When a bug shows up in code that three different AI agents touched last week, the first question is "which AI wrote this?" Origin is a new open-source CLI tool built to answer that.
Origin hooks into Claude Code, Cursor, Gemini, and Codex. It tags every commit with metadata: which agent generated the code, what prompt was used, which model ran, and what it cost. Run origin blame on any file and each line shows [AI] or [HU] - like standard git blame, but distinguishing human-written code from AI-generated code.
All data lives in git notes, so there's no external server, no account, and no lock-in. It works fully offline and stores everything inside your existing git repository.
Who Needs This
Teams where multiple developers use different AI coding tools are the obvious audience. If one person uses Claude Code, another uses Cursor, and a third uses Codex, tracking which AI produced which code is currently impossible without digging through individual tool logs (if those even exist).
The cost-tracking angle is also practical. Knowing that a particular refactoring session cost $14 in Opus inference and produced 200 lines that later needed manual fixes is useful data for deciding how and when to use AI assistance.
The tool is MIT-licensed and available on GitHub. It's early-stage - the team built it after realizing nobody on their own team could trace AI-authored code back to its source when debugging. Given how fast AI-assisted development is spreading, the "who wrote this" problem is only going to get more common.