Start the server
- On the Mac, open the Local API server or Developer page and choose the models to serve.
- Start the listener and copy the address, model identifier, and authentication details shown by Noema.
- For another device, enable local-network serving and pair it with the provided QR flow or configure the displayed credentials.
- Keep Noema running and the Mac awake while a client is using it.
Supported endpoint families
| Endpoint | Use |
|---|---|
| /v1/models | Discover models exposed by the server. |
| /v1/chat/completions | OpenAI-compatible chat clients. |
| /v1/responses | Clients using the Responses request format. |
| /v1/messages | Clients using the Anthropic Messages request format. |
| /v1/messages/count_tokens | A token-count estimate for Messages clients. |
Pairing, trusted hosts, and request log
- Use authentication and the Authorized devices controls to manage paired clients. Revoke access for a device you no longer use.
- Add a custom DNS hostname to Trusted hosts when needed. The server rejects untrusted Host names; enter the hostname it actually receives.
- Use the request log to inspect incoming requests, model selection, and errors. Logging can be controlled in server settings.
- Keep the listener on a network you control. Off-Grid and server availability controls still apply.
Troubleshooting
For authentication errors, check the selected credential and paired device. For an unknown model, copy an identifier returned by /v1/models. For a rejected host, check the hostname and Trusted hosts setting.
If a client cannot connect, check that the server is running, the Mac is awake, the port and local-network settings match, and both devices can reach the selected address. A model loading or memory error requires fixing the host model setup.