How It Works
From ticket to PR,
on machines you control.
Install the CLI, register a worker, ship work. The orchestrator watches — you stay in control.
Visualize your AI infrastructure — live. Click around to see how it works.
⌘+scroll to zoom · drag to pan · click a star
The Product
Two pillars, one control center.
A workforce of AI agents, and the machines they work on. ClusterCode runs both from the same place, so the work and the infrastructure never drift apart.
The work
AI Workforce
Assign a ticket and it goes to an agent, not to your evening. Claude, Codex, and Copilot pick up your tickets, hand off to each other, and their results come back verified.
- Tickets assigned to Claude, Codex, and Copilot
- Results verified before they reach you
- Approval gates are yours to define
- A launch hands you the machine; a Run delivers the result.
The machines
Infrastructure
Every agent gets its own DevBox: an isolated environment, container or full VM, that you can watch, take over, and throw away when the work is done.
- Disposable DevBoxes, container or full VM
- A live terminal on every one of them
- Observable from the fleet down to a single box
- Your fleet or our cloud, same control center
The Flow
From work arriving to a merge-ready PR.
Two pillars run this path together: an AI workforce that does the work, and the infrastructure that gives it somewhere safe to do it. Every step is tagged below, so you can see which pillar is carrying which beat.
- 01
Work arrives
You paste a ticket, or a Schedule finds it on your board for you.
- 02AI Workforce
Control center assigns an agent
Claude, Codex, or Copilot, picked for the job at hand.
- 03Infrastructure
Agent works in an isolated DevBox
On your own fleet, or on our cloud workers. Same DevBox either way.
- 04AI Workforce
Results verified
Tests, review passes, and visual checks run before the work reaches you.
- 05
Your approval gates
As hands-on or as autonomous as you allow. You set where the gates sit.
- 06
Merge-ready PR
Waiting in your repository, with the whole run history behind it.
Throughput
Ticket in. PR out.
That path is not a queue of one. Runs happen in parallel, up to your plan's concurrency, spread across every worker you have connected.
Get a worker
Your machine or a Cloud worker — one CLI command connects it.
Add your AI agent
Build a DevBox image with Claude Code, Copilot, or Codex — your own key.
Ship work
Paste a ticket URL — or run on a schedule. DevBoxes spin up automatically.
Review the run
Run completes — you're notified in-app and by email. Review the agent's commits and any PRs it opened, then merge.
Getting Started
Get set up in four steps.
Setup is how you join; the flow above is what it does. The clustercode CLI is where you start — it handles login, health checks, and worker registration. The orchestrator dashboard is where you watch, manage, and iterate.
Install the CLI
30 seconds. Node.js 18+ required.
The ClusterCode CLI is your entry point — it handles login, worker registration, onboarding, and health checks. Install it globally once, then run doctor: it checks everything ClusterCode needs, including authentication, worker configuration, container runtime (Podman or Docker), and connectivity to the orchestrator. If anything fails, clustercode onboard walks you through fixing it interactively — including installing Podman, configuring WSL2 on Windows, and selecting your tenant.
npm install -g @clustercode/cli # Check the environment before anything else clustercode doctor
✓ Logged in as you@example.com ✓ Worker registered (My Team) ✓ Podman 5.1.0 — running ✓ Orchestrator reachable (console.clustercode.io)
Register a worker
Any machine with Podman or Docker. Mac, Linux, or Windows.
After login, clustercode worker starts the worker agent on this machine — it connects outbound to the orchestrator over a WebSocket, so no inbound ports or firewall rules are needed. Prefer not to use your own hardware? Provision a Cloud worker from the dashboard instead — it connects the same way, you just don't manage the machine.
# Log in first (opens browser for OAuth) clustercode login # Start the worker on this machine clustercode worker
Starting worker... ✓ Connected to console.clustercode.io ✓ Worker "my-laptop" online
Build a DevBox image
Done through the UI. Pick your engines. Pro+ unlocks full customisation.
Open Dashboard → Images → Build. Choose a target worker (BYOM or Cloud), select Linux or Windows, and pick which AI engines to include — Claude Code, GitHub Copilot, or Codex. Each engine uses your own API key; nothing is bundled. Pro+ users unlock the Advanced section: add language runtimes, toggle extra dev tools, supply a custom base image, or inject arbitrary Containerfile instructions. Hit Build. The image compiles on the worker and appears in your image list within a few minutes — ready for Agent Runs and Schedules.
Ship work
One-off Agent Run, or automated with Schedules + Nova Autopilot.
Paste a ticket URL in the dashboard and hit Run — a DevBox spins up, the AI agent gets full ticket context, and starts coding. Watch it work in the live terminal. You can type corrections, check test output, or close the browser — the DevBox keeps running, and the agent's session is still there when you come back, from any machine. Your logins travel with you too — every DevBox comes up already authenticated, on any machine. Or take the other path: launch a DevBox on its own. Same machine, same ticket context, agent CLI installed — but nothing runs until you start the agent yourself in the terminal. A launch gives you the machine; a Run gives you the result. For automated overnight work: configure a Schedule with Nova Autopilot to query your ticket board for "Ready for Dev" items and launch DevBoxes for each one automatically. When a run finishes, ClusterCode notifies you in-app and by email. Review the commits and any pull request the agent opened, merge what looks good, and the DevBox shuts down on its own, freeing the worker for the next task.
Architecture
Three layers. One flow.
Orchestrator— ClusterCode’s hosted brain. The dashboard you log into at console.clustercode.io. Manages workers, DevBox images, schedules, and Nucleus — the conversational fleet manager you can simply ask (“launch a DevBox for ticket #55”). You don’t self-host it — we run it for you.
Workers — lightweight agents running on machines you control (or our Cloud workers). They connect outbound to the orchestrator over WebSocket — no inbound ports, no firewall changes. One CLI command starts them.
DevBoxes — isolated environments that Workers launch. Each DevBox gets the full ticket context, the AI agent, and your tools. They orbit the worker in the Observatory, pulsing while the agent works.
Ready to ship your first run?
Register your worker in under 5 minutes. clustercode doctor tells you if anything needs fixing before you start.