How to Verify AI Agent Tool Calls
Last updated: 2026-10-03
In short
After an agent calls a tool, Zambo stores a verifiable receipt for the recorded execution. A reviewer can open a live receipt at its /run/ URL or paste its identifier into the verifier. The receipt checks the saved record, not whether the model was right.
Verification steps
- The agent calls a tool through Zambo.
- Zambo stores a verifiable receipt for the recorded call and result.
- Open a live example receipt, or paste its UUID into the receipt verifier.
The route shape
A human receipt URL uses the /run/<receipt-id> route. For a real example, open https://zambo.dev/run/b08cfe77-6a52-47e7-9ade-b01fbcd5e543. The verifier at /verify/ accepts the receipt UUID.
What the check means
A valid integrity check supports the claim that the saved record matches its commitment. It does not prove that the model reasoned correctly or that an external system completed an unobserved side effect. Read the execution receipt definition for the evidence boundary.
Frequently asked questions
How to Verify AI Agent Tool Calls
Open the tool call's public /run/ receipt URL, or paste its receipt identifier into the Zambo verifier. Review the recorded result and integrity checks.
Can I use a receipt URL as the receipt identifier?
Open the URL directly, or copy its UUID and paste the UUID into the verifier.
Does a verified tool call prove the answer was correct?
No. It checks the saved record and its integrity. Output correctness requires a separate review.