LM Studio Setup

LM Studio runs language models in a separate app or service. Summit connects to its OpenAI-compatible API for transcript cleanup, summaries, and chat. You can run it on this Mac or another computer you manage.

If you want a model inside Summit without maintaining a server, use Local AI Setup instead.

Start the LM Studio server

1. Install LM Studio on the computer that will run the model. Check its current system requirements for that computer.
2. Download a text instruction or chat model that fits the computer’s memory. Test that it can answer a short prompt in LM Studio.
3. Open LM Studio’s Developer tab and turn on Start server. See the official server guide.
4. Note the server address, port, and model identifier. The usual same-Mac address is http://localhost:1234.
5. If server authentication is enabled, create or obtain a token from LM Studio’s server settings.

Keep LM Studio or its server service running while Summit uses it. Downloading a model does not by itself start an API server.

Connect Summit

1. Open Settings → AI Models.
2. Select Remote (Cloud or Self-Hosted).
3. Set Provider to Other.
4. Enter http://localhost:1234/v1 as the Base URL if LM Studio is on this Mac and uses the default port.
5. Enter the server token in API key if required. Otherwise leave it blank.
6. Click Refresh Models and select the model. If discovery is unavailable, enter the identifier shown by LM Studio manually.
7. Send a short question in Summit Chat or use Reprocess Summary on a test recording.

Summit uses LM Studio’s OpenAI-compatible /v1 endpoints. Do not substitute LM Studio’s native /api/v1 path. Settings take effect as you change them.

Use a different computer

localhost always means the computer running Summit. To reach LM Studio on another computer:

1. Configure LM Studio to listen on a network interface reachable from your Mac.
2. Use the server computer’s address and configured port in Summit, keeping the /v1 suffix.
3. Configure authentication and network access for that server.

See LM Studio’s server start options for binding and authentication guidance. The selected server receives the text sent by Summit, so consider who can access that machine and its logs.

Match context to the model

Leave Summit’s Chat context limit at 0 initially. If you change it, choose a value the loaded model and LM Studio configuration can support. Larger context requires more memory on the server.

A smaller model or shorter context can help with memory errors. For incomplete answers, check whether the transcript and prompt provide the needed facts before changing generation settings.

Troubleshooting

Symptom	What to check
Connection refused	LM Studio’s server is running, and Summit’s host and port match it.
Authentication error	The API key field contains the token expected by the server, or is blank when authentication is disabled.
No models listed	A suitable model is downloaded and available to the server. Try entering its exact identifier.
Refresh works but summaries fail	Confirm the selected model can generate text and has enough memory and context. Inspect the server’s error output.
Works on the server computer only	The server is listening beyond localhost and the Mac can reach its address and port.
Chat or summaries are slow	Check server memory pressure and other workloads; try a smaller model and compare the same test recording.

Privacy boundaries

Summit’s AI requests send text to the configured server, including relevant transcript and meeting context. They do not upload the recorded audio files. A same-Mac localhost endpoint keeps that connection on this Mac; a network endpoint sends the text to the other computer.

This does not disable iCloud sync, export, or the MCP server. Review those separately when deciding where your meeting data may be available.