Skip to content

MuAiFlow

A workflow for building software with several AIs, where one plans, another reviews and nothing runs without human approval.

  • Laboratory
  • 2026
  • Product, Engineering
  • Node.js
  • TypeScript
  • VS Code

Repository ↗

Technical detail

It came from an everyday problem: generate a plan in one AI, copy it to another for review, go back to the first to execute. The person became the middleman between tools.

MuAiFlow replaces the copying with a file-based protocol. The plan lives in Markdown, moves from draft to reviewed, and stops to wait for approval. Review is mandatory for migrations, authentication, external integrations and large changes. It works with Claude Code, Codex and Crush in any role, with one rule: whoever reviews is not whoever planned.

It includes an npm package, installable skills and a router that sends each task to a model of the right size.

The VS Code extension handles an everyday detail: typing @ in a Markdown file opens a list of the project's files and folders and inserts the relative path. The list is cached in memory and only rebuilt when a file is created or deleted, or when the settings change. The search is capped at 15,000 files so large monorepos do not freeze.

One decision worth recording: the @ trigger listens to document changes instead of using a completion provider. A provider only responds if the extension is already active, so anyone who opened a .md and typed @ right away hit a race condition. A 100 ms debounce avoids opening several pickers when typing @@ or pasting text.

It is the process we use in our own projects.

You talk to the people who build it.

30 minutes by video, in English or Portuguese, with the person who will lead your project. No sales pitch.

When booking, three quick questions: company, context and timeline.

Book a call Prefer to write? LinkedIn