Connect an engine
Use this guide to add a model server and verify a first response. GOAT uses one active engine at a time; you can keep several saved configurations.
Before you begin
Start the server using its own controls. Note its root URL, authentication requirement and available models. Use Models and engines for the concepts and the engine reference for the exact contract.
Add the connection
- Open Settings → Engine and add an engine.
- Choose the matching preset, or Custom…. Presets fill conventional URLs and choose capability metadata behavior; they do not install a server.
- Enter the actual root URL. GOAT appends the API paths. Avoid a URL that redirects to another address.
- Supply the API key if required. Choose Test, review the result and save.
- Select the engine and a model, then send a short text request. Verify that the response completes before trying images or tools.
The engine list is stored in ~/.goat/config/engines.json under the default GOAT Home. Credentials use an owner-only file, not the chat database. See storage.
If connection fails
Check that the server is running at the saved address. A 401 or 403 usually requires reviewing credentials. A working browser address may still be a dashboard rather than the model API.
JUDAS can block even a local engine in Block connections mode. LAN services may also need macOS Local Network permission. Changing JUDAS policy disconnects existing integrations; reconnect deliberately afterward.
If basic chat works but tools or reasoning do not, inspect the model/server capability and template settings. Do not infer support from a model name. The compatibility table distinguishes implementation coverage from live qualification.