Agentdiff

agentdiff — Know What Every AI Agent Wrote

  Git-native AI code provenance. Records which AI agent wrote which line, signs each attribution with ed25519, and stores it in your own git history — across Claude Code, Cursor, Copilot, Codex, Windsurf, OpenCode, and Gemini.

agentdiff is an open-source, git-native AI code provenance tool. It hooks into every major AI coding agent — Claude Code, Cursor, Codex, Copilot, Windsurf, OpenCode, Gemini — and writes a permanent, commit-scoped attribution record to your repository. Each record captures the agent name, model, prompt excerpt, and exact line ranges, signed with ed25519. All of it queryable from the CLI, no server required.

Watch the launch demo to see the attribution workflow end-to-end. Full guide: AI code provenance.

Where agentdiff fits

"AI code provenance" gets confused with adjacent tools that answer different questions. agentdiff is the source-line authorship layer — it records which agent wrote which line, as signed evidence. The others stack with it rather than compete:

Provenance is recorded at authorship time and signed, so it is durable evidence — not a post-hoc estimate. See AI code provenance tools for the full category map.

Install

Requirements: Python 3.7+ on PATH, Git 2.20+, Rust 1.85+ (installed automatically if missing)

Other install methods

Quick Start

That's it. From here every commit is attributed to whichever agent (or human) wrote it.
Note: agentdiff configure installs capture scripts globally, but capture only fires in repos where agentdiff init has been run (the .git/agentdiff/ directory must exist). Running configure on its own does not track any repo — you must also run agentdiff init inside each repo you want to track.
AGENTS.md: agentdiff configure also writes (or updates) an ##…