Spaghetti Code FoundrySCF

Open source · MIT · runs on your own machine

Your own AI coding factory: from GitHub issue to merged code

Put one label on a GitHub issue. Spaghetti Code Foundry asks the open questions, plans, codes, tests, reviews and merges the story with Claude Code and Codex, and asks you only when it has to.

Install the Foundry See how it builds issues View on GitHub 
git clone https://github.com/MeloMar-IT/spaghetti-code-foundry.git
cd spaghetti-code-foundry
npm install && npm run build && npm link
scf ui      # web UI at http://localhost:4777
The Foundry's Home: what needs you, what is running and what finished
Home shows first what waits for you, with one button each.

How it works

  1. Label an issuePut Factory_go on a GitHub issue or a whole epic. That is the only label you set.
  2. Questions firstThe Foundry asks its open questions on the issue, up front. Answer them, or reply /defaults to take its recommendations.
  3. Plan, code, test, reviewEach story gets a plan with a risk score, a review of the plan, code on its own feature branch, tests, a code review by a second model, and docs.
  4. Merged, once a day to mainFinished stories merge into develop. Once a day you get one pull request from develop to main to merge.

Why teams use it

  • Hands-off, with you in control

    Every plan gets a risk score from 0 to 100. Risky plans wait for your approval; stories that are too big are split into smaller ones.

  • Always clear what happens next

    Home shows what waits for you. The board shows where every story is and why. A health line says when something is wrong with the Foundry itself.

  • Several stories at once

    Runs lock the parts of the code they change, so stories in different areas are built in parallel without merge conflicts.

  • Two agents, any model

    Steps run on Claude Code or the Codex CLI, on Anthropic, OpenAI or local models such as Ollama and LM Studio. One vendor's model reviews the other's work.

  • Your own flows

    Work is described as flows: YAML pipelines of agent steps, shell steps, approvals and branches. Edit them visually, reuse blocks, or let any AI assistant write one.

  • Safe by default

    Every run works in its own copy of the code. Pushes to protected branches and pushes that contain secrets are blocked. Approval steps, budgets and sandboxing when you want them.

  • See everything

    Live logs, readable transcripts of every agent step, diffs, a dashboard of costs, and resume, retry and approve buttons.

  • Turn ideas into stories

    Refinement helps you write good stories with an AI architect that reads your code, asks the right questions and checks each draft against your Definition of Ready.

  • Accounts and roles

    An admin makes flows, watchers and settings. Users run the flows the admin published, on their own repositories, and see only their own runs.

All features, with screenshots

Built for the way you already work

The Foundry follows gitflow: each story on its own feature branch, merged into develop when it is done, and one release pull request a day to main. It reads your issues, comments on them, and labels them so you can follow every story on GitHub too.

It runs on your own Mac, with your own accounts for Claude Code, Codex and GitHub. Your code and your keys stay with you.

How the daily release works

The board: where every story is, and why
The board: where every story is, and who has the next move.

Start here

All how-to guides · The complete user guide

Frequently asked questions

What is Spaghetti Code Foundry?

An open-source AI coding factory that runs on your own machine. It runs flows: pipelines of steps that code, test, review and ship changes with AI coding agents. Its main use: put one label on a GitHub issue, and the Foundry asks the open questions, plans, codes, tests, reviews and merges the story.

Which AI agents and models does it use?

Steps run on Claude Code or OpenAI's Codex CLI, on Anthropic, OpenAI or local models (Ollama, LM Studio). Routing rules pick the model per step, with fallbacks when a model hits a limit. See choosing models.

What do I need to run it?

Node.js 20 or newer and git, Claude Code logged in, and the GitHub CLI (gh) logged in for the GitHub flows. The Codex CLI, Ollama or LM Studio and Docker are optional. It is built and used on macOS. See how to install it.

Does my code leave my machine?

The Foundry runs on your own machine and works on a copy of your repository for every run. The agents it starts use the AI providers you configure, the same way they do when you use Claude Code or Codex yourself.

How do I stay in control of what it changes?

Every plan gets a risk score; risky plans wait for your approval. Pushes to protected branches and pushes that contain secrets are blocked, and all work lands on develop first. You merge one pull request a day to main. See keeping your code and secrets safe.

Can I write my own pipelines?

Yes. Flows are YAML files you edit visually or as text, with a block library of reusable steps. Any AI assistant can write one for you from the flow format reference. See how to write a flow.

What does it cost?

The Foundry itself is free and open source under the MIT licence. You pay for the AI subscriptions or API usage of the agents and models you choose. The dashboard shows what each run cost. See controlling costs.

Is it only for GitHub?

The issue watchers and the daily release pull request work with GitHub. You can also run any flow by hand on a local repository with scf run. See your first run.

Give your backlog to the Foundry

Install it in a few minutes, connect a repository, and put one label on an issue.

Install the FoundryStar it on GitHub