Assign the right work
MUON turns a mission into explicit jobs, roles, scopes, ownership, and checkpoints before agents begin working.
The orchestration layer for AI software engineering
Make your coding agents work as one.
Give MUON one mission. It plans the work, assigns clear responsibilities, dispatches governed Claude Code and Codex workers, keeps their context aligned, and brings your team back for the decisions that matter.
Governed Claude Code + Codex workersShared crew memoryHuman approval before merge
Private beta. Bring your existing provider access and model usage.
Inspect the workflowOLLAMA
Optional local recallNot a coding workerCLAUDE CODE
Implementation / reviewCHILD 01CHILD 02
CENTRAL ORCHESTRATION
MUON
MUONPlans · assigns · coordinates · governs
UNIFIED CREW MEMORY
CURSOR
Human workspaceReadiness-only pathCODEX
Implementation / reviewCHILD 03CHILD 04
HUMAN AUTHORITYApprove fan-out · merge · shipOne mission record
The coordination gap
More agents create more output, but they also create more briefs, handoffs, overlap, and review work for a person to manage.
MUON handles the coordination work your engineers should not have to.
Keep the crew aligned
Each worker receives relevant memory and a scoped view of peer progress, current targets, duplicate-work signals, and conflict warnings.
Review one result
MUON keeps handoffs, checks, changed files, and approval evidence attached to the mission so your team can review the outcome, not reconstruct it.
One mission, one coordinated crew
How MUON runs the crew.
MUON turns a request into assigned work, keeps agents aware of the mission and each other, and returns one reviewable result to your team.
Plan
Turn intent into clear jobs
MUON records the goal, breaks work into bounded responsibilities, and proposes which lane should own each step.
Assign
Create a governed crew
Claude Code and Codex workers receive explicit briefs, workspace boundaries, budgets, lineage, and authority limits. Parallel work waits for approval.
Coordinate
Keep every worker informed
MUON supplies relevant memory and peer activity before edits, carries structured handoffs forward, and flags overlap before agents collide.
Review
Decide with the full record
Your team reviews checks, changes, handoffs, approvals, and the exact worktree result before deciding whether it can merge.
Shared context without shared confusion
Stop copying context between agent windows.
Each worker keeps memory for its own job. MUON also supplies the relevant crew context it needs before editing, including current targets, peer progress, confirmed decisions, duplicate-work signals, and conflict warnings.
Per-agent memoryUnified crew memoryConfirmed-only reuseOptional local Ollama assist
Explore shared memoryUse MUON where you work
One mission record across Desktop, CLI, and TUI.
Move from a visual command center to scripted or terminal control without splitting jobs, approvals, memory, or history across separate systems.
Desktop
Full workspaceMission chat, Crew, shared memory, approvals, changes, governed review, agent sessions, native terminals, and audit activity share one workspace-bound interface.
CLI
Complete control pathOnboard providers, run tasks, apply workflows, inspect status, review approvals, request a second opinion, run checks, and file merge approval from scripts or a terminal.
TUI
Operational parityObserve the same jobs, gates, memory, workflow state, and panic-stop outcome in a keyboard-first interface designed for narrow terminal layouts.
A role for every tool
MUON runs the system. Each integration does a defined job.
Claude Code and Codex do governed coding work. Cursor preserves a human-owned workspace path. Ollama can strengthen local memory recall. The current product does not pretend those roles are interchangeable.
Claude CodeCodexOllamaCursor
Orchestrator or worker
Claude Code
Dispatch-readyMUON can assign bounded implementation, investigation, repair, or review work and create governed child jobs with enforced lineage and limits.
Orchestrator or worker
Codex
Dispatch-readyMUON can assign bounded implementation, investigation, repair, or review work and create governed child jobs with enforced lineage and limits.
Human workspace
Cursor
Readiness-onlyMUON can detect the Cursor agent CLI and preserve a human-owned continuity path. Cursor is not a governed dispatch or QA lane in the current build.
Local recall assist
Ollama
OptionalA loopback Ollama instance can strengthen local memory recall without becoming a coding worker. MUON falls back to built-in lexical recall when it is absent.
Trust model
More agent throughput without giving up control.
Your workspace stays local
Core engineering state stays on the machine your team controls. The working loop does not require a cloud control plane.
Authority stays bounded
Approvals are limited to the exact action, payload, workspace, run, and expiry. Network, merge, and ship actions always return to a person.
Delivery stays reviewable
Checks, changed files, handoffs, approval history, and the reviewed artifact remain connected to the work that produced them.