Open source git alternatives don't f**k, ours does.
A self-hostable code host for repositories, pull requests, issues, and CI, with agents that pick up check.failed and review.requested_changes before you get to them.
$ git remote add origin https://open-git.com/you/app.git$ git push -u origin main Enumerating objects: 14, done.remote: Resolving deltas: 100% (6/6), done.remote:remote: Open a pull request for 'main':remote: https://open-git.com/you/app/pulls/newremote:To https://open-git.com/you/app.git * [new branch] main -> mainAdd rate limiting to the public API
you wants to merge 4 commits into main
Autopilot pushed a fix for the one failing check.
Keep GitHub. Try open-git anyway.
Mirror a GitHub repository into open-git and keep the two in sync. Your team keeps pushing where they already push, nothing about their day changes, and you get to run real work through open-git before committing to anything.
Nothing to schedule
No migration window, no freeze, no weekend. The mirror is live while everything carries on as it is.
Your history, unchanged
Plain git on both ends. Every commit, branch, and tag stays exactly what it was.
Reversible by design
If it isn't for you, stop mirroring. The repository is already whole in both places.
Autopilot works the pull request while you sleep.
A check goes red at 2am. Autopilot reads the failing job's logs, works in a sandbox, pushes a commit, re-runs the check, and leaves a summary of what it changed and why.
What sets it off
check.failed- A required check goes red on the pull request.
review.requested_changes- A reviewer blocks the pull request.
review_comment.created- Someone leaves a comment on a line.
manual- You ask for it, from the timeline.
What keeps it in bounds
- Your key, your model
- Runs on your own OpenRouter key against whichever model you pick.
- A second opinion on retry
- Nominate a different model for the retry when Autopilot's own fix fails CI.
- Attempt ceiling
- Consecutive attempts on one pull request are capped. Three by default, then it stops and leaves it to you.
- Monthly spend cap
- A dollar limit per calendar month, per organization. Autopilot stops when it is reached.
- Your test command
- It proves the fix with the command you already trust, not a guess at one.
Review the prompt, not just the diff.
When a change comes from Claude Code, Codex, or Cursor, open-git captures the session — prompts, transcripts, tool calls — and threads it into the pull request timeline. Read what the agent was told before you read what it wrote.
Claude Code
Codex
Cursor
Any other agentCheckpoints are stored on a metadata branch inside your own repository, in a format compatible with Entire — so the record travels with the code rather than living in our database. Off by default on public repositories.
The rest of the code host, included.
Not modules, not an integration budget. The surfaces a team needs to ship are part of the same install.
Built-in CI
Workflows in .opengit/workflows, executed on runners you host. Container per step, no minutes to buy.
Campfire
Per-repository chat with channels, so the decision and the code it changed live in the same place.
Docs
Point at a directory of Markdown and it becomes a docs tab on the repository.
Issues and reviews
Issues, pull requests, and reviews with the conversation attached to the diff.
Insights
Contribution and delivery views built from your own data, not a sampled export.
Access control
Organizations, teams, and per-repository roles with capability-level permissions.
It's yours, including the parts that usually aren't.
One compose up and it runs. CI executes on hardware you own, the data sits in a Postgres you can query, and the source is open for you to read before you trust it with anything.