Frigate 0.18: UI setup, AI chat and motion search

This walkthrough covers seven features in Frigate 0.18, with examples of browser configuration and local AI integration. The presenter shows global and per-camera settings, including masks, zones and object detection. Most configuration changes apply without restarting, though enabling MQTT still requires a restart.

Profiles store named camera presets such as home, away and night mode. The example assigns a night profile to a camera and enables bear tracking as an override. For generative AI, the presenter connects llama.cpp through a base URL and selects Gemma 4E4B. Installation belongs to an earlier video; this walkthrough starts with the model service already available.

The chat interface exposes an AI agent with tools for searching events and managing cameras. Both demonstrated requests have problems: live context returns no detected objects instead of a frame description, and disabling recording returns an unknown feature error. The presenter questions whether the model caused the first result.

A separate embedding provider handles semantic search because, according to the presenter, the selected Gemma model lacks vector embeddings. That setup requires a Frigate restart. The remaining sections show region-based motion search, a motion heat map and named cases that group related video exports.