Connect local Ollama

Prepare the server

1. Install Ollama and consult the official documentation.
2. Choose a model with tools from the library. Download it, for example with ollama pull llama3.1, if that model is still available. This suggestion exists in the runtime; confirm compatibility and resources on your device.
3. Keep the Ollama service running. Use ollama serve only if no server is already running.
4. Confirm the exact installed name with ollama list.

Configure Browser Mate

1. Select Ollama (local) in Settings → LLM Model.
2. Enter the installed identifier, including the tag when required.
3. Use http://localhost:11434/v1/chat/completions for the service on the default port. Adjust host/port only if your server uses other values. The runtime normalizes Ollama endpoints to the OpenAI-compatible route.
4. Leave the key empty: Ollama is the only menu option that does not require a key.
5. Save and test the connection. Test text first; text-only models do not accept screenshots.

Keep the server local

Do not expose the service to the Internet to fix an error. If there is an origin refusal, check Ollama’s allowed-origin options and authorize only the origin required by your extension installation; avoid global wildcards.

If it doesn’t work

Check the running service, port, model name, available memory, firewall and origin permissions. A model that chats well may not be able to run tools reliably. Check the capability report instead of assuming the server guarantees support.

Local does not mean zero auxiliary communications

Inference on this endpoint stays on your computer, but the extension may make auxiliary requests to public catalogs and exchange rates. It is not a promise of completely offline operation. See Privacy.