Overview
mD-AI hands repeatable work to AI agents. It runs coding agents in parallel on a server you control, each task in its own Git worktree and branch, so several fixes or features move at once while you review the diffs and merge what works.
An orchestrator, which you type or dictate to from any device, manages projects, tasks and agents for you, and notifies your phone when an agent finishes, needs you or fails.
Who it is for
- Developers and small teams who want agents working on their repos while they are away from the desk.
- People who use Claude Code, Codex, OpenCode or other CLI agents and want them side by side in one place.
- Anyone who wants the agents, code and history on their own server.
What it automates
- Parallel agents, each in its own Git worktree and branch.
- Project flows: investigate, implement, review and deploy, stopping only where a flow says to ask you.
- Diffs, pull requests and CI checks from one place.
- Work that keeps running on the server after the browser closes.
Built-in automation
- Scheduled automations that trigger the orchestrator with your own prompt, saved as templates.
- A nightly self-review: a read-only run that summarises the outcome log, proposes fixes and saves a report.
- Phone notifications when an agent finishes, needs you or fails.
Tech stack
TypeScript, React 19, Vite, Tailwind CSS, Node.js 24, SQLite (better-sqlite3, Drizzle), node-pty, Claude Agent SDK, Docker Compose, Caddy.