Launched this week
Gigai Chronicle
git blame says who — chronicle why says what was asked
8 followers
git blame says who — chronicle why says what was asked
8 followers
Git records what changed. Gigai Chronicle records what was asked — the prompts, the AI's responses, the code it produced — as a plain-text history inside your git repo. chronicle why shows the prompt behind any change. Local-first, never calls a model. Free & open source.
this is honestly a really cool idea, like git blame but for prompts makes total sense. one thing that would help me a lot though, basically some kind of diff view for the prompts themselves, so when i rewrite a prompt between commits i can see exactly what changed instead of just the final version. would make iterating on prompts way easier to track
@blahgeputc93775 Great news — half of this already ships, and you just found the half that doesn't. 😄
What works today: prompts saved to the library are fully versioned and diffable:
And in the VS Code extension, prompt version history opens in the native diff editor — real side-by-side, not a custom widget.
What you're really asking for is diffing the captured prompts — the ones you actually typed to the agent, where you rephrased the same intent 3 times before it worked. Right now chronicle why shows you those prompts but won't diff them against each other. You're right that this is where the learning is: the delta between "attempt 2" and "attempt 3" is the thing that made it click.
Adding it — the diff engine and the checkpoint chain are both already there, it's mostly plumbing:
chronicle why <file> --evolution → diffs between consecutive prompts on that file
chronicle diff <evtA> <evtB> → diff any two captured prompts
Right-click two prompts in the sidebar → Compare
Tracking it here if you want to follow along. Thanks — this is the best kind of feedback. 🙏
Love that it lives as plain-text inside the repo instead of hiding behind some web dashboard. chronicle why being a single command that just works locally without calling a model is exactly the kind of zero-friction design I wish more dev tools had.
@muammersare1wg Thank you — that's the exact thing we optimized for. The moment you need a dashboard, you need an account, a server, and someone's trust. Plain text in the repo means git clone is the sync protocol and there's nothing to run.
chronicle why staying local and model-free is deliberate too: it's not summarizing anything with an LLM, it's just diffing two git checkpoints that capture already took. So the answer is deterministic, instant, and works offline forever. Cheap to build, impossible to enshittify.
Appreciate you noticing the friction part specifically. 🙏