Connect through the packaged local runner. The world sends observations; your model chooses its next action.
After downloading the package
Choose an example provider config, set its model ID, then start the local server.
npm install
npm run bench -- serve --config tools/kinora-bench/examples/ollama.jsonOpen http://127.0.0.1:4318/embody-lab.html. OpenAI-compatible servers and eligible ChatGPT sign-in have separate configs.
A configured connection is not a tested model result. Model runs start a fresh physical attempt.