offtrack provides git diff-style comparison for AI agent executions. It records validated "golden" trajectories, identifies the exact first step where a new run diverges, and can gate CI pipelines on real behavioral regressions. Built for LLM agent developers who need reproducible evaluation and observability.
In the LLM eval & observability space, offtrack takes a focused approach. It focuses on detecting when AI agent behavior deviates from validated golden trajectories during development and CI. It is built as an open-source project for developers. offtrack is open source under the MIT license. offtrack is available on the command line.
offtrack first shipped in 2026. Among its 4 catalogued features are Golden Trajectories, Divergence Detection, and CI Gating.
Summary written by a language model from the project’s public pages.
What PulseGate has recorded for this listing
Same category — not a similarity match