dryfire

Git-native regression testing for LLM agent tool loops. Assert on the ordered trajectory — which tools your agent called, in what order, with what arguments — not the final prose.

dryfire runs the tool-calling loop itself with your tools mocked deterministically, so every run is free, offline, and identical. It's local-first — no account, no server, no database — and the exit code is the API, so it drops straight into CI.

pip install dryfire
dryfire init      # scaffold a keyless example
dryfire run       # green, offline, no API key

  ✓ reports_the_weather                 2 turns   0 tok   —   0.0s

1 cases   1 passed   0 failed   —   0.0s

Where to start

• Get started — from zero to a green suite to CI, with an annotated example. Start here.
• Core: Mocks · Cassettes · CI
• v0.3 — judgment & comparison: LLM-as-judge · Flakiness · Compare
• Reference: SPEC · Architecture · How it compares

What it is not

dryfire is a pre-deployment unit test, and deliberately not more: not production observability, not a hosted dashboard or account product, not dataset/labeling management, not an agent framework. It stays a zero-infra local CLI.