PandaProbe

Trace agent runs and validate reusable fixes for failures

Website

Meet PandaProbe

PandaProbe is a developer platform for tracing AI agents, evaluating their work, and testing repairs. It records model calls, tool use, subagent activity, timing, and outputs to build a structured history of a run. Evaluations examine both the final result and the steps taken, helping teams identify failures that may not appear as ordinary execution errors.

Its repair harness proposes scoped rules and tests them through developer-controlled replays or trials. Rules that improve measured outcomes can be kept in a persistent workspace for future tasks. A command-line interface lets agents inspect traces and evaluations directly, while framework integrations add this workflow around an existing orchestration system.

Discussion

Tried it? Share a useful thought or ask a question.

Product details

Status
Launched
Listed

Explore related topics