MadoHub Docs

Introduction

MadoHub is a canvas-based desktop app for running and orchestrating multiple AI coding agents.

MadoHub is a native desktop application that gives you an infinite canvas for running multiple AI coding agents simultaneously. Instead of juggling terminal tabs, you arrange agents visually — and route outputs between them automatically.

Core concepts

  • Canvas — An infinite, zoomable workspace where you place tiles. Pan, zoom, and organize spatially like a design tool.
  • Tiles — Containers for terminals, code, notes, images, and file trees. Each tile is draggable and resizable.
  • Agent detection — MadoHub monitors terminal output and process trees to identify which AI agent is running (Claude Code, Codex, Amp, OpenCode, or a generic shell) and tracks its state in real-time.
  • Routing — Connect tiles to automatically pass one agent's output to another as an actionable prompt. MadoHub uses AI to transform the output into context-aware instructions.
  • MadoAgent — A natural-language orchestrator (Cmd+Shift+M) that can drive your canvas and agents on your behalf.
  • Flows — Reusable multi-agent templates (like peer review or pipeline workflows) you can save and re-run.
  • Fleet/Mission Control — A higher-level view for overseeing and coordinating many agents at once.

What agents does it support?

Any terminal-based AI agent works in MadoHub's terminal tiles. Agent detection and routing currently have deep integration with:

  • Claude Code — Full support including JSONL session reading, completion detection based on its spinner-and-duration output pattern, and permission dialog awareness.
  • OpenAI Codex CLI — Detected via prompt patterns and branding markers.
  • Amp and OpenCode — Detected via backend process-tree classification.
  • Any shell command — Falls back to generic shell detection.

Tech stack

MadoHub is built with Tauri v2 — a Rust backend with a React + TypeScript frontend. This means native performance, small binary size, and deep OS integration. Workspace state is stored locally; AI requests go only to the provider you configure.

On this page