What it does Frigate turns a pile of IP cameras into a proper surveillance system with realtime object detection. It records continuously, detects motion, identifies objects with TensorFlow, and integrates tightly with Home Assistant. Everything runs locally; no subscription, no cloud dependency.

The interesting bit The project is deliberately stingy with compute. It uses lightweight motion detection to decide where to run full object detection, and farms that work out to separate processes. The README is explicit that you want a GPU or AI accelerator like Google Coral—CPUs need not apply for serious setups.

Key highlights

• Tight Home Assistant integration via a custom component
• MQTT support for wiring into other automation systems
• RTSP re-streaming to avoid drowning your cameras in connection requests
• WebRTC and MSE for low-latency live viewing
• Built-in mask/zone editor and multi-camera scrubbing UI
• Retention policies tied to detected objects, not just blind time-based deletion

Caveats

• The README notes AI accelerators are “highly recommended”; CPU-only performance is unclear and likely disappointing at scale
• Trademark restrictions apply: the Frigate name and logo are owned by Frigate, Inc. and not covered under the MIT license

Verdict Home Assistant users with Coral accelerators or spare GPUs should look here first. If you’re running a single camera on a Raspberry Pi without extras, expect to read the docs carefully—or look elsewhere.