Version Control Built for AI Engineering
Treating prompts like production code with immutable commit history, automated dual-model evaluations, and local-first developer tooling.
Our Mission
Prompts govern production application behavior, but engineering teams still manage them as raw strings scattered across codebases. When a prompt change breaks output schemas or causes regressions, there is rarely a commit log showing what changed, why, or how to roll back.
Git for Prompts brings the discipline of code versioning to prompts: immutable commits, visual diffs, automated regression tests, and zero-lockin local storage.
Architecture & Design Principles
1. Local-First Engineering
Developers should never be blocked by network connectivity or cloud outages. The CLI operates on a local SQLite database directly inside your project (.gitforprompts/). You can create prompts, commit snapshots, and run evaluations completely offline in your terminal.
2. Transactional Concurrency
When syncing prompt versions across teams, cloud mutations use database transaction locks to guarantee version numbers never collide or overwrite existing history.
3. Zero Key Custody
Your LLM API keys belong to you. Local evaluations run using your workstation environment variables. Cloud API credentials are stored solely as non-reversible SHA-256 hashes for authenticated prompt delivery.
4. Autonomous Agent Readiness
Engineered for both human developers and autonomous coding agents. Features full support for the Model Context Protocol (MCP), structured schemas, standard llms.txt endpoints, and spec-compliant Markdown content negotiation.
100% Open Source
Git for Prompts is fully open-source under the MIT license. We welcome contributions from developers, researchers, and AI engineers across the world.