Run your first AI debate with LM Studio
Use a local model server to give AI Arena Lite's agents a place to run their models. You can start with one model shared by the cast, then experiment with different assignments.
1. Prepare Lite and LM Studio
Install AI Arena Lite for Windows. In LM Studio, download a model that fits your hardware, load it, and start the local server from its Developer area. LM Studio also supports starting the server with lms server start. See LM Studio's server instructions.
2. Discover the server
Open Models in Lite's AI Lab. Lite looks for running local servers and combines their model catalogues. If the server does not appear, use Settings → Provider connection → Find servers.
For a remote server or an uncommon port, expand Advanced, enter the server address and optional token, then choose Connect to address. Lite detects the supported APIs.
3. Load and assign a model
Select the model in Models. If it is unloaded and a Load model action is available, use that action and wait for confirmation. Under Assign to, enable Default for roles without an explicit assignment, or select individual agents and the Narrator.
A model being loaded does not automatically mean it is assigned to a role. Check the assignments before starting. Different roles can use different models when your servers and hardware support them.
4. Set a topic and run one turn
Open Match Setup and choose your cast and a topic. For example: “Should a small team prioritise new features or reliability?” Give agents different perspectives, then run 1 Turn in AI Lab.
Review the response before choosing Auto Chat. You can add an Operator message to request evidence, challenge an assumption, or bring the discussion back to the topic.
If the first turn is blocked
- Confirm the LM Studio server is running and reachable from this computer.
- Check that the selected model is loaded and assigned to the active roles.
- Read Lite's readiness or status message before retrying.
For model context settings, routing, and recovery, follow the full first-run walkthrough and provider troubleshooting guide.
Local model speed and memory needs depend on the model and the server's hardware settings. Treat an AI debate as material to inspect, not as proof that its answers are correct.