Origin CLI

Every agent. Every prompt. Every line. In your git repo.
The AI coding history layer for developers and teams.

---

Origin runs silently next to any AI coding agent — Claude Code, Cursor, Codex, Gemini CLI, Aider, Windsurf, Copilot, and more — and captures every session: prompts, files touched, tokens, cost, diffs. All of it lives in your git repo as notes and refs. git clone brings the history with the code. No server, no login, no API keys required.

Install

npm i -g https://getorigin.io/cli/origin-cli-latest.tgz

Quick start

origin enable                     # auto-detect agents, install git hooks
# ...code with any AI agent — Origin tracks automatically
origin blame src/index.ts       # see which AI wrote each line

That's it. Nothing to configure.

---

The four beats

Origin maps to one idea per verb. Everything else is detail.

01 — Every agent.

Auto-detected. Claude Code, Cursor, Codex, Gemini CLI, Aider, Windsurf, Copilot, Continue, Cody, Cline, Codeium, Roo, Kilo. One command picks up whichever tool you use.

origin enable           # detect agents + install hooks
origin agents         # list detected agents
origin status         # show the active session

02 — Every prompt.

Each turn is captured as a snapshot: prompt text, model, files touched, diff, tokens, cost, duration.

origin snapshot       # list per-prompt snapshots in the current session
origin sessions       # list all sessions
origin session <id>   # replay a single session
origin rewind --to <sha>   # restore working tree to any snapshot

03 — Every line.

Line-level attribution across agents and sessions. Point at any line, get the exact prompt that wrote it.

origin blame <file>          # per-line AI/human + model per line
origin why <file>:<line>     # the exact prompt behind one line
origin diff                  # annotated diff, AI vs human
origin search "auth bug"     # full-text search across prompts

04 — In your git repo.

Nothing leaves your machine. Sessions live in refs/notes/origin and the origin-sessions branch — the history moves with the repo, not through a server.

refs/notes/origin              per-commit model / session / cost / tokens
refs/notes/origin-memory       cross-session memory (what past sessions did)
refs/notes/origin-memory-brief distilled continuation brief for the next session
refs/notes/origin-acceptance   how much of a past agent's output survived
refs/notes/origin-repo-brief   repo brief (what this repo is) — opt-in
origin-sessions                transcripts, prompts, file changes
~/.origin/config.json          CLI config (machine-local)

No telemetry by default. Opt in with origin config set telemetry true.

---

Memory that follows the repo

Agents forget everything between sessions. Origin gives them a memory that lives in the repo — so the next agent (any agent) starts where the last one left off. Two layers, both stored as git notes. Git itself never clones or pulls refs/notes/*, so Origin installs a fetch refspec the first time it syncs a repo; after that your ordinary git pull carries memory down with it, and the post-merge hook makes it readable. A clone on a machine without the Origin CLI won't see it.

Session memory — what past sessions did. On session end (or on every commit — your choice), Origin writes a compact entry per session to refs/notes/origin-memory: summary, files touched, open TODOs. At the next session start it distills these into a short brief — recent changes, frequently touched files, carried-over TODOs — and injects it into the new agent's context. 100% local, no LLM, no server.

origin context memory                      # what previous sessions did here
origin config set memoryUpdate commit      # refresh memory on every commit
                                           #   (default: session-end · also: both)
origin context clear --memory-only         # wipe it

Repo brief — what this repo is. Opt-in. A one-time, cached LLM summary of the repo's purpose, architecture, entry points, and gotchas, injected at session start so an agent is oriented before its first prompt. Generating it sends a bounded bundle (README, manifests, file tree, recent commit subjects) to Anthropic using your own key; the result is cached in refs/notes/origin-repo-brief and only regenerated when the repo drifts.

origin context brief --enable     # turn it on for this repo
origin context brief              # show the cached brief
origin context brief --refresh    # regenerate now (uses your Anthropic key)

Session memory is on by default and never leaves your machine. The repo brief is the only piece that calls an external service, and only after you enable it.

---

Supported agents

Full session capture (native hooks): Claude Code · Cursor · Codex CLI · Gemini CLI · GitHub Copilot · Windsurf · Antigravity Commit-time detection: Aider · Cody · Continue · Codeium · Cline · Amp · Junie · OpenCode · Rovo Dev · Droid

Detection runs on CLI availability, IDE extension inspection, MCP config, and process detection at commit time.

---

More commands

The CLI has 50+ commands covering review, governance, handoff, memory, TODOs, time travel, reports, audit, and CI integration. See DOCS.md or run origin --help.

Commonly used beyond the four beats:

origin stats                 # AI vs human stats for the repo
origin handoff show          # pass context to the next agent
origin recap                 # end-of-day summary
origin backfill --apply      # retroactively tag old commits
origin policies              # list active governance policies
origin doctor                # diagnose stuck sessions
origin upgrade               # update to latest

---

For teams

getorigin.io adds the team layer on top of the CLI: live dashboard, per-user cost attribution, model and budget policies, PR compliance checks, audit trails, and GitHub App / Slack integrations. Free for solo developers, $29/user/month for teams.

origin login      # authenticate with your Origin instance
origin enable       # register the machine + install hooks

---

License

MIT