Fluent Flow

Automated code review orchestrator for AI coding agents. Manages the review-fix loop so agents focus on writing code — not deciding what to do next.

How It Works

1. An AI agent opens a PR
2. Fluent Flow runs an automated review
3. If the review fails, the agent is notified with specific feedback
4. The agent pushes a fix, triggering a re-review
5. On pass, the PR is auto-merged. On repeated failure, a human is escalated

Packages

Package	Description
fluent-flow	Server — review engine, agent registry, MCP server, webhook handler
fluent-flow-runner	CLI runner — connects to the server, picks up work, spawns agents locally

Getting Started

Server

cd packages/fluent-flow
cp .env.example .env   # configure DATABASE_URL, GITHUB_TOKEN, MCP_AUTH_TOKEN
npm install
npm run dev

See server docs for full setup, concepts, and configuration.

Runner

# 1. Register an agent and generate a token (see runner docs)
# 2. Run
fluent-flow-runner --token ff_<token> --server http://localhost:3847 --verbose

See runner docs for step-by-step setup.

Docker

docker compose up -d --build

Documentation

• Design specs — Architecture and design decisions
• Implementation plans — Step-by-step build plans

Development

This is a Turborepo monorepo.

npm install         # install all dependencies
npm test            # run all tests
npm run build       # build all packages

License

MIT