Screenshots From the README of goruck/home-generative-agent.
About
Home Generative Agent is an AI conversation agent that controls Home Assistant and writes automations from plain language. It adds camera analysis with optional face recognition, long-term semantic memory kept in a PostgreSQL database with pgvector, and an anomaly engine called Sentinel that raises proactive alerts. Safety decisions stay deterministic, with the language model advising but never acting on them. It works with cloud model providers or fully local ones, though the integration itself needs the vendor's cloud connection. Camera notifications have cooldown settings per camera and for the whole house, and Ring event recordings can be analyzed through ring-mqtt. It is set up in the UI, adds image and sensor entities, and can post repair notices.
Automatic summary of the README and code of goruck/home-generative-agent. Something wrong, or are you the maintainer and want your own text here? Get in touch
Supports
Uses the vendor's cloud Needs an internet connection and an account with the vendor.
From the Marketplace Installed through the Marketplace, formerly HACS.
Actively maintained Its author released or pushed a change in the last six months.
Provides
- Sensors
Setting it up
Install it from the Marketplace and restart Home Assistant, then add it under Settings, Devices and services.
After setup
- Face recognition API URL (optional)
- InstructionsInstruct how the LLM should respond. This can be a template.
- Camera description language (optional)Language for VLM camera image descriptions, e.g. 'Czech'. Leave empty for English. The internal 'Scene unchanged.' reply always stays in English.
- Additional camera analysis instructions (optional)Extra instructions appended to the VLM camera analysis prompt, e.g. 'Ignore cars in the driveway'. They never replace the built-in rules.
- Control Home Assistant
- Video analyzer mode
- Enable perceptual-hash frame filter (dHash)Skip visually near-identical frames before VLM analysis. Warning: enabling this removes inter-frame motion context that the summary model uses to narrate movement. Only enable if a static scene generates excessive duplicates and you accept that motion continuity may be lost.
- Analyze Ring event recordings (ring-mqtt, Ring Protect)When a ring-mqtt event_select entity publishes a signed recordingUrl with a new eventId, download the event's MP4, extract up to 8 frames with ffmpeg and analyze those instead of the retained interval snapshot, which on battery cameras can be up to 10 minutes older than the event. Needs a Ring Protect subscription (cameras without a recordingUrl keep using snapshots) and the ffmpeg binary the official Home Assistant image ships.
- Motion sensor → camera overrides (optional)One override per line: binary_sensor.X_motion: camera.Y — use this when automatic resolution picks the wrong camera. Leave blank to rely on device-based and name-based resolution.
- Video Analyzer Notify Service (optional)
- Camera notification cooldown (seconds)0 turns it off. After a camera notification sounds, later ones from that camera within this many seconds (600 at most) replace the same notification quietly with the latest caption and image, instead of sounding again. With face recognition on, the first unknown person in that period still sounds. Works only with a single companion-app notify service (notify.mobile_app_...).
- House notification cooldown (seconds)0 turns it off. One event seen by several cameras makes each camera notify. After a camera notification sounds, a notification from any camera within this many seconds (60 at most) still posts its own card, but quietly, and that camera then stays quiet for its camera notification cooldown, or this many seconds if longer, counted from that sound. With the camera cooldown off, every camera, including the one that sounded, is quiet for this many seconds. With face recognition on, the first unknown person in that period still sounds. Start around 15 seconds: each camera's video is analyzed separately, so one event can reach the cameras 10 to 30 seconds apart. Works only with a single companion-app notify service (notify.mobile_app_...); verified on iOS only.
- Enable face recognition
- Context management method
- Max tokens in chat context
Texts from the code of goruck/home-generative-agent.
Troubleshooting
- Repair noticesNotices it can show in Settings, Repairs:Home Generative Agent is running without a database, Critical action PIN is not protecting spoken commands, Sentinel's unknown-person rules cannot fire, Memory store holds another embedding model's vectors
From the README
A Home Assistant integration that brings a generative AI agent into your smart home. Talk to your home, create automations in plain English, analyze camera footage, and get proactive alerts — all powered by your choice of cloud or local LLMs. HGA is a single integration that gives you conversational control over every HA entity, camera understanding with face recognition, long-term semantic memory, and the Sentinel anomaly engine.
Creating an automation in plain English — the agent writes the YAML, registers it, and it shows up in the HA automation editor.
Shows in Vulpo
Chart card, Apple Watch.Information
- Installs
- 331 sharing analytics
- GitHub stars
- 367
- Connection
- Uses the vendor's cloud
- Upkeep
- Updated in the last 6 months
- Version
- v3.44.10,
- Category
- AI and voice, AI models
- Kind
- Marketplace integration
- Needs
- Needs the Marketplace (HACS on 2026.10 and older)
- Domain
home_generative_agent- Issues
- Report an issue














