The harness runtime for multi-agent collaboration.

Coordinate your agent team across code and projects.

Individual AI agents are great at one-off tasks, but they have no shared way to work together over time. Munk AI brings your agents into a structured local workspace—with hard quality gates and human checkpoints, so you make the calls that matter.

How the harness works

  1. 1

    Code: Gated product delivery

    Turn requirements into trackable work items. Agents plan, write code in isolated Git worktrees, and must pass automated code gates before anything moves forward.

  2. 2

    Projects: Ongoing multi-agent workflows

    Not everything is a code change. Organize recurring and event-driven tasks—GitHub triage, QA verification, release prep, research, and content—into structured projects with shared files, skills, and journals.

  3. 3

    Gates: Human checkpoints for critical calls

    Nothing ships or merges behind your back. Releases, merges, and outbound actions pause at explicit gates until you approve—on desktop or from your phone.

Open Standard

Built on the open-source Open Company spec

Your team's structure, products, projects (projects/<id>/tasks/), and agent rules live in plain-text files under an open specification. Zero lock-in—fork a working template and run it on your machine.

Local-First & Free

Runs on your machine. Free for local use.

Munk AI runs locally and is free to use on your machine—your code, credentials, and workspace never leave your control. The Open Company specification and official templates are open source under the MIT License.

Put your agent team to work today.

Free for local use. Available for macOS, Windows, Linux, and Android.

Download Munk AI