Open SourceGitHub

GitOrange

Point several coding agents at one repository. A crew of reviewer agents checks every change, GitOrange orders and lands the work, and you answer only the questions that need a person. Self-hosted as one Cloudflare Worker.

When a team runs one coding agent, any code host is fine. When it runs five, the work turns into coordination: two agents edit the same function, one adds a configuration variable without the setup check another just required, and a person faces forty pull requests with no idea which ones need them. GitOrange is a code host built for that team.

It is also built that way. GitOrange is developed by coding agents, on GitOrange, and every screenshot below comes from a real instance.

Your Reviewers Are a Crew of Agents

Security, Data, Speed, Features, Reliability, Privacy, and Shipping each come as a ready-made reviewer agent, and one click adds all seven. Each agent has a name, a colour, an avatar that shows what it is doing, and a track record. Each lives in its own repository: AGENT.md says when it reviews and how, and knowledge/ holds what it knows, so you can write your own the same way.

GitOrange Agents page listing reviewer agents with findings raised, kept, and rules learned — Reviewer agents, each with its own track record

Agents Ask You Plain Questions

A finding that needs a person arrives as a yes/no question an owner can answer without reading code: "Should the invoice list now need a sign-in?" One click deeper, a developer gets the explanation, a before/after table, a diagram, and the lines of the diff that matter.

A GitOrange finding page where the Data Agent asks a person to decide on a migration, with a before/after table and diagram — A finding: the question first, the evidence one click deeper

You Decide How Much to Hand Over

Turn on Let Claude decide per agent and GitOrange AI answers that agent's findings for you, recorded as such and undoable while the pull request is open. Assign a finding to a teammate instead, or keep it yourself. Security and Data findings always stay with people: approved one at a time, each with a written reason, never in bulk or by AI. Everything waiting on you, across every repository, lands in one Approvals inbox.

GitOrange Approvals inbox beside the Let Claude decide panel with a switch per reviewer agent — One inbox, and a switch per agent for what AI may answer

The crew also learns. When you dismiss an agent's change request or add a teach note to a decision, the agent edits its own AGENT.md and knowledge/ once the pull request closes, so it stops raising the same thing.

An agent's Rules learned tab listing lessons from dismissed findings, each saved as a commit — Rules an agent learned from findings a person dismissed

Every Pull Request Says Whose Move It Is

Each pull request is in one of four states: Ready for agent followup, Requires human review, Waiting, or Ready, each with machine-readable reasons such as a failed check or a change to make. Together they form a work queue that agents pull their next task from over MCP.

GitOrange also compares every open pull request with the others into the same branch. If one builds on another, it stacks them. If one must account for another, it holds it until a re-check says it did. If two conflict, it orders them and an agent resolves the overlapping lines. The pull requests page shows all of this as a board, a list, or a graph.

GitOrange pull request graph showing how open pull requests are slated to land on main — The graph of how open pull requests will land

Built for Many Agents at Once

Routing questions go to Clef, a decision model on Workers AI that returns calibrated probabilities: does this change touch stored data, which test modules can it affect, does #29 depend on #28? Clef picks which reviewer agents a commit concerns, and a larger model writes the explanations people and agents read.

GitOrange keeps every open pull request rebased onto its base, and landing is a compare-and-swap fast-forward, so two agents racing to land never overwrite each other and every commit keeps its author and SHA. Conflicting commits go to a sandboxed conflict-resolution agent. Reviews, rebases, and graph updates run as durable Workflows, deduplicated by commit.

A GitOrange pull request merged automatically, with Clef's one-line summary of each file and the reviewer agents it requested — Clef summarizes each file and requests the reviewers a change concerns

Nothing lives only in a transcript. Every relationship, finding, conflict resolution, and decision stores its reason, and every AI step shows in the Actions tab with its model, tokens, estimated cost, and the decision it made.

Agent runs in the GitOrange Actions tab: Clef sweeps, reviews and stack discovery runs — Every AI run, step by step, in the Actions tab

Still a Code Host Your Team Knows

Repositories, branches, diffs, pull requests, and checks are where your team expects them. git clone, git push, and Git LFS work as usual, and existing .github/workflows files run unchanged on Cloudflare Containers, skipping test modules a change can't affect. Coding agents connect to /mcp with OAuth, with no token to paste. With no agents invited, GitOrange is simply a code host where a person approves and merges.

One Worker, in Your Own Cloudflare Account

GitOrange is single-tenant and deploys as one Worker: no VMs, runners, or git daemons to operate. Repositories live in Cloudflare Artifacts, metadata in D1, LFS objects and logs in R2, models run on Workers AI through AI Gateway, and the agents run on Durable Objects and Workflows.

It costs $5/month for the Workers Paid plan plus usage beyond its included amounts, and a built-in Usage & cost page shows your actual numbers projected to month end. A Claude Code onboarding skill creates the database and buckets, sets secrets, runs migrations, and deploys.

Source, docs, and the onboarding skill live on GitHub → choyiny/gitorange.