Frigate NVR Updates: GenAI Integration and Camera Management

Transcript

Good morning, this is your Frigate NVR update for March 9th, 2026.

NickM-27 merged support for GenAI embeddings in semantic search, adding the ability to use generative AI models for object recognition and search functionality. The 346-line addition includes new embedding APIs and updated documentation for the semantic search configuration.

Josh Hawkins delivered several significant updates. He merged the ability to delete cameras through both API endpoints and the frontend interface, providing a 578-line addition that includes proper cleanup utilities and internationalization support. His environment variable fixes resolve timing bugs in the FRIGATE_PLUS integration and add comprehensive test coverage.

Hawkins also pushed through frontend improvements, introducing an OptionalFieldWidget for configuration forms and optimizing motion preview rendering to reduce DOM overhead. The thumbnail encoding logic received fixes to ensure proper re-encoding regardless of quality settings.

The enrichment configuration system was refactored to use single-subscriber architecture with topic-based routing, simplifying how face recognition, license plate detection, and bird classification modules handle configuration updates.

Additional fixes include improved preview retrieval error handling, corrected go2rtc configuration processing, and enhanced license plate detection that now publishes box coordinates via MQTT.

NickM-27's ONNX dependency updates bring full support for NVIDIA's RTX 50-series GPUs through updated TensorRT requirements.

What's next: The GenAI integration opens possibilities for more sophisticated object recognition, while the camera management improvements suggest continued focus on user experience enhancements.

That's your Frigate NVR briefing. We'll be back tomorrow with more development updates.