Most AI coding tools have a weird hidden cost:
They use an LLM to decide what another LLM should do next.
That means you are spending tokens on routing, not just on the actual coding.
I built RedQueen to flip that model.
What is RedQueen?
RedQueen is an open-source, self-hosted orchestrator for AI coding agents.
Instead of asking an LLM, “what should happen next?”, RedQueen uses a deterministic state machine to move work through a software development pipeline:
spec → code → review → test → human review
No token spend for orchestration.
No mysterious agent loop.
No “trust me bro” routing logic.
Just a debuggable workflow you can read, edit, and run yourself.
Why this matters
Autonomous coding gets messy when one giant prompt is responsible for everything.
RedQueen breaks the work into isolated AI workers:
- one worker writes the spec
- one writes the code
- one reviews the code
- one tests it
- humans approve at configurable gates
Each phase has a focused job.
And the workflow itself lives in config, not vibes.
Want to add a security review step? Add it.
Want to remove spec review? Remove it.
Want the agent to stop and wait for a human before touching production code? Make that a gate.
The opinionated part
I do not think the future of AI coding is one giant autonomous blob.
I think it looks more like a tiny dev team:
- focused workers
- explicit handoffs
- retry limits
- human review gates
- issue tracker sync
- source control integration
- boring, readable orchestration
Boring is good.
Boring is how you debug things at 2 a.m.
Quickstart
npm install -g redqueen
redqueen init -y
redqueen start
Open the dashboard, label a GitHub issue, and watch it move through the pipeline.
RedQueen currently dispatches Claude Code workers and supports GitHub Issues and Jira, with the adapter pattern designed so more integrations can be added cleanly.
Why I made it
I wanted AI coding agents that felt less like slot machines and more like build systems.
A build system does not spend tokens deciding whether to compile.
A CI pipeline does not ask an LLM whether tests should run.
So why should an AI coding workflow?
RedQueen is my attempt at making autonomous coding more:
- deterministic
- inspectable
- self-hosted
- configurable
- human-controlled
- cheap to orchestrate
It is still early, rough around the edges, and very much a v0.1 preview.
But the core idea is simple:
Stop orchestrating with LLMs.
Start commanding them.
Check it out here:
GitHub stars, bug reports, brutal feedback, and weird workflow ideas are all welcome.
Top comments (0)