Automate your repository tasks like reviewing PRs, triaging issues, or writing weekly team reports. Fully open source, using your favorite coding agent.
Website · Discord · Issues

Your team adopted Cursor, Claude Code, or Amp. Velocity went up. But so did the time you spend on code reviews.

AI tools are great at generating working code, but they don’t know your team’s patterns. They miss your naming conventions, ignore your error-handling style, and introduce subtle inconsistencies that compound over time.

You’re now the human guardrail, catching style drift across dozens of PRs each week.

DevFleet is an open-source agent that reviews AI-generated pull requests and flags pattern drift with clear, actionable comments.

It understands your codebase conventions and points out when AI-generated code doesn’t match:

• Inconsistent naming or file organization
• Missed abstractions your team already uses
• Error handling that doesn’t follow your patterns
• Style violations that slip past linters

Think of it as a senior engineer who’s memorized your style guide and never gets tired.

1. AI agent opens a PR
      ↓
2. DevFleet receives the webhook
      ↓
3. Reviews the diff against your patterns
      ↓
4. Posts inline comments on style drift
      ↓
5. You review with context, not guesswork

We offer a hosted version of DevFleet at https://www.devfleet.ai.

You can find the full self-hosting guide here: devfleet.ai/docs/self-hosting

1. Clone the repo and install dependencies:

   git clone https://github.com/dev-fleet/devfleet.git
   cd devfleet && pnpm install

2. Set up environment variables:

   cp apps/web/.env.example apps/web/.env

3. Create a GitHub App with:

   • Permissions: Contents, Issues, Pull requests, Checks (Read & Write)
   • Events: Issue comment, Issues, Pull request

4. Add your API keys to .env:

   • GITHUB_APP_* credentials
   • ANTHROPIC_API_KEY
   • E2B_API_KEY

5. Start the database and app:

   npm run db:docker:up
   npm run db:migrate
   npm run dev

6. Open http://localhost:3001

We welcome contributions! Whether it's bug fixes, new features, or documentation improvements.

Built by engineers who got tired of reviewing AI drift.

Questions? Open an issue or reach out on Twitter/X.