Summary

Broccoli OSS automates converting Linear tickets into shipped GitHub PRs using AI (Claude, Codex). It targets engineering teams seeking integrated AI assistance, offering a self-hosted solution within their Google Cloud environment for enhanced data privacy and control.

How It Works

Broccoli uses two Cloud Run workloads (service, runner) backed by PostgreSQL. The runner leverages Claude/Codex CLIs with customizable prompts. Deployment within the user's GCP project ensures full control over infrastructure, keys, and data, avoiding third-party dependencies.

Quick Start & Requirements

• Primary Install/Run: Deployment via bootstrap script (deploy/bootstrap.sh) after initial setup. Google Cloud Build recommended for image creation.
• Prerequisites: GCP project with billing; authenticated gcloud CLI; GitHub, Linear, GCP Console access; GitHub App (Contents, PRs, Issues: read/write; Metadata: read-only; PR review, PR events); dedicated Linear bot user with API key; OpenAI/Anthropic API keys; Python 3.12+ and uv.
• Estimated Setup Time: ~30 minutes.
• Links: GitHub App (README -> Deploy -> 1. Create GitHub App), Linear bot (README -> Deploy -> 2. Designate Linear bot user), Deployment guide (README -> Deploy).

Highlighted Details

• AI Automation: Transforms Linear tickets into reviewable PRs, handling planning and implementation.
• Self-Hosted & Data Control: Deploys to user's GCP, keeping data within their tenancy.
• Customizable Prompts: Users can fork, tune, and version prompt templates.
• AI Code Review: Claude/Codex provide actionable comments and can push fix commits.
• Production-Grade: Uses Cloud Run, Secret Manager, durable job state.

Maintenance & Community

Includes CI via GitHub Actions, CONTRIBUTING.md, and CHANGELOG.md. Specific maintainer details, sponsorships, or community channels (Discord/Slack) are not explicitly mentioned.

Licensing & Compatibility

• License: MIT License.
• Compatibility: MIT permits commercial use and closed-source linking. Project is described as "production-grade" and "self-contained."

Limitations & Caveats

Initial setup involves manual steps across GCP, GitHub, and Linear. Relies on external OpenAI/Anthropic APIs (requires active accounts/billing). Users manage their own GCP resources and costs. While contributions are welcomed via GitHub, explicit community forums are not detailed. Swapping LLMs requires forking runner handlers and prompt templates.