If you already run Home Assistant and have a PC with a decent graphics card, you can give Assist, Home Assistant's built-in voice assistant, a local AI brain. The Ollama integration connects Home Assistant to an Ollama server on your network and adds conversation agents that can chat and, if you allow it, control your devices. This guide follows Home Assistant's and Ollama's documentation; Forge PC hasn't tested particular models with Home Assistant, so we haven't recommended one on performance.
What you need
Home Assistant with an administrator account.
A PC running Ollama on the same network. Ollama runs on Windows, macOS and Linux. Our guide to running AI models locally with Ollama and LM Studio covers installing it.
A model that supports tools if you want device control. Home Assistant says only tool-capable models can use its AI tools, and points to the "tools" filter in the Ollama library.
Step 1: Make Ollama reachable on your network
By default Ollama only listens on the PC it runs on. Home Assistant's docs send you to Ollama's FAQ section "How can I expose Ollama on my network?", which means setting the OLLAMA_HOST environment variable to 0.0.0.0:11434. On Windows:
Quit Ollama from the taskbar.
Open Settings, search for environment variables and choose Edit environment variables for your account.
Add OLLAMA_HOST with the value 0.0.0.0:11434.
Click OK and start Ollama again.
Note the PC's local IP address (run ipconfig). Our guide to sharing Ollama on your local network has more detail. Only do this on a network you trust.
Step 2: Add the Ollama integration
In Home Assistant, go to Settings > Devices & services.
Select Add Integration in the bottom right and choose Ollama.
Enter the URL of your Ollama server, such as http://192.168.1.10:11434. Leave the API key empty unless your server requires one.
Home Assistant's Ollama options include:
Option | What it does (per Home Assistant) |
|---|
Model | The Ollama model to use. Downloaded models are listed first. The default is qwen3:4b-instruct. If the model isn't on your server, Home Assistant downloads it when you save. |
Instructions | How the AI should respond to your requests |
AI tools | Sets of tools the agent can use, such as Device control. Only tool-capable models can use them. |
Context window size | Tokens the model can take as input, including your exposed entities. Default 8192, minimum 2048. Larger supports more entities but uses more memory. |
Max history messages | Messages kept per conversation. Default 20; 0 keeps all. |
Keep alive | Seconds Ollama keeps the model in memory after a message. -1 keeps it loaded (the default); 0 unloads it straight away. |
Think before responding | Lets supported models reason before answering. Slower, but can give better answers. Off by default. |
The Keep alive default of -1 means the model stays in your GPU's memory. That's good for quick answers, but if the same PC is your gaming PC, consider a shorter value. Our guide to keeping Ollama models loaded or unloading them explains how keep-alive works.
Step 4: Let it control devices (carefully)
To let the assistant switch lights or read sensors, Home Assistant's docs suggest:
Expose only the entities you want the assistant to use. Home Assistant warns that local models are often smaller than cloud ones, and smaller models are more likely to make mistakes.
Add a conversation agent named something like "Ollama devices", choose a tool-capable model and select Device control under AI tools.
Optionally add a second agent, such as "Ollama chat", with the same model and no AI tools, because some smaller models chat less well with tools enabled.
Create an Assist assistant for each conversation agent.
Home Assistant also notes that your sentence triggers are checked first: if one matches, it answers and the AI agent never sees the request.
Choosing hardware
A bigger context window and a bigger model both need more VRAM. Home Assistant says the context window includes your exposed entities, so more entities need a larger value. See how much VRAM you need, LLM context length and VRAM and the best GPU for local LLMs. If you'd rather not leave a gaming PC on all day, our best mini PCs guide is a starting point for a dedicated box.
Our verdict
Home Assistant's Ollama integration is the simplest way to put a local model behind Assist: expose Ollama on your network with OLLAMA_HOST, add the integration with your server's URL, pick a tool-capable model, and expose only the entities you want it to control.