The asynchronous Git commit polisher. Write code, fire your standard commit, and let the ghost daemon rewrite it perfectly using AI in the background.
Run git commit -m "fix". The intercepting post-commit hook finishes instantly.
A background process orphans itself to the OS, completely freeing your terminal.
The daemon asks the LLM, builds a replacement from the recorded commit, and atomically advances the ref only if it has not moved.
If you push early, the exact ref updates are queued until the safe replacement finishes.
Track your time saved and AI-polished commit counts. A native dashboard inside your IDE gives you tangible insight into how much repetitive work the daemon has eliminated.
Never lose the "Original Draft" intent. The interactive timeline highlights which commits were ghost-authored by AI, revealing the exact models, generation latency, and your original rough thoughts.
Native VS Code and IntelliJ IDEA plugins query the CLI for external app state, logs, and safe controls without creating project files.
Synchronous CLI tools (like aicommits) and deep agents (Copilot) are great if you want to micromanage your commits. But they force you to wait 10-30 seconds to generate and review. Git AI is built for flow-state developers: it stays entirely out of your way and polishes asynchronously.
No. Git AI records the exact commit SHA and ref, then uses an atomic compare-and-swap update. If the ref moved or any step fails, Git and workspace files stay unchanged.
By default, the pre-push hook queues the exact ref updates while polishing. If replay cannot be proven safe or credentials are unavailable, Git AI stops and asks you to retry manually.
Git AI natively supports the OpenAI, Anthropic, Gemini, and DeepSeek APIs. It also supports locally hosted models like Ollama.
No, the core engine is a standalone Go binary that operates purely via standard Git hooks. It works universally.
Install the CLI via script or grab an IDE plugin.
curl -fsSL https://raw.githubusercontent.com/daidi/git-ai/main/install.sh | bash