git has GitHub; agent-trace has TracesHub. An open-source CLI records which lines of your code were written by AI — which model, via which tool, and the prompt behind them. Deterministic, local-first, one ledger across Cursor, Claude Code, and other agents. Coding agents, when making new changes, use the CLI tool to surface context hidden behind lines of code.

Hey! I built this because git blame has quietly stopped telling the truth. It says I wrote a line that Cursor or Claude Code actually wrote, and the prompt that explains why is buried in some old chat. So agent-trace records per-line AI attribution - which model, which tool, which conversation - at commit time. Local-first: hooks never touch the network; nothing leaves your machine unless you push, git-style. The part that excited me most: agents can read that record too. I asked one to clean up "magic number" constants, and it stopped mid-task after pulling the original prompt. The numbers came from a support ticket, so they weren’t safe to delete. That felt like the future. The CLI and service are open source / self-hostable, and TracesHub is the hosted option (think git → GitHub), free to use. I’m here all day, so ask me anything and I welcome your hard feedbacks including why this shouldn’t exist :)

finally something that solves a real headache for me - chasing down which AI wrote what across cursor and claude code. local-first and open source is a great call.

TracesHub

Line-level AI attribution for your codebase

Line-level AI attribution for your codebase