Skip to content

Autopilot Core Documentation

Welcome to the Autopilot Core documentation.

autopilot-core serves as the centralized control plane for the Coding-Autopilot-System platform. It acts as an organizational AI autopilot, orchestrating autonomous issue intake, bug fixing, and pull request generation for opted-in repositories across the organization.

Overview

The autopilot platform is composed of three primary parts: - autopilot-core (This Repository): The control plane. Manages org-wide intake governance, schedules the operator, creates pull requests, and maintains rollout visibility. - ci-autopilot: The runtime pattern. Contains the Python agent and workflow assets executed by runners to perform the actual repairs. - autopilot-demo: A demonstration repository. Provides a safe target for illustrating the end-to-end failure-to-fix loop.

Key Features

  • Centralized Control Plane: Operates via explicit GitHub Issues queueing rather than making opaque direct mutations to source code.
  • Auditable Lifecycle: Every step—from CI failure to intake issue, operator run, fix branch generation, and PR creation—is transparent and traceable in GitHub.
  • Guardrailed Execution: Uses explicit label-gating (e.g., autofix + queued), skips risky work, and verifies fixes before creating Pull Requests.
  • Automated Rollout: An installer workflow automatically distributes the intake automation to any repository in the organization that opts in.

Getting Started

To learn more about the technical details, read the Architecture Guide.

If you are looking to set up or configure the operator, please refer to the Operator Runbook and the Repo Onboarding Runbook.