I Built an Investigation-First Debugger for AI Agents — TraceMotive v0.3

I Built an Investigation-First Debugger for AI Agents — TraceMotive v0.3

TraceMotive is an open-source, local-first debugging tool designed to analyze AI agent executions. It helps identify the first behavioral divergence that can be supported by structural evidence when comparing two executions: a reference run and a changed run. The tool focuses on producing deterministic findings such as changes in tool input, output, error observations, added or removed tools, or altered execution subtrees.

It also detects context-only changes like model updates, request parameter modifications, or trace status changes.

However, TraceMotive avoids making assumptions or guessing the root cause. For instance, it may surface uncertain findings when evidence is ambiguous or unavailable. The new comparison interface prioritizes the investigation state and first supported investigation point, providing additional behavioral observations and context uncertainty.

TraceMotive keeps data privacy in mind, maintaining loopback-only data persistence, redacting content before transport, and not collecting any analytics or telemetry. It does not claim to prove causality but rather identify the first supported behavioral divergence to guide further investigation. The next steps involve broader real-world integration validation and gathering feedback on the usefulness of the investigation workflow for real agent failures. TraceMotive is available on GitHub and can be installed via pip install tracemotive == 0.3.0.