A 24/7 engine for the AI coding loop.

Munk AI watches for work, plans changes, directs coding agents, verifies on real devices, repairs failures, and delivers passing results. One continuous loop — without prompting every step.

Read the Docs

Loop

Start with one Loop.

One durable Loop runs Plan → Code → Gates → Deploy → Verify → Delivery. Gates and Verify can fail into Repair; delivery opens only after verify passes.

Orchestrate a recoverable engineering loop — not a one-shot pipeline.

Graph

When one Loop isn’t enough.

Complex demand needs more than one Loop. Compose them into a Graph, then let each Loop advance — by dependency, in parallel where edges allow.

Loop A runs Intake — the foundation of a complex demand.

Product family

Two products. One feedback story.

Munk AI owns the engineering loop on the desktop. Munk Test owns device-level verification inside that loop. They connect — they are not the same product.

Desktop orchestration

Munk AI

Decides when to plan, code, run gates, deploy, verify, repair, and deliver. Holds wait / resume / retry / max-round policy in the workflow — not in a single agent chat.

Work item → durable workflow
Multi-role, multi-stage, multi-round repair
Delivery gated on gates + verification

CLI verification engine

Munk Test

Local-first, multi-platform automation testing runtime (CLI). Open source today. In the Munk AI loop it is a primary verify surface — not the orchestrator.

Plan / run / review / verify change
Android, iOS, and Web
CLI, MCP, Local API, Web UI

Why orchestration

Built for loops that must survive reality.

Agents write code. Orchestration decides when they stop, retry, wait for humans, and ship.

Durable workflow runtime

Park, resume, and memoize steps so long-running jobs survive restarts and human checkpoints.

Policy stays in the workflow

Wait, resume, retry, and delivery gating are workflow concerns — not hidden inside a provider adapter.

Verification-gated delivery

Delivery ready only when code gates pass and verification verdict is passed.

Human remains in control of merge

GitHub delivery creates or updates a Draft PR. Auto-merge of GitHub PRs is intentionally out of scope.

Debuggable by design

Job timeline, debug projections, and structured evidence keep the loop inspectable when something goes wrong.

Boundaries

Product boundaries.

Private preview

Munk AI is not publicly available yet. This page introduces the product — private preview access will open later.

Orchestration, not a coding agent

It does not replace your coding provider. It decides when coding, gates, deploy, verify, and repair run.

Not a replacement for Munk Test

Munk Test remains the open-source CLI verification engine. Munk AI may call it as part of verify — it does not absorb that product.

Waitlist

Get notified for private preview.

Follow updates on@iBoyCoder