flowforge

Ship software from a single prompt

A multi-agent LangGraph pipeline that turns one line of intent into a spec, a plan, working code, parallel quality reviews, triaged issues, and a pull request.

Get started View on GitHub
pip install swe-forge
Works with Claude Code · Codex · GitHub Copilot
prompt clarify spec · plan tasks fan-out code review security audit test engineer triage ship pull request

LangGraph pipeline

A durable 10-node graph — clarify, spec, plan, fan-out, gates, triage, ship — auditable at every step.

Deep Agents

Each node runs as a LangChain Deep Agent with per-run budgets and a diff-based secret scan before any file is written.

Quality gates

Code review, security audit, and test engineering run in parallel, then findings are triaged into GitHub issues.

How it works

  1. clarify Ambiguous prompts are resolved before any code is written.
  2. spec · plan A specification and a dependency-ordered plan are produced.
  3. fan-out Independent tasks execute in parallel, then merge into one state.
  4. gates Three quality gates produce structured, severity-tagged findings.
  5. triage Findings are deduped, prioritized, and filed as GitHub issues.
  6. ship A feature branch and pull request — with release flow when gates pass.

See it run

Watch FlowForge take a prompt all the way to a pull request.

Watch the FlowForge demo video

Get started

1
pip install swe-forge
2
swe-forge setup
3
swe-forge run "build tic-tac-toe web app"

The PyPI distribution is swe-forge; the Python module is flowforge. Generated projects land in ~/flowforge-workspace/<slug>/.