Watch It Work

See an AI agent's every step.

Most agents are a black box: a prompt goes in, an answer comes out. This one shows the work. Every step, every tool call, the reviewer gate that blocks a wrong answer and sends it back, and a running token and cost counter. Watch the recorded demo, or run your own with your key.

The task

Task give the agent something with a number in it

Agent model who runs the task (the reviewer gate stays cheap)

Model optional

API key Anthropic · used for this run only, never stored

The agent must use a calculator and a word-count tool instead of doing the math itself. The reviewer checks the final answer against what the tools returned, and blocks it if a number doesn't match.

The run

Press Play demo to watch a recorded run, including the reviewer catching a wrong word count and forcing a retry. Or paste a key and run your own.

Want this working in your business?

Book a 30-min call →