Hey all,

We just published a Railway template for SWE-AF, our open-source autonomous engineering team runtime.

You give it a goal and a repo URL. It spins up hundreds of Claude Code instances as a coordinated engineering team - PM, architect, coders, QA, reviewers, merger - each working in isolated git worktrees communicating with each other. One API call, zero hand-holding. It plans, architects, builds, iterates, tests, resolves merge conflicts semantically, and opens a complete production ready draft PR when it's done.

curl -X POST "$YOUR_RAILWAY_URL/api/v1/execute/async/swe-planner.build" \
  -H "Content-Type: application/json" \
  -d '{
    "input": {
      "goal": "Add JWT auth to all API endpoints",
      "repo_url": "https://github.com/you/your-repo"
    }
  }'

That's literally it. Walk away, come back to a production ready PR.

We've been using it on our own repo PR #179 was built entirely by SWE-AF. 10 issues, 217 tests passing, $19 in API costs. No human touched the code.

The bigger stress test: we told it to build a Rust-based Python compiler from scratch. 175 agents, 40 minutes, $53 total. Working compiler, 253x faster than CPython on short snippets.

It's not just Claude either - supports OpenRouter (DeepSeek, Qwen, MiniMax, 200+ models), OpenAI, Gemini. Pick your model per role if you want.

The template deploys the control plane + PostgreSQL + the SWE-AF agent node. Should run fine on hobby tier for trying it out.

Repo: https://github.com/Agent-Field/SWE-AF/

Railway Template: https://railway.com/deploy/swe-af

Apache 2.0.

Would love to hear what people build with it. Happy to answer questions about the setup.