Connect Ollama for AI debate and collaboration
AI Arena Lite can use models served by Ollama alongside other compatible servers. Start with one available model and a short prompt, then explore different roles and workspaces.
1. Start Ollama and make a model available
Install Ollama using its official Windows instructions, then download a model suitable for your hardware through Ollama. Its Windows application runs in the background and exposes a local API at http://localhost:11434.
You also need AI Arena Lite for Windows. Lite connects to the server; Ollama supplies and runs the model.
2. Find the model in Lite
Open Models in AI Lab. Lite discovers running local servers, including Ollama, and shows their model catalogues together. Use Settings → Provider connection → Find servers to refresh discovery.
If you use a different address or a remote server, open Advanced and use Connect to address with that server's address and any required token. Keep remote access limited to the people and devices you intend to allow.
3. Assign the model to your agents
Select a model and use Assign to. Enable Default to supply roles that have no explicit model, or assign models to individual roles. If Lite offers a provider-supported load action, wait for the server to confirm its state.
Each assignment keeps the model's serving connection. A model with the same name on another server remains a separate choice.
4. Choose a workspace
- AI Lab: Set a topic and cast in Match Setup, then choose 1 Turn to inspect the opening response.
- AI Collaborate: Confirm the team's model routing, enter a short prompt, and follow the team's contributions and final answer.
- Agent: Select a project folder for a software task and review command proposals before approving execution.
A useful first collaboration prompt is: “Draft a simple plan for organising a small community event. Have the team check the budget, accessibility, and likely problems.”
If the server or model does not appear
Check that Ollama is running and the model is available in Ollama itself. Refresh discovery in Lite. A server being online, a model being loaded, and a model being assigned are separate states.
See the model routing guide or provider troubleshooting guide for the next step.