New: Cookbooks and AI ExplanationsStep-by-Step recipes to solve problems connected to Roadmaps and Cheat Sheets. Need more details? Use AI buttons for structured and simple explanations with concrete examples throughout the whole platform.Take a look
Your agent finished and handed back an answer that reads like it did exactly what you asked, but there is no record of which tool call actually produced it.
What you'll have at the end
A structured log line for every tool call, with its name, arguments, and result, that lets you replay a run step by step
You need
An agent that already calls one or more tools during a run, through your own code sitting between the model's tool-call request and the function that actually runs it.
Not covered
Wiring these lines into a hosted tracing platform, and deciding whether a failed call should retry automatically instead of only being recorded; this recipe only makes sure a written, redacted line exists for every call, whatever it returns.
Leans on
One bad tool call crashes the whole agent run
If a failing call should retry automatically instead of only being written to the log, add a wrapper that catches the failure and decides whether to try again.
Should your agent ask before it deletes, sends, or pays?
If the call this log is recording can delete, send, or pay for something, decide whether it needs a person's go-ahead before it fires, not only a record after it did.
Checked 15 Aug 2026
Part of the AI Agents cookbook