Every agent. Every prompt. Every line. In your git repo.
The AI coding history layer for developers and teams.
AI agents write more and more of your code, but git only records who committed.
Six months from now, git blame will point at a teammate — not at the agent, not
at the prompt that produced the change, not at what it cost or what else changed
in that session. The prompt — the actual source of the code — dies with the
terminal window.
Origin fixes that. It runs silently beside your AI coding agent and records every
session — prompts, per-prompt diffs, files touched, tokens, cost — into your git
repo as notes and refs. git clone brings the AI history along with the code.
Supported agents: Claude Code · Cursor · Codex · Gemini CLI (auto-detected). Antigravity, Windsurf and Aider support is experimental.
npm i -g https://getorigin.io/cli/origin-cli-latest.tgz
origin enable # auto-detects your agents, installs hooksThe CLI runs standalone by default: no account, no server — everything stays in your repo and on your machine. (Connecting it to the team platform is where logins and API keys come in.)
There is no step 2 — after origin enable, just code with your agent like you
always do. Origin captures each session automatically. When your first AI session
has run:
$ origin sessions
Sessions (1 total)
8f41ac02 claude-fable-5 ENDED 4 files $0.87 12m agoEvery session is broken into per-prompt snapshots — what you asked, what the agent changed, what it cost:
origin session 8f41ac02 # replay the session, prompt by prompt
origin snapshot # per-prompt snapshots of the current sessionAnd once AI-written code is in your repo, attribution is line-level:
$ origin blame src/auth.ts
Line Tag Agent Model Author Content
─────────────────────────────────────────────────────────────
41 [AI] claude claude-fable-5 dev const token = rotate(session)
42 [AI] claude claude-fable-5 dev await store.save(token)
43 [HU] dev // TODO: audit log
Summary: AI: 2 (67%) Human: 1 (33%)
$ origin why src/auth.ts:41 # the exact prompt behind that line"Which AI wrote this line, and why?" — When something breaks, don't stop at
who committed. origin blame separates AI from human lines per agent and model;
origin why file:line pulls up the exact prompt that produced the line;
origin search "refresh token" full-text-searches every prompt ever run against
the repo.
"What did the agent actually do?" — Agents touch more than you asked for.
origin session <id> replays a session turn by turn: prompt, diff, files,
tokens, cost. origin recap gives you the end-of-day summary; origin stats
the AI-vs-human ratio for the whole repo.
"The agent made a mess — get me back." — Every prompt is a snapshot.
origin rewind --to <snapshot> restores the working tree to the moment before
things went sideways — finer-grained than your last commit.
"The history must travel with the code." — Everything lives in git, not in a vendor database:
refs/notes/origin per-commit agent / model / session / cost
origin-sessions branch transcripts, prompts, per-prompt file changes
~/.origin/config.json CLI config (machine-local)
Clone the repo, get the history. Works offline. No telemetry by default.
"Old repo, no history?" — origin backfill --apply retroactively tags past
commits by detecting the agents that authored them.
The CLI answers "what happened in my repo." getorigin.io answers "what is AI doing across my team": a live dashboard of every session, per-user and per-agent cost, budgets that actually block overspend, model and content policies enforced at commit time, PR compliance checks, and audit trails.
This is the part that needs an account: origin login authenticates the CLI and
issues the API key that links your sessions to your org. Free for solo
developers, $29/user/month for teams.
The CLI has 50+ commands — review, governance, handoff between agents, session
memory, TODO tracking, reports, CI integration. See DOCS.md or run
origin --help.
MIT