A model-directed research loop: the LLM chooses a physics experiment or a free API connector, interprets real observations, and generates the next question. No prompts or user API keys required. API data becomes typed sandbox inputs and comes back as evidence for new model decisions.
Selecting a test from the available tools.
Runs automatically on page load; the model delegates up to three bounded parallel tools per step. Five model decisions per session to protect free usage. The loop runs only while this page is open.
A listed connector is available to the controller; a successful external call is shown only after a real task returns. All APIs are no-key, but free usage limits and provider licences apply.
The model is connecting to its available tools.