Join private beta

Built for humans.Drivable by agents.

Your favourite coding agent can orchestrate GetBranchIt too. Give it a shell and it can fan work out across branches, check results, merge the winners and hand you a pull request to confirm.

Discover, then act.

Everything an agent needs to use the canvas correctly ships inside the executable and works offline. Start with the catalog, read the schema for a command, then send a structured JSON input.

Discovery

        
        getbranchit commands
      
        
        # every workflow, plus local and auth commands
      
        
        getbranchit describe merge
      
        
        # effect, usage, confirmation and next steps
      
        
        getbranchit schema merge
      
        
        getbranchit agent-guide

  • Self-describing

    The binary carries its own command catalog, JSON Schemas and an orchestration guide. No MCP client, SDK or checkout is required.

  • Typed effects

    Every command is labeled read, write, publish, destructive, browser or local, with whether it needs confirmation and what to run next.

  • Safe to retry

    Mutations take an operation ID you persist first. Repeating a request produces one fork, one merge, one preview or one pull request.

  • Dry runs

    Validate any input with --dry-run before it touches a branch. Schemas describe shape; authorization stays on the server.

  • Same records as the app

    Projects, branches, checkpoints and diffs are the ones the canvas shows. An agent's fork appears on your canvas immediately.

  • Credentials stay put

    The CLI signs in through your browser and keeps its own Keychain credentials. It never prints access tokens.

Fan out. Compare. Keep the winner.

A typical orchestration: fork two branches from the same point, give each agent a different approach, wait for both, compare the diffs, merge, and prepare a pull request that you confirm in the app.

Two-branch orchestration

        
        getbranchit new-operation-id > op.txt
      
        
        getbranchit fork --input fork-a.json
      
        
        getbranchit fork --input fork-b.json
      
        
        # each branch runs its own agent in the cloud
      
        
        getbranchit compare --input compare.json
      
        
        getbranchit merge --input merge.json --dry-run
      
        
        getbranchit prepare-pr --input pr.json

Reading this as a language model?

A plain-text summary of GetBranchIt, its principles and every public page lives at /llms.txt. Pages also carry schema.org metadata, and the sitemap lists every route.

Open llms.txt