Watch AI think

Type a prompt and watch the real pipeline behind the answer — tokenization, the model's live decision to use tools, real Wikipedia search or calculation calls, and the final answer streaming in token by token. Nothing here is simulated.

Built for education, not production use — it runs on free-tier and small local models by design, not production-grade infrastructure. It doesn't have guaranteed real-time or perfectly precise data, and can occasionally misstate a fact or approximate a number instead of using a tool result. Every real tool call and its actual output is shown in the step log below, so you can always check what really happened.