Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Power Local LLM
Local AI Agents & Tool Use

Home Generative Agent Review: AI Agent for Home Assistant

·6 min readBy Hans Kuepper · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI

Home Generative Agent (version 3.44.10, released on 23 October 2026) is a free, MIT-licensed Home Assistant integration that adds an LLM agent to Assist and works with a local Ollama server or cloud models. It needs a running Home Assistant, a PostgreSQL database with pgvector and at least one model provider.

Key Takeaways

  • Maintenance: the latest stable release, version 3.44.10, is from 23 October 2026 and the main branch was updated on 1 November 2026, so no stale-project warning applies.
  • Model: none is included; Ollama, OpenAI-compatible servers, OpenAI, Gemini and Anthropic are supported.
  • Cost: the software is free under the MIT licence; only a cloud provider, if you choose one, bills you.
  • Setup: several steps across a database app, HACS and a model provider rather than one installer.
  • Maker: a single developer (GitHub user goruck), not a company.

📍 In One Sentence

Home Generative Agent is an open-source AI agent for Home Assistant that lets you talk to your home, create automations in plain language and ask about camera images.

💬 In Plain Terms

You install it from HACS into your existing Home Assistant, connect a language model such as a local Ollama server, and then pick it as the conversation agent in Assist.

📌Note: This review is based on the GitHub repository, its documentation and the Home Assistant Community thread, checked on 1 November 2026. PromptQuorum has not installed, run or benchmarked Home Generative Agent, so reliability, speed and the quality of the automations it writes are not assessed.

Read in:
🇺🇸en🇩🇪de🇫🇷fr🇯🇵ja🇨🇳zh🇪🇸es🇧🇷pt🇸🇦ar🇰🇷ko

Home Generative Agent (HGA) is a free, MIT-licensed custom integration for Home Assistant that adds a tool-using AI agent to the Assist conversation interface. Built on LangGraph, it can control the devices you expose to it, write automations from plain-language requests, describe camera images and keep long-term memory. It uses the model provider you choose, from a local Ollama server to OpenAI, Gemini or Anthropic. This review is based on the project's GitHub repository, its Home Assistant Community thread and GitHub Releases; PromptQuorum has not installed or run it. It is listed in the PromptQuorum directory.

Home Generative Agent Review: AI Agent for Home Assistant

What Is Home Generative Agent?

Home Generative Agent is an AI agent for Home Assistant that acts on your home instead of only passing your words to a language model. Per the README it is built on LangGraph and uses tools to control entities, query history, watch cameras and write real automations.

It is a Home Assistant custom integration installed through HACS, not a standalone app. Its Sentinel engine watches for security and safety issues with fixed rules, and the README says the language model only advises there and never switches anything; this is the maker's description and was not checked. The opening post of the Community thread still names an older version than GitHub, so this review follows GitHub.

Get It

Home Generative Agent is installed through HACS into an existing Home Assistant; there is no standalone download.

Option
How to get it
HACS (recommended)Search for it in HACS, which lists it in its default repository, or use the Open-in-HACS button in the README; then restart Home Assistant and add the integration
Manual installDownload home_generative_agent.zip from GitHub Releases and copy it into Home Assistant as the installation guide describes
DatabaseThe PostgreSQL with pgvector app on Home Assistant OS or Supervised, or your own PostgreSQL server
WebsiteHome Assistant Community thread, the project's homepage on GitHub
DocumentationGuides in the repository: installation, configuration, Sentinel, cameras and architecture
Privacy policyNone found in the repository or the Community thread
Source codegithub.com/goruck/home-generative-agent, MIT licence

Version reviewed: 3.44.10, released on 23 October 2026 per GitHub Releases; a pre-release, 3.45.0-beta.5, also exists and was not reviewed. The Home Assistant steps happen in its web interface, but pulling Ollama models uses a terminal, and the whole chain of services takes more effort than a single install.

Requirements

Home Assistant 2026.8.0 or newer, HACS, a PostgreSQL database with pgvector and at least one model provider are required.

Home Assistant

What the sources say:
2026.8.0 is the minimum (from HGA 3.43.3); 2026.9 and 2026.10 are listed as supported

Database

What the sources say:
PostgreSQL with pgvector; without it only in-memory storage works, so long-term memory, face recognition and the baseline anomaly detector are unavailable

Model provider

What the sources say:
OpenAI, Gemini, Anthropic, Ollama or any OpenAI-compatible server

Local models (optional)

What the sources say:
The guide lists Ollama models such as gpt-oss, qwen3:8b and qwen3-vl:8b; it gives no minimum RAM or VRAM

Face recognition (optional)

What the sources say:
An external face service that you set up separately

Voice (optional)

What the sources say:
The OpenAI speech APIs or a local OpenAI-compatible server such as Speaches; about 2 GB of VRAM is recommended for the large faster-whisper model

Which hardware suits which model is not stated beyond this; it depends on the model server, not on Home Generative Agent.

How to Get Started

The steps follow the README and the installation guide; PromptQuorum has not run them.

  1. 1
    Set up the database
    Why it matters: Install the PostgreSQL with pgvector app on Home Assistant OS or Supervised, or point the integration at your own PostgreSQL server.
  2. 2
    Install from HACS and restart
    Why it matters: Install Home Generative Agent from HACS, restart Home Assistant, then add the integration under Settings → Devices & Services.
  3. 3
    Add a model provider
    Why it matters: Click + Model Provider and choose Edge (Ollama or an OpenAI-compatible server) or Cloud (OpenAI, Gemini or Anthropic).
  4. 4
    Run Setup
    Why it matters: Click + Setup and pick Basic, which enables all features with recommended defaults, or Advanced to configure each one.
  5. 5
    Choose it as the conversation agent
    Why it matters: Select Home Generative Agent under Settings → Voice assistants, then expose the entities you want it to see.

Features Confirmed by the Sources

Every item below comes from the README or the documentation; none has been tested here.

  • Conversation. Natural-language control and status questions for exposed entities, with streaming replies.
  • Automations. Writes and registers automations from chat; entity IDs and service calls are checked against Home Assistant first, and an optional Critical Action PIN holds automations that would unlock a door or open a garage.
  • Cameras. Describes camera images and runs motion-triggered analysis; the README names Axis, Ring via ring-mqtt, Reolink and UniFi Protect, plus any camera that exposes a motion entity.
  • Sentinel. Rule-based anomaly detection and a security audit of the Home Assistant setup, with phone alerts and optional language-model triage.
  • Memory and tools. Long-term semantic memory in pgvector, and extra tools such as web search through MCP servers or a Home Assistant LLM API.
  • Voice. Its own speech-to-text and text-to-speech providers for Assist pipelines, using OpenAI or a local server.

Data Flow and Privacy

Where your data goes depends on the model provider you choose: the README says Home Generative Agent can run fully locally with Ollama or an OpenAI-compatible server, with cloud providers as the alternative.

With a cloud provider, your prompts and the camera images you ask it to analyse reach that provider under its own policy. The README adds that Sentinel and the audit tool redact MAC and IP addresses and client hostnames before anything is shown to a language model. Face recognition uses a separate face service, whose storage this review did not examine.

📌Note: These are the maker's statements, not audit results. PromptQuorum has not inspected the code or the network traffic.

Who Should Use It

  • Home Assistant users with a spare GPU machine who want a conversational agent that stays on their network through Ollama.
  • People who want automations drafted from plain-language requests and will review each one in the automation editor.
  • Owners of supported cameras who want questions answered about what a camera sees, using a vision model they host.

What We Could Not Verify

  • Reliability. How often the agent finds the right entity or writes a correct automation is untested.
  • Safety claims. The Critical Action PIN and the statement that the model never switches anything in Sentinel are the maker's descriptions.
  • Local-model quality. The guide recommends models but publishes no benchmarks for tool use or vision on local hardware.
  • Data handling. The face service and cloud-provider data flows were not inspected.

Competitors and Alternatives

Home Assistant Assist

Platforms:
Home Assistant (web, Android, iOS)
Price / license:
Free / Apache-2.0
Key difference:
Built-in voice pipeline; LLM agents are optional add-ons
Articles about Home Assistant Assist (1)

nanobot

Platforms:
Windows, macOS, Linux
Price / license:
Free / MIT
Key difference:
General self-hosted agent, not tied to Home Assistant
Articles about nanobot (5)

PicoClaw

Platforms:
Linux, macOS, Windows, Android
Price / license:
Free / MIT
Key difference:
Small general-purpose agent, not tied to Home Assistant
Articles about PicoClaw (1)

Only Home Assistant Assist is in the same smart-home segment; nanobot and PicoClaw are general agents. Confirm each app's current price, license and platforms on its own listing before choosing.

Compare in directory

Frequently Asked Questions

Does Home Generative Agent replace Home Assistant Assist?

No. You select it as the conversation agent inside Assist, and it can add its own speech-to-text and text-to-speech providers to an Assist pipeline.

Why does the agent say it cannot find a device?

It only sees entities exposed to Assist and finds them by name and area, so the README advises exposing the entity and giving it a clear name and area.

Does it work on Home Assistant Container or Core?

The bundled database app needs Home Assistant OS or Supervised, so on Container or Core the installation guide points you to your own PostgreSQL server with pgvector.

Verdict

Home Generative Agent is a free, MIT-licensed way to give an existing Home Assistant a tool-using AI agent that can run against a local Ollama server or a cloud model, and it is actively maintained. Against that, setup spans several services, many features depend on optional components such as the face service, and PromptQuorum has not tested reliability, privacy behaviour or local-model quality. It suits Home Assistant users who are comfortable with HACS and a model server; anyone using it for locks or safety automations should check every generated automation themselves.

Sources

← Back to Power Local LLM

MCP