strip() or None,
+                        "tool_calls": None,
+                        "finish_reason": "stop",
+                    },
+                )
+        except (TimeoutException, ResponseError, ConnectionError) as e:
+            logger.warning("Ollama streaming error: %s", str(e))
+            yield (
+                "message",
+                {
+                    "content": None,
+                    "tool_calls": None,
+                    "finish_reason": "error",
+                },
+            )
+        except Exception as e:
+            logger.warning(
+                "Unexpected error in Ollama chat_with_tools_stream: %s", str(e)
+            )
+            yield (
+                "message",
+                {
+                    "content": None,
+                    "tool_calls": None,
+                    "finish_reason": "error",
+                },
+            )
--- a/frigate/genai/utils.py
+++ b/frigate/genai/utils.py
@@ -0,0 +1,70 @@
+"""Shared helpers for GenAI providers and chat (OpenAI-style messages, tool call parsing)."""
+
+import json
+import logging
+from typing import Any, List, Optional
+
+logger = logging.getLogger(__name__)
+
+
+def parse_tool_calls_from_message(
+    message: dict[str, Any],
+) -> Optional[list[dict[str, Any]]]:
+    """
+    Parse tool_calls from an OpenAI-style message dict.
+
+    Message may have "tool_calls" as a list of:
+      {"id": str, "function": {"name": str, "arguments": str}, ...}
+
+    Returns a list of {"id", "name", "arguments"} with arguments parsed as dict,
+    or None if no tool_calls. Used by Ollama and LlamaCpp (non-stream) responses.
+    """
+    raw = message.get("tool_calls")
+    if not raw or not isinstance(raw, list):
+        return None
+    result = []
+    for tool_call in raw:
+        function_data = tool_call.get("function") or {}
+        try:
+            arguments_str = function_data.get("arguments") or "{}"
+            arguments = json.loads(arguments_str)
+        except (json.JSONDecodeError, KeyError, TypeError) as e:
+            logger.warning(
+                "Failed to parse tool call arguments: %s, tool: %s",
+                e,
+                function_data.get("name", "unknown"),
+            )
+            arguments = {}
+        result.append(
+            {
+                "id": tool_call.get("id", ""),
+                "name": function_data.get("name", ""),
+                "arguments": arguments,
+            }
+        )
+    return result if result else None
+
+
+def build_assistant_message_for_conversation(
+    content: Any,
+    tool_calls_raw: Optional[List[dict[str, Any]]],
+) -> dict[str, Any]:
+    """
+    Build the assistant message dict in OpenAI format for appending to a conversation.
+
+    tool_calls_raw: list of {"id", "name", "arguments"} (arguments as dict), or None.
+    """
+    msg: dict[str, Any] = {"role": "assistant", "content": content}
+    if tool_calls_raw:
+        msg["tool_calls"] = [
+            {
+                "id": tc["id"],
+                "type": "function",
+                "function": {
+                    "name": tc["name"],
+                    "arguments": json.dumps(tc.get("arguments") or {}),
+                },
+            }
+            for tc in tool_calls_raw
+        ]
+    return msg
--- a/frigate/util/config.py
+++ b/frigate/util/config.py
@@ -438,6 +438,13 @@ def migrate_018_0(config: dict[str, dict[str, Any]]) -> dict[str, dict[str, Any]
    """Handle migrating frigate config to 0.18-0"""
    new_config = config.copy()

+    # Migrate GenAI to new format
+    genai = new_config.get("genai")
+
+    if genai and genai.get("provider"):
+        genai["roles"] = ["embeddings", "vision", "tools"]
+        new_config["genai"] = {"default": genai}
+
    # Remove deprecated sync_recordings from global record config
    if new_config.get("record", {}).get("sync_recordings") is not None:
        del new_config["record"]["sync_recordings"]
--- a/frigate/util/services.py
+++ b/frigate/util/services.py
@@ -121,7 +121,7 @@ def get_cpu_stats() -> dict[str, dict]:
        pid = str(process.info["pid"])
        try:
            cpu_percent = process.info["cpu_percent"]
-            cmdline = " ".join(process.info["cmdline"]).rstrip()
+            cmdline = process.info["cmdline"]

            with open(f"/proc/{pid}/stat", "r") as f:
                stats = f.readline().split()
@@ -155,7 +155,7 @@ def get_cpu_stats() -> dict[str, dict]:
                "cpu": str(cpu_percent),
                "cpu_average": str(round(cpu_average_usage, 2)),
                "mem": f"{mem_pct}",
-                "cmdline": clean_camera_user_pass(cmdline),
+                "cmdline": clean_camera_user_pass(" ".join(cmdline)),
            }
        except Exception:
            continue
--- a/web/package-lock.json
+++ b/web/package-lock.json
--- a/web/package.json
+++ b/web/package.json
@@ -71,6 +71,8 @@
    "react-icons": "^5.5.0",
    "react-konva": "^18.2.10",
    "react-router-dom": "^6.30.3",
+    "react-markdown": "^9.0.1",
+    "remark-gfm": "^4.0.0",
    "react-swipeable": "^7.0.2",
    "react-tracked": "^2.0.1",
    "react-transition-group": "^4.4.5",
--- a/web/public/locales/en/common.json
+++ b/web/public/locales/en/common.json
@@ -127,6 +127,7 @@
    "cancel": "Cancel",
    "close": "Close",
    "copy": "Copy",
+    "copiedToClipboard": "Copied to clipboard",
    "back": "Back",
    "history": "History",
    "fullscreen": "Fullscreen",
@@ -245,6 +246,7 @@
    "uiPlayground": "UI Playground",
    "faceLibrary": "Face Library",
    "classification": "Classification",
+    "chat": "Chat",
    "user": {
      "title": "User",
      "account": "Account",
--- a/web/public/locales/en/views/chat.json
+++ b/web/public/locales/en/views/chat.json
@@ -0,0 +1,24 @@
+{
+  "title": "Frigate Chat",
+  "subtitle": "Your AI assistant for camera management and insights",
+  "placeholder": "Ask anything...",
+  "error": "Something went wrong. Please try again.",
+  "processing": "Processing...",
+  "toolsUsed": "Used: {{tools}}",
+  "showTools": "Show tools ({{count}})",
+  "hideTools": "Hide tools",
+  "call": "Call",
+  "result": "Result",
+  "arguments": "Arguments:",
+  "response": "Response:",
+  "send": "Send",
+  "suggested_requests": "Try asking:",
+  "starting_requests": {
+    "show_recent_events": "Show recent events",
+    "show_camera_status": "Show camera status"
+  },
+  "starting_requests_prompts": {
+    "show_recent_events": "Show me the recent events from the last hour",
+    "show_camera_status": "What is the current status of my cameras?"
+  }
+}
--- a/web/src/App.tsx
+++ b/web/src/App.tsx
@@ -27,6 +27,7 @@ const Settings = lazy(() => import("@/pages/Settings"));
 const UIPlayground = lazy(() => import("@/pages/UIPlayground"));
 const FaceLibrary = lazy(() => import("@/pages/FaceLibrary"));
 const Classification = lazy(() => import("@/pages/ClassificationModel"));
+const Chat = lazy(() => import("@/pages/Chat"));
 const Logs = lazy(() => import("@/pages/Logs"));
 const AccessDenied = lazy(() => import("@/pages/AccessDenied"));

@@ -106,6 +107,7 @@ function DefaultAppView() {
              <Route path="/logs" element={<Logs />} />
              <Route path="/faces" element={<FaceLibrary />} />
              <Route path="/classification" element={<Classification />} />
+              <Route path="/chat" element={<Chat />} />
              <Route path="/playground" element={<UIPlayground />} />
            </Route>
            <Route path="/unauthorized" element={<AccessDenied />} />
--- a/web/src/components/chat/ChatEventThumbnailsRow.tsx
+++ b/web/src/components/chat/ChatEventThumbnailsRow.tsx
@@ -0,0 +1,42 @@
+import { useApiHost } from "@/api";
+
+type ChatEventThumbnailsRowProps = {
+  events: { id: string }[];
+};
+
+/**
+ * Horizontal scroll row of event thumbnail images for chat (e.g. after search_objects).
+ * Renders nothing when events is empty.
+ */
+export function ChatEventThumbnailsRow({
+  events,
+}: ChatEventThumbnailsRowProps) {
+  const apiHost = useApiHost();
+
+  if (events.length === 0) return null;
+
+  return (
+    <div className="flex min-w-0 max-w-full flex-col gap-1 self-start">
+      <div className="scrollbar-container min-w-0 overflow-x-auto">
+        <div className="flex w-max gap-2">
+          {events.map((event) => (
+            <a
+              key={event.id}
+              href={`/explore?event_id=${event.id}`}
+              target="_blank"
+              rel="noopener noreferrer"
+              className="relative aspect-square size-32 shrink-0 overflow-hidden rounded-lg"
+            >
+              <img
+                className="size-full object-cover"
+                src={`${apiHost}api/events/${event.id}/thumbnail.webp`}
+                alt=""
+                loading="lazy"
+              />
+            </a>
+          ))}
+        </div>
+      </div>
+    </div>
+  );
+}
--- a/web/src/components/chat/ChatMessage.tsx
+++ b/web/src/components/chat/ChatMessage.tsx
@@ -0,0 +1,208 @@
+import { useState, useEffect, useRef } from "react";
+import ReactMarkdown from "react-markdown";
+import remarkGfm from "remark-gfm";
+import { useTranslation } from "react-i18next";
+import copy from "copy-to-clipboard";
+import { toast } from "sonner";
+import { FaCopy, FaPencilAlt } from "react-icons/fa";
+import { FaArrowUpLong } from "react-icons/fa6";
+import { Button } from "@/components/ui/button";
+import { Textarea } from "@/components/ui/textarea";
+import {
+  Tooltip,
+  TooltipContent,
+  TooltipTrigger,
+} from "@/components/ui/tooltip";
+import { cn } from "@/lib/utils";
+
+type MessageBubbleProps = {
+  role: "user" | "assistant";
+  content: string;
+  messageIndex?: number;
+  onEditSubmit?: (messageIndex: number, newContent: string) => void;
+  isComplete?: boolean;
+};
+
+export function MessageBubble({
+  role,
+  content,
+  messageIndex = 0,
+  onEditSubmit,
+  isComplete = true,
+}: MessageBubbleProps) {
+  const { t } = useTranslation(["views/chat", "common"]);
+  const isUser = role === "user";
+  const [isEditing, setIsEditing] = useState(false);
+  const [draftContent, setDraftContent] = useState(content);
+  const editInputRef = useRef<HTMLTextAreaElement>(null);
+
+  useEffect(() => {
+    setDraftContent(content);
+  }, [content]);
+
+  useEffect(() => {
+    if (isEditing) {
+      editInputRef.current?.focus();
+      editInputRef.current?.setSelectionRange(
+        editInputRef.current.value.length,
+        editInputRef.current.value.length,
+      );
+    }
+  }, [isEditing]);
+
+  const handleCopy = () => {
+    const text = content?.trim() || "";
+    if (!text) return;
+    if (copy(text)) {
+      toast.success(t("button.copiedToClipboard", { ns: "common" }));
+    }
+  };
+
+  const handleEditClick = () => {
+    setDraftContent(content);
+    setIsEditing(true);
+  };
+
+  const handleEditSubmit = () => {
+    const trimmed = draftContent.trim();
+    if (!trimmed || onEditSubmit == null) return;
+    onEditSubmit(messageIndex, trimmed);
+    setIsEditing(false);
+  };
+
+  const handleEditCancel = () => {
+    setDraftContent(content);
+    setIsEditing(false);
+  };
+
+  const handleEditKeyDown = (e: React.KeyboardEvent<HTMLTextAreaElement>) => {
+    if (e.key === "Enter" && !e.shiftKey) {
+      e.preventDefault();
+      handleEditSubmit();
+    }
+    if (e.key === "Escape") {
+      handleEditCancel();
+    }
+  };
+
+  if (isUser && isEditing) {
+    return (
+      <div className="flex w-full max-w-full flex-col gap-2 self-end">
+        <Textarea
+          ref={editInputRef}
+          value={draftContent}
+          onChange={(e) => setDraftContent(e.target.value)}
+          onKeyDown={handleEditKeyDown}
+          className="min-h-[80px] w-full resize-y rounded-lg bg-primary px-3 py-2 text-primary-foreground placeholder:text-primary-foreground/60"
+          placeholder={t("placeholder")}
+          rows={3}
+        />
+        <div className="flex items-center gap-2 self-end">
+          <Button
+            variant="ghost"
+            size="sm"
+            className="text-muted-foreground hover:text-foreground"
+            onClick={handleEditCancel}
+          >
+            {t("button.cancel", { ns: "common" })}
+          </Button>
+          <Button
+            variant="select"
+            size="icon"
+            className="size-9 rounded-full"
+            disabled={!draftContent.trim()}
+            onClick={handleEditSubmit}
+            aria-label={t("send")}
+          >
+            <FaArrowUpLong size="16" />
+          </Button>
+        </div>
+      </div>
+    );
+  }
+
+  return (
+    <div
+      className={cn(
+        "flex flex-col gap-1",
+        isUser ? "items-end self-end" : "items-start self-start",
+      )}
+    >
+      <div
+        className={cn(
+          "rounded-lg px-3 py-2",
+          isUser ? "bg-primary text-primary-foreground" : "bg-muted",
+        )}
+      >
+        {isUser ? (
+          content
+        ) : (
+          <ReactMarkdown
+            remarkPlugins={[remarkGfm]}
+            components={{
+              table: ({ node: _n, ...props }) => (
+                <table
+                  className="my-2 w-full border-collapse border border-border"
+                  {...props}
+                />
+              ),
+              th: ({ node: _n, ...props }) => (
+                <th
+                  className="border border-border bg-muted/50 px-2 py-1 text-left text-sm font-medium"
+                  {...props}
+                />
+              ),
+              td: ({ node: _n, ...props }) => (
+                <td
+                  className="border border-border px-2 py-1 text-sm"
+                  {...props}
+                />
+              ),
+            }}
+          >
+            {content}
+          </ReactMarkdown>
+        )}
+      </div>
+      <div className="flex items-center gap-0.5">
+        {isUser && onEditSubmit != null && (
+          <Tooltip>
+            <TooltipTrigger asChild>
+              <Button
+                variant="ghost"
+                size="icon"
+                className="size-7 text-muted-foreground hover:text-foreground"
+                onClick={handleEditClick}
+                aria-label={t("button.edit", { ns: "common" })}
+              >
+                <FaPencilAlt className="size-3" />
+              </Button>
+            </TooltipTrigger>
+            <TooltipContent>
+              {t("button.edit", { ns: "common" })}
+            </TooltipContent>
+          </Tooltip>
+        )}
+        {isComplete && (
+          <Tooltip>
+            <TooltipTrigger asChild>
+              <Button
+                variant="ghost"
+                size="icon"
+                className="size-7 text-muted-foreground hover:text-foreground"
+                onClick={handleCopy}
+                disabled={!content?.trim()}
+                aria-label={t("button.copy", { ns: "common" })}
+              >
+                <FaCopy className="size-3" />
+              </Button>
+            </TooltipTrigger>
+            <TooltipContent>
+              {t("button.copy", { ns: "common" })}
+            </TooltipContent>
+          </Tooltip>
+        )}
+      </div>
+    </div>
+  );
+}
--- a/web/src/components/chat/ChatStartingState.tsx
+++ b/web/src/components/chat/ChatStartingState.tsx
@@ -0,0 +1,89 @@
+import { Button } from "@/components/ui/button";
+import { Input } from "@/components/ui/input";
+import { FaArrowUpLong } from "react-icons/fa6";
+import { useTranslation } from "react-i18next";
+import { useState } from "react";
+import type { StartingRequest } from "@/types/chat";
+
+type ChatStartingStateProps = {
+  onSendMessage: (message: string) => void;
+};
+
+export function ChatStartingState({ onSendMessage }: ChatStartingStateProps) {
+  const { t } = useTranslation(["views/chat"]);
+  const [input, setInput] = useState("");
+
+  const defaultRequests: StartingRequest[] = [
+    {
+      label: t("starting_requests.show_recent_events"),
+      prompt: t("starting_requests_prompts.show_recent_events"),
+    },
+    {
+      label: t("starting_requests.show_camera_status"),
+      prompt: t("starting_requests_prompts.show_camera_status"),
+    },
+  ];
+
+  const handleRequestClick = (prompt: string) => {
+    onSendMessage(prompt);
+  };
+
+  const handleSubmit = () => {
+    const text = input.trim();
+    if (!text) return;
+    onSendMessage(text);
+    setInput("");
+  };
+
+  const handleKeyDown = (e: React.KeyboardEvent<HTMLInputElement>) => {
+    if (e.key === "Enter" && !e.shiftKey) {
+      e.preventDefault();
+      handleSubmit();
+    }
+  };
+
+  return (
+    <div className="flex size-full flex-col items-center justify-center gap-6 p-8">
+      <div className="flex flex-col items-center gap-2">
+        <h1 className="text-4xl font-bold text-foreground">{t("title")}</h1>
+        <p className="text-muted-foreground">{t("subtitle")}</p>
+      </div>
+
+      <div className="flex w-full max-w-2xl flex-col items-center gap-4">
+        <p className="text-center text-sm text-muted-foreground">
+          {t("suggested_requests")}
+        </p>
+        <div className="flex w-full flex-wrap justify-center gap-2">
+          {defaultRequests.map((request, idx) => (
+            <Button
+              key={idx}
+              variant="outline"
+              className="max-w-sm text-sm"
+              onClick={() => handleRequestClick(request.prompt)}
+            >
+              {request.label}
+            </Button>
+          ))}
+        </div>
+      </div>
+
+      <div className="flex w-full max-w-2xl flex-row items-center gap-2 rounded-xl bg-secondary p-4">
+        <Input
+          className="h-12 w-full flex-1 border-transparent bg-transparent text-base shadow-none focus-visible:ring-0 dark:bg-transparent"
+          placeholder={t("placeholder")}
+          value={input}
+          onChange={(e) => setInput(e.target.value)}
+          onKeyDown={handleKeyDown}
+        />
+        <Button
+          variant="select"
+          className="size-10 shrink-0 rounded-full"
+          disabled={!input.trim()}
+          onClick={handleSubmit}
+        >
+          <FaArrowUpLong size="18" />
+        </Button>
+      </div>
+    </div>
+  );
+}
--- a/web/src/components/chat/ToolCallBubble.tsx
+++ b/web/src/components/chat/ToolCallBubble.tsx
@@ -0,0 +1,88 @@
+import { useState } from "react";
+import { useTranslation } from "react-i18next";
+import {
+  Collapsible,
+  CollapsibleContent,
+  CollapsibleTrigger,
+} from "@/components/ui/collapsible";
+import { Button } from "@/components/ui/button";
+import { cn } from "@/lib/utils";
+import { ChevronDown, ChevronRight } from "lucide-react";
+
+type ToolCallBubbleProps = {
+  name: string;
+  arguments?: Record<string, unknown>;
+  response?: string;
+  side: "left" | "right";
+};
+
+export function ToolCallBubble({
+  name,
+  arguments: args,
+  response,
+  side,
+}: ToolCallBubbleProps) {
+  const { t } = useTranslation(["views/chat"]);
+  const [open, setOpen] = useState(false);
+  const isLeft = side === "left";
+  const normalizedName = name
+    .replace(/_/g, " ")
+    .split(" ")
+    .map((word) => word.charAt(0).toUpperCase() + word.slice(1).toLowerCase())
+    .join(" ");
+
+  return (
+    <div
+      className={cn(
+        "rounded-lg px-3 py-2",
+        isLeft
+          ? "self-start bg-muted"
+          : "self-end bg-primary text-primary-foreground",
+      )}
+    >
+      <Collapsible open={open} onOpenChange={setOpen}>
+        <CollapsibleTrigger asChild>
+          <Button
+            variant="ghost"
+            size="sm"
+            className={cn(
+              "h-auto w-full min-w-0 justify-start gap-2 whitespace-normal p-0 text-left text-xs hover:bg-transparent",
+              !isLeft && "hover:text-primary-foreground",
+            )}
+          >
+            {open ? (
+              <ChevronDown size={12} className="shrink-0" />
+            ) : (
+              <ChevronRight size={12} className="shrink-0" />
+            )}
+            <span className="break-words font-medium">
+              {isLeft ? t("call") : t("result")} {normalizedName}
+            </span>
+          </Button>
+        </CollapsibleTrigger>
+        <CollapsibleContent>
+          <div className="mt-2 space-y-2">
+            {isLeft && args && Object.keys(args).length > 0 && (
+              <div className="text-xs">
+                <div className="font-medium text-muted-foreground">
+                  {t("arguments")}
+                </div>
+                <pre className="scrollbar-container mt-1 max-h-32 overflow-auto whitespace-pre-wrap break-words rounded bg-muted/50 p-2 text-[10px]">
+                  {JSON.stringify(args, null, 2)}
+                </pre>
+              </div>
+            )}
+            {!isLeft && response && response !== "" && (
+              <div className="text-xs">
+                <div className="font-medium opacity-80">{t("response")}</div>
+                <pre className="scrollbar-container mt-1 max-h-32 overflow-auto whitespace-pre-wrap break-words rounded bg-primary/20 p-2 text-[10px]">
+                  {response}
+                </pre>
+              </div>
+            )}
+          </div>
+        </CollapsibleContent>
+      </Collapsible>
+    </div>
+  );
+}
--- a/web/src/hooks/use-navigation.ts
+++ b/web/src/hooks/use-navigation.ts
@@ -6,7 +6,7 @@ import { isDesktop } from "react-device-detect";
 import { FaCompactDisc, FaVideo } from "react-icons/fa";
 import { IoSearch } from "react-icons/io5";
 import { LuConstruction } from "react-icons/lu";
-import { MdCategory, MdVideoLibrary } from "react-icons/md";
+import { MdCategory, MdChat, MdVideoLibrary } from "react-icons/md";
 import { TbFaceId } from "react-icons/tb";
 import useSWR from "swr";
 import { useIsAdmin } from "./use-is-admin";
@@ -18,6 +18,7 @@ export const ID_EXPORT = 4;
 export const ID_PLAYGROUND = 5;
 export const ID_FACE_LIBRARY = 6;
 export const ID_CLASSIFICATION = 7;
+export const ID_CHAT = 8;

 export default function useNavigation(
  variant: "primary" | "secondary" = "primary",
@@ -82,7 +83,15 @@ export default function useNavigation(
          url: "/classification",
          enabled: isDesktop && isAdmin,
        },
+        {
+          id: ID_CHAT,
+          variant,
+          icon: MdChat,
+          title: "menu.chat",
+          url: "/chat",
+          enabled: isDesktop && isAdmin && config?.genai?.model !== "none",
+        },
      ] as NavData[],
-    [config?.face_recognition?.enabled, variant, isAdmin],
+    [config?.face_recognition?.enabled, config?.genai?.model, variant, isAdmin],
  );
 }
--- a/web/src/lib/const.ts
+++ b/web/src/lib/const.ts
@@ -1,3 +1,6 @@
+/** ONNX embedding models that require local model downloads. GenAI providers are not in this list. */
+export const JINA_EMBEDDING_MODELS = ["jinav1", "jinav2"] as const;
+
 export const supportedLanguageKeys = [
  "en",
  "es",
--- a/web/src/pages/Chat.tsx
+++ b/web/src/pages/Chat.tsx
@@ -0,0 +1,226 @@
+import { Button } from "@/components/ui/button";
+import { Input } from "@/components/ui/input";
+import { FaArrowUpLong } from "react-icons/fa6";
+import { useTranslation } from "react-i18next";
+import { useState, useCallback } from "react";
+import axios from "axios";
+import { ChatEventThumbnailsRow } from "@/components/chat/ChatEventThumbnailsRow";
+import { MessageBubble } from "@/components/chat/ChatMessage";
+import { ToolCallBubble } from "@/components/chat/ToolCallBubble";
+import { ChatStartingState } from "@/components/chat/ChatStartingState";
+import type { ChatMessage } from "@/types/chat";
+import {
+  getEventIdsFromSearchObjectsToolCalls,
+  streamChatCompletion,
+} from "@/utils/chatUtil";
+
+export default function ChatPage() {
+  const { t } = useTranslation(["views/chat"]);
+  const [input, setInput] = useState("");
+  const [messages, setMessages] = useState<ChatMessage[]>([]);
+  const [isLoading, setIsLoading] = useState(false);
+  const [error, setError] = useState<string | null>(null);
+
+  const submitConversation = useCallback(
+    async (messagesToSend: ChatMessage[]) => {
+      if (isLoading) return;
+      const last = messagesToSend[messagesToSend.length - 1];
+      if (!last || last.role !== "user" || !last.content.trim()) return;
+
+      setError(null);
+      const assistantPlaceholder: ChatMessage = {
+        role: "assistant",
+        content: "",
+        toolCalls: undefined,
+      };
+      setMessages([...messagesToSend, assistantPlaceholder]);
+      setIsLoading(true);
+
+      const apiMessages = messagesToSend.map((m) => ({
+        role: m.role,
+        content: m.content,
+      }));
+
+      const baseURL = axios.defaults.baseURL ?? "";
+      const url = `${baseURL}chat/completion`;
+      const headers: Record<string, string> = {
+        "Content-Type": "application/json",
+        ...(axios.defaults.headers.common as Record<string, string>),
+      };
+
+      await streamChatCompletion(url, headers, apiMessages, {
+        updateMessages: (updater) => setMessages(updater),
+        onError: (message) => setError(message),
+        onDone: () => setIsLoading(false),
+        defaultErrorMessage: t("error"),
+      });
+    },
+    [isLoading, t],
+  );
+
+  const sendMessage = useCallback(() => {
+    const text = input.trim();
+    if (!text || isLoading) return;
+    setInput("");
+    submitConversation([...messages, { role: "user", content: text }]);
+  }, [input, isLoading, messages, submitConversation]);
+
+  const handleEditSubmit = useCallback(
+    (messageIndex: number, newContent: string) => {
+      const newList: ChatMessage[] = [
+        ...messages.slice(0, messageIndex),
+        { role: "user", content: newContent },
+      ];
+      submitConversation(newList);
+    },
+    [messages, submitConversation],
+  );
+
+  return (
+    <div className="flex size-full justify-center p-2">
+      <div className="flex size-full flex-col xl:w-[50%] 3xl:w-[35%]">
+        {messages.length === 0 ? (
+          <ChatStartingState
+            onSendMessage={(message) => {
+              setInput("");
+              submitConversation([{ role: "user", content: message }]);
+            }}
+          />
+        ) : (
+          <div className="scrollbar-container flex min-h-0 w-full flex-1 flex-col gap-2 overflow-y-auto">
+            {messages.map((msg, i) => {
+              const isStreamingPlaceholder =
+                i === messages.length - 1 &&
+                msg.role === "assistant" &&
+                isLoading &&
+                !msg.content?.trim() &&
+                !(msg.toolCalls && msg.toolCalls.length > 0);
+              if (isStreamingPlaceholder) {
+                return <div key={i} />;
+              }
+              return (
+                <div key={i} className="flex flex-col gap-2">
+                  {msg.role === "assistant" && msg.toolCalls && (
+                    <>
+                      {msg.toolCalls.map((tc, tcIdx) => (
+                        <div key={tcIdx} className="flex flex-col gap-2">
+                          <ToolCallBubble
+                            name={tc.name}
+                            arguments={tc.arguments}
+                            side="left"
+                          />
+                          {tc.response && (
+                            <ToolCallBubble
+                              name={tc.name}
+                              response={tc.response}
+                              side="right"
+                            />
+                          )}
+                        </div>
+                      ))}
+                    </>
+                  )}
+                  <MessageBubble
+                    role={msg.role}
+                    content={msg.content}
+                    messageIndex={i}
+                    onEditSubmit={
+                      msg.role === "user" ? handleEditSubmit : undefined
+                    }
+                    isComplete={
+                      msg.role === "user" ||
+                      !isLoading ||
+                      i < messages.length - 1
+                    }
+                  />
+                  {msg.role === "assistant" &&
+                    (() => {
+                      const isComplete = !isLoading || i < messages.length - 1;
+                      if (!isComplete) return null;
+                      const events = getEventIdsFromSearchObjectsToolCalls(
+                        msg.toolCalls,
+                      );
+                      return <ChatEventThumbnailsRow events={events} />;
+                    })()}
+                </div>
+              );
+            })}
+            {(() => {
+              const lastMsg = messages[messages.length - 1];
+              const showProcessing =
+                isLoading &&
+                lastMsg?.role === "assistant" &&
+                !lastMsg.content?.trim() &&
+                !(lastMsg.toolCalls && lastMsg.toolCalls.length > 0);
+              return showProcessing ? (
+                <div className="self-start rounded-lg bg-muted px-3 py-2 text-muted-foreground">
+                  {t("processing")}
+                </div>
+              ) : null;
+            })()}
+            {error && (
+              <p className="self-start text-sm text-destructive" role="alert">
+                {error}
+              </p>
+            )}
+          </div>
+        )}
+        {messages.length > 0 && (
+          <ChatEntry
+            input={input}
+            setInput={setInput}
+            sendMessage={sendMessage}
+            isLoading={isLoading}
+            placeholder={t("placeholder")}
+          />
+        )}
+      </div>
+    </div>
+  );
+}
+
+type ChatEntryProps = {
+  input: string;
+  setInput: (value: string) => void;
+  sendMessage: () => void;
+  isLoading: boolean;
+  placeholder: string;
+};
+
+function ChatEntry({
+  input,
+  setInput,
+  sendMessage,
+  isLoading,
+  placeholder,
+}: ChatEntryProps) {
+  const handleKeyDown = (e: React.KeyboardEvent<HTMLInputElement>) => {
+    if (e.key === "Enter" && !e.shiftKey) {
+      e.preventDefault();
+      sendMessage();
+    }
+  };
+
+  return (
+    <div className="flex w-full flex-col items-center justify-center rounded-xl bg-secondary p-2">
+      <div className="flex w-full flex-row items-center gap-2">
+        <Input
+          className="w-full flex-1 border-transparent bg-transparent shadow-none focus-visible:ring-0 dark:bg-transparent"
+          placeholder={placeholder}
+          value={input}
+          onChange={(e) => setInput(e.target.value)}
+          onKeyDown={handleKeyDown}
+          aria-busy={isLoading}
+        />
+        <Button
+          variant="select"
+          className="size-10 shrink-0 rounded-full"
+          disabled={!input.trim() || isLoading}
+          onClick={sendMessage}
+        >
+          <FaArrowUpLong size="16" />
+        </Button>
+      </div>
+    </div>
+  );
+}
--- a/web/src/pages/Explore.tsx
+++ b/web/src/pages/Explore.tsx
@@ -23,6 +23,7 @@ import { toast } from "sonner";
 import useSWR from "swr";
 import useSWRInfinite from "swr/infinite";
 import { useDocDomain } from "@/hooks/use-doc-domain";
+import { JINA_EMBEDDING_MODELS } from "@/lib/const";

 const API_LIMIT = 25;

@@ -293,7 +294,12 @@ export default function Explore() {
  const modelVersion = config?.semantic_search.model || "jinav1";
  const modelSize = config?.semantic_search.model_size || "small";

-  // Text model state
+  // GenAI providers have no local models to download
+  const isGenaiEmbeddings =
+    typeof modelVersion === "string" &&
+    !(JINA_EMBEDDING_MODELS as readonly string[]).includes(modelVersion);
+
+  // Text model state (skipped for GenAI - no local models)
  const { payload: textModelState } = useModelState(
    modelVersion === "jinav1"
      ? "jinaai/jina-clip-v1-text_model_fp16.onnx"
@@ -328,6 +334,10 @@ export default function Explore() {
  );

  const allModelsLoaded = useMemo(() => {
+    if (isGenaiEmbeddings) {
+      return true;
+    }
+
    return (
      textModelState === "downloaded" &&
      textTokenizerState === "downloaded" &&
@@ -335,6 +345,7 @@ export default function Explore() {
      visionFeatureExtractorState === "downloaded"
    );
  }, [
+    isGenaiEmbeddings,
    textModelState,
    textTokenizerState,
    visionModelState,
@@ -358,10 +369,11 @@ export default function Explore() {
    !defaultViewLoaded ||
    (config?.semantic_search.enabled &&
      (!reindexState ||
-        !textModelState ||
-        !textTokenizerState ||
-        !visionModelState ||
-        !visionFeatureExtractorState))
+        (!isGenaiEmbeddings &&
+          (!textModelState ||
+            !textTokenizerState ||
+            !visionModelState ||
+            !visionFeatureExtractorState))))
  ) {
    return (
      <ActivityIndicator className="absolute left-1/2 top-1/2 -translate-x-1/2 -translate-y-1/2" />
--- a/web/src/types/chat.ts
+++ b/web/src/types/chat.ts
@@ -0,0 +1,16 @@
+export type ToolCall = {
+  name: string;
+  arguments?: Record<string, unknown>;
+  response?: string;
+};
+
+export type ChatMessage = {
+  role: "user" | "assistant";
+  content: string;
+  toolCalls?: ToolCall[];
+};
+
+export type StartingRequest = {
+  label: string;
+  prompt: string;
+};
--- a/web/src/utils/chatUtil.ts
+++ b/web/src/utils/chatUtil.ts
@@ -0,0 +1,193 @@
+import type { ChatMessage, ToolCall } from "@/types/chat";
+
+export type StreamChatCallbacks = {
+  /** Update the messages array (e.g. pass to setState). */
+  updateMessages: (updater: (prev: ChatMessage[]) => ChatMessage[]) => void;
+  /** Called when the stream sends an error or fetch fails. */
+  onError: (message: string) => void;
+  /** Called when the stream finishes (success or error). */
+  onDone: () => void;
+  /** Message used when fetch throws and no server error is available. */
+  defaultErrorMessage?: string;
+};
+
+type StreamChunk =
+  | { type: "error"; error: string }
+  | { type: "tool_calls"; tool_calls: ToolCall[] }
+  | { type: "content"; delta: string };
+
+/**
+ * POST to chat/completion with stream: true, parse NDJSON stream, and invoke
+ * callbacks so the caller can update UI (e.g. React state).
+ */
+export async function streamChatCompletion(
+  url: string,
+  headers: Record<string, string>,
+  apiMessages: { role: string; content: string }[],
+  callbacks: StreamChatCallbacks,
+): Promise<void> {
+  const {
+    updateMessages,
+    onError,
+    onDone,
+    defaultErrorMessage = "Something went wrong. Please try again.",
+  } = callbacks;
+
+  try {
+    const res = await fetch(url, {
+      method: "POST",
+      headers,
+      body: JSON.stringify({ messages: apiMessages, stream: true }),
+    });
+
+    if (!res.ok) {
+      const errBody = await res.json().catch(() => ({}));
+      const message = (errBody as { error?: string }).error ?? res.statusText;
+      onError(message);
+      onDone();
+      return;
+    }
+
+    const reader = res.body?.getReader();
+    const decoder = new TextDecoder();
+    if (!reader) {
+      onError("No response body");
+      onDone();
+      return;
+    }
+
+    let buffer = "";
+    let hadStreamError = false;
+
+    const applyChunk = (data: StreamChunk) => {
+      if (data.type === "error") {
+        onError(data.error);
+        updateMessages((prev) =>
+          prev.filter((m) => !(m.role === "assistant" && m.content === "")),
+        );
+        return "break";
+      }
+      if (data.type === "tool_calls" && data.tool_calls?.length) {
+        updateMessages((prev) => {
+          const next = [...prev];
+          const lastMsg = next[next.length - 1];
+          if (lastMsg?.role === "assistant")
+            next[next.length - 1] = {
+              ...lastMsg,
+              toolCalls: data.tool_calls,
+            };
+          return next;
+        });
+        return "continue";
+      }
+      if (data.type === "content" && data.delta !== undefined) {
+        updateMessages((prev) => {
+          const next = [...prev];
+          const lastMsg = next[next.length - 1];
+          if (lastMsg?.role === "assistant")
+            next[next.length - 1] = {
+              ...lastMsg,
+              content: lastMsg.content + data.delta,
+            };
+          return next;
+        });
+        return "continue";
+      }
+      return "continue";
+    };
+
+    for (;;) {
+      const { done, value } = await reader.read();
+      if (done) break;
+      buffer += decoder.decode(value, { stream: true });
+      const lines = buffer.split("\n");
+      buffer = lines.pop() ?? "";
+      for (const line of lines) {
+        const trimmed = line.trim();
+        if (!trimmed) continue;
+        try {
+          const data = JSON.parse(trimmed) as StreamChunk & { type: string };
+          const result = applyChunk(data as StreamChunk);
+          if (result === "break") {
+            hadStreamError = true;
+            break;
+          }
+        } catch {
+          // skip malformed JSON lines
+        }
+      }
+      if (hadStreamError) break;
+    }
+
+    // Flush remaining buffer
+    if (!hadStreamError && buffer.trim()) {
+      try {
+        const data = JSON.parse(buffer.trim()) as StreamChunk & {
+          type: string;
+          delta?: string;
+        };
+        if (data.type === "content" && data.delta !== undefined) {
+          updateMessages((prev) => {
+            const next = [...prev];
+            const lastMsg = next[next.length - 1];
+            if (lastMsg?.role === "assistant")
+              next[next.length - 1] = {
+                ...lastMsg,
+                content: lastMsg.content + data.delta!,
+              };
+            return next;
+          });
+        }
+      } catch {
+        // ignore final malformed chunk
+      }
+    }
+
+    if (!hadStreamError) {
+      updateMessages((prev) => {
+        const next = [...prev];
+        const lastMsg = next[next.length - 1];
+        if (lastMsg?.role === "assistant" && lastMsg.content === "")
+          next[next.length - 1] = { ...lastMsg, content: " " };
+        return next;
+      });
+    }
+  } catch {
+    onError(defaultErrorMessage);
+    updateMessages((prev) =>
+      prev.filter((m) => !(m.role === "assistant" && m.content === "")),
+    );
+  } finally {
+    onDone();
+  }
+}
+
+/**
+ * Parse search_objects tool call response(s) into event ids for thumbnails.
+ */
+export function getEventIdsFromSearchObjectsToolCalls(
+  toolCalls: ToolCall[] | undefined,
+): { id: string }[] {
+  if (!toolCalls?.length) return [];
+  const results: { id: string }[] = [];
+  for (const tc of toolCalls) {
+    if (tc.name !== "search_objects" || !tc.response?.trim()) continue;
+    try {
+      const parsed = JSON.parse(tc.response) as unknown;
+      if (!Array.isArray(parsed)) continue;
+      for (const item of parsed) {
+        if (
+          item &&
+          typeof item === "object" &&
+          "id" in item &&
+          typeof (item as { id: unknown }).id === "string"
+        ) {
+          results.push({ id: (item as { id: string }).id });
+        }
+      }
+    } catch {
+      // ignore parse errors
+    }
+  }
+  return results;
+}
--- a/web/src/utils/i18n.ts
+++ b/web/src/utils/i18n.ts
@@ -46,6 +46,7 @@ i18n
      "components/icons",
      "components/player",
      "views/events",
+      "views/chat",
      "views/explore",
      "views/live",
      "views/settings",
--- a/web/src/views/settings/UiSettingsView.tsx
+++ b/web/src/views/settings/UiSettingsView.tsx
@@ -196,7 +196,7 @@ export default function UiSettingsView() {
                </SelectTrigger>
                <SelectContent>
                  <SelectGroup>
-                    {[1, 2, 3, 5, 8, 10, 12, 15].map((timeout) => (
+                    {[1, 2, 3, 4, 5, 6, 7, 8, 9, 10].map((timeout) => (
                      <SelectItem
                        key={timeout}
                        className="cursor-pointer"
Author	SHA1	Message	Date
Nicolas Mowen	88bad3423b	Set model	2026-02-26 15:28:58 -07:00
Nicolas Mowen	f3cda9020b	Don't require download check	2026-02-26 14:33:08 -07:00
Nicolas Mowen	0c333ec28a	Fix sending images	2026-02-26 14:33:08 -07:00
Nicolas Mowen	de986c7430	undo	2026-02-26 14:33:08 -07:00
Nicolas Mowen	dd2d7aca19	Basic docs	2026-02-26 14:33:08 -07:00
Nicolas Mowen	3f1bf1ae12	Add support for embedding via genai	2026-02-26 14:33:08 -07:00
Nicolas Mowen	d6e8cad32f	Add embed API support	2026-02-26 14:33:07 -07:00
Nicolas Mowen	699d5ffa28	Support GenAI for embeddings	2026-02-26 14:32:33 -07:00
Nicolas Mowen	f400e91ede	Add a starting state for chat	2026-02-26 08:38:59 -07:00
Nicolas Mowen	3bac4b15ae	Add thumbnail images to object results	2026-02-26 08:38:59 -07:00
Nicolas Mowen	b2c424ad73	Add support for markdown tables	2026-02-26 08:38:59 -07:00
Nicolas Mowen	c18846ac62	Fix loading	2026-02-26 08:38:59 -07:00
Nicolas Mowen	b65ae76f0c	Cleanup UI bubbles	2026-02-26 08:38:59 -07:00
Nicolas Mowen	5faf5e0d84	Cleanup UI and prompt	2026-02-26 08:38:59 -07:00
Nicolas Mowen	6837b9c89a	Cleanup	2026-02-26 08:38:59 -07:00
Nicolas Mowen	f04df4a144	Add sub label to event tool filtering	2026-02-26 08:38:59 -07:00
Nicolas Mowen	e42f70eeec	Implement message editing	2026-02-26 08:38:59 -07:00
Nicolas Mowen	e7b2b919d5	Improve default behavior	2026-02-26 08:38:59 -07:00
Nicolas Mowen	c68b7c9f46	Improvements to UI	2026-02-26 08:38:59 -07:00
Nicolas Mowen	8184ec5c8f	Add copy button	2026-02-26 08:38:59 -07:00
Nicolas Mowen	ef448a7f7c	Fix tool calling	2026-02-26 08:38:58 -07:00
Nicolas Mowen	f841ccdb63	Undo	2026-02-26 08:38:58 -07:00
Nicolas Mowen	4b6228acd9	Full streaming support	2026-02-26 08:38:58 -07:00
Nicolas Mowen	0b8d1ce568	Support streaming	2026-02-26 08:38:58 -07:00
Nicolas Mowen	9ad7a2639f	Improve UI handling	2026-02-26 08:38:58 -07:00
Nicolas Mowen	089c2c1018	Add title	2026-02-26 08:38:58 -07:00
Nicolas Mowen	3e97f9e985	Show tool calls separately from message	2026-02-26 08:38:58 -07:00
Nicolas Mowen	eb9f16b4fa	More time parsing improvements	2026-02-26 08:38:58 -07:00
Nicolas Mowen	45c6be47d2	Reduce fields in response	2026-02-26 08:38:58 -07:00
Nicolas Mowen	5a6c62a844	Adjust timing format	2026-02-26 08:38:58 -07:00
Nicolas Mowen	f29fbe14ca	Improvements	2026-02-26 08:38:58 -07:00
Nicolas Mowen	cc941ab2db	Add markdown	2026-02-26 08:38:58 -07:00
Nicolas Mowen	56b3ebe791	processing	2026-02-26 08:38:58 -07:00
Nicolas Mowen	6fdfe22f8c	Add chat history	2026-02-26 08:38:58 -07:00
Nicolas Mowen	0cf713985f	Add basic chat page with entry	2026-02-26 08:38:58 -07:00
Nicolas Mowen	dc39d2f0ef	Set model in llama.cpp config	2026-02-26 08:38:52 -07:00
Nicolas Mowen	e6387dac05	Fix import issues	2026-02-26 08:38:52 -07:00
Nicolas Mowen	c870ebea37	Cleanup	2026-02-26 08:38:52 -07:00
Nicolas Mowen	56a1a0f5e3	Support getting client via manager	2026-02-26 08:38:52 -07:00
Nicolas Mowen	67a245c8ef	Convert to roles list	2026-02-26 08:38:52 -07:00
Nicolas Mowen	a072600c94	Add config migration	2026-02-26 08:38:52 -07:00
Nicolas Mowen	b603678b26	GenAI client manager	2026-02-26 08:38:52 -07:00
Nicolas Mowen	8793650c2f	Fix frame time access	2026-02-26 08:38:42 -07:00
Nicolas Mowen	9c8dd9a6ba	Adapt to new Gemini format	2026-02-25 09:19:56 -07:00
nulledy	507b495b90	ffmpeg Preview Segment Optimization for "high" and "very_high" (#21996 ) * Introduce qmax parameter for ffmpeg preview encoding Added PREVIEW_QMAX_PARAM to control ffmpeg encoding quality. * formatting * Fix spacing in qmax parameters for preview quality	2026-02-25 09:02:08 -07:00
nulledy	3525f32bc2	Allow API Events to be Detections or Alerts, depending on the Event Label (#21923 ) * - API created events will be alerts OR detections, depending on the event label, defaulting to alerts - Indefinite API events will extend the recording segment until those events are ended - API event start time is the actual start time, instead of having a pre-buffer of record.event_pre_capture * Instead of checking for indefinite events on a camera before deciding if we should end the segment, only update last_detection_time and last_alert_time if frame_time is greater, which should have the same effect * Add the ability to set a pre_capture number of seconds when creating a manual event via the API. Default behavior unchanged * Remove unnecessary _publish_segment_start() call * Formatting * handle last_alert_time or last_detection_time being None when checking them against the frame_time * comment manual_info["label"].split(": ")[0] for clarity	2026-02-25 09:02:08 -07:00
Josh Hawkins	ac142449f1	Improve jsmpeg player websocket handling (#21943 ) * improve jsmpeg player websocket handling prevent websocket console messages from appearing when player is destroyed * reformat files after ruff upgrade	2026-02-25 09:02:08 -07:00
FL42	47b89a1d60	feat: add X-Frame-Time when returning snapshot (#21932 ) Co-authored-by: Florent MORICONI <170678386+fmcloudconsulting@users.noreply.github.com>	2026-02-25 09:02:08 -07:00
Eric Work	cdcf56092c	Add networking options for configuring listening ports (#21779 )	2026-02-25 09:02:08 -07:00
Nicolas Mowen	08ee2e21de	Add live context tool to LLM (#21754 ) * Add live context tool * Improve handling of images in request * Improve prompt caching	2026-02-25 09:02:08 -07:00
Nicolas Mowen	9ab4dd4538	Update to ROCm 7.2.0 (#21753 ) * Update to ROCm 7.2.0 * ROCm now works properly with JinaV1 * Arcface has compilation error	2026-02-25 09:02:08 -07:00
Josh Hawkins	fe5441349b	Offline preview image (#21752 ) * use latest preview frame for latest image when camera is offline * remove frame extraction logic * tests * frontend * add description to api endpoint	2026-02-25 09:02:08 -07:00
Nicolas Mowen	a4b1cc3a54	Implement LLM Chat API with tool calling support (#21731 ) * Implement initial tools definiton APIs * Add initial chat completion API with tool support * Implement other providers * Cleanup	2026-02-25 09:02:08 -07:00
John Shaw	99e25661b2	Remove parents in remove_empty_directories (#21726 ) The original implementation did a full directory tree walk to find and remove empty directories, so this implementation should remove the parents as well, like the original did.	2026-02-25 09:02:08 -07:00
Nicolas Mowen	20360db2c9	Implement llama.cpp GenAI Provider (#21690 ) * Implement llama.cpp GenAI Provider * Add docs * Update links * Fix broken mqtt links * Fix more broken anchors	2026-02-25 09:02:08 -07:00
John Shaw	3826d72c2a	Optimize empty directory cleanup for recordings (#21695 ) The previous empty directory cleanup did a full recursive directory walk, which can be extremely slow. This new implementation only removes directories which have a chance of being empty due to a recent file deletion.	2026-02-25 09:02:08 -07:00
Nicolas Mowen	3d5757c640	Refactor Time-Lapse Export (#21668 ) * refactor time lapse creation to be a separate API call with ability to pass arbitrary ffmpeg args * Add CPU fallback	2026-02-25 09:02:08 -07:00
Eugeny Tulupov	86100fde6f	Update go2rtc to v1.9.13 (#21648 ) Co-authored-by: Eugeny Tulupov <eugeny.tulupov@spirent.com>	2026-02-25 09:02:08 -07:00
Josh Hawkins	28b1195a79	Fix incorrect counting in sync_recordings (#21626 )	2026-02-25 09:02:08 -07:00
Josh Hawkins	b6db38bd4e	use same logging pattern in sync_recordings as the other sync functions (#21625 )	2026-02-25 09:02:08 -07:00
Josh Hawkins	92c6b8e484	Media sync API refactor and UI (#21542 ) * generic job infrastructure * types and dispatcher changes for jobs * save data in memory only for completed jobs * implement media sync job and endpoints * change logs to debug * websocket hook and types * frontend * i18n * docs tweaks * endpoint descriptions * tweak docs	2026-02-25 09:02:07 -07:00
Josh Hawkins	9381f26352	Add media sync API endpoint (#21526 ) * add media cleanup functions * add endpoint * remove scheduled sync recordings from cleanup * move to utils dir * tweak import * remove sync_recordings and add config migrator * remove sync_recordings * docs * remove key * clean up docs * docs fix * docs tweak	2026-02-25 09:02:07 -07:00
Nicolas Mowen	e0180005be	Add API to handle deleting recordings (#21520 ) * Add recording delete API * Re-organize recordings apis * Fix import * Consolidate query types	2026-02-25 09:02:07 -07:00
Nicolas Mowen	2041798702	Exports Improvements (#21521 ) * Add images to case folder view * Add ability to select case in export dialog * Add to mobile review too	2026-02-25 09:02:07 -07:00
Nicolas Mowen	3d23b5de30	Add support for GPU and NPU temperatures (#21495 ) * Add rockchip temps * Add support for GPU and NPU temperatures in the frontend * Add support for Nvidia temperature * Improve separation * Adjust graph scaling	2026-02-25 09:02:07 -07:00
Andrew Roberts	209bb44518	Camera-specific hwaccel settings for timelapse exports (correct base) (#21386 ) * added hwaccel_args to camera.record.export config struct * populate camera.record.export.hwaccel_args with a cascade up to camera then global if 'auto' * use new hwaccel args in export * added documentation for camera-specific hwaccel export * fix c/p error * missed an import * fleshed out the docs and comments a bit * ruff lint * separated out the tips in the doc * fix documentation * fix and simplify reference config doc	2026-02-25 09:02:07 -07:00
Nicolas Mowen	88462cd6c3	Refactor temperature reporting for detectors and implement Hailo temp reading (#21395 ) * Add Hailo temperature retrieval * Refactor `get_hailo_temps()` to use ctxmanager * Show Hailo temps in system UI * Move hailo_platform import to get_hailo_temps * Refactor temperatures calculations to use within detector block * Adjust webUI to handle new location --------- Co-authored-by: tigattack <10629864+tigattack@users.noreply.github.com>	2026-02-25 09:02:07 -07:00
Nicolas Mowen	c2cc23861a	Export filter UI (#21322 ) * Get started on export filters * implement basic filter * Implement filtering and adjust api * Improve filter handling * Improve navigation * Cleanup * handle scrolling	2026-02-25 09:02:07 -07:00
Josh Hawkins	2b46084260	Camera connection quality indicator (#21297 ) * add camera connection quality metrics and indicator * formatting * move stall calcs to watchdog * clean up * change watchdog to 1s and separately track time for ffmpeg retry_interval * implement status caching to reduce message volume	2026-02-25 09:02:07 -07:00
Nicolas Mowen	67466f215c	Case management UI (#21299 ) * Refactor export cards to match existing cards in other UI pages * Show cases separately from exports * Add proper filtering and display of cases * Add ability to edit and select cases for exports * Cleanup typing * Hide if no unassigned * Cleanup hiding logic * fix scrolling * Improve layout	2026-02-25 09:02:07 -07:00
Josh Hawkins	e011424947	refactor vainfo to search for first GPU (#21296 ) use existing LibvaGpuSelector to pick appropritate libva device	2026-02-25 09:02:07 -07:00
Nicolas Mowen	a1a0051dd7	implement case management for export apis (#21295 )	2026-02-25 09:02:07 -07:00
Nicolas Mowen	ff331060c3	Create scaffolding for case management (#21293 )	2026-02-25 09:02:07 -07:00
Nicolas Mowen	7aab1f02ec	Update version	2026-02-25 09:02:07 -07:00