By recording agent execution at non-deterministic boundaries, Chronicle enables regression testing of LLM agents without re-running expensive model calls, catching bugs that would otherwise be hard to reproduce due to non-determinism.
Chronicle is a testing tool that records LLM agent runs at decision points and replays them to catch regressions. It captures non-deterministic model outputs and tool interactions as immutable snapshots, then selectively replays recorded boundaries while executing new code live—turning past failures into reproducible regression tests that run in CI/CD pipelines.