Introduction

Release scope

This documentation describes BranchRunner Community Edition 1.0.0, the first self-hosted release. Read release status for what the release includes, which configurations were verified live and its known limitations.

What BranchRunner is

BranchRunner is a development control plane for turning specified work into reviewed changes. It coordinates tasks, isolated workspaces, coding agents, validation, independent review, signed evidence and explicit human approvals. Its only delivery is a draft pull request. It never merges or deploys.

It does not replace your Git host, CI system, model provider or engineering judgment. It connects them through workflows whose state stays inspectable.

Core model

• A task describes the requested outcome and its numbered requirements.
• A run is one attempt to deliver that task.
• A workspace isolates the checkout and services one run uses.
• An agent profile names a coding agent, its prompt, instructions and limits, and an ordered list of provider connections.
• A workflow runs six stages: Plan, Implement, Validate, Independent review, Approval and Handoff (a draft pull request).
• Evidence connects changes, validation and reviews to requirements. It is signed by the installation.

Where things run

The control plane (web UI, API and PostgreSQL) runs on your server. Runtime workers run repository checkouts, coding agents and validation in hardened containers, either on the same server or on other hosts you enroll. In 1.0.0, a worker on another host runs Codex subscription and API-key stages, while Claude subscription stages run only on the worker on the server. GitHub access goes through a GitHub App that the installation creates on your own account.

Choose a path

Install BranchRunner with the quickstart, understand tasks and runs, or read security to report a vulnerability.