I built PIT because managing 50+ LLM prompts in Google Docs broke my production AI 3 times.

PIT is "git for prompts" — semantic version control that understands LLM behavior, not just text changes.

Key features: • Bisect: Binary search to find which prompt version broke • Replay: Test same input across all versions instantly • Query: Find versions with "success_rate > 0.9" • Patches: Share prompt improvements like code reviews • Hooks: CI/CD for prompts (pre-commit security scans)

332 tests. Production-ready. Open source.

pip install prompt-pit

Would love feedback from anyone managing prompts in production!