Control plane for AI coding agents
Run coding agents across your machines. Ship only what's verified.
Agent Orchestration schedules coding tasks onto your own workers, runs Claude Code, Codex, Gemini CLI, OpenCode or Aider there, recovers from limits and crashes, and completes a task only after your tests, type checks and build pass.
Self-hostable · latest release v0.1.1 · License pending
Control plane
queue · scheduler · policies · audit · API
Outbound HTTPS and WebSocket from each worker
Worker · laptop
macOS
web-app
project
Claude Code
agent
Own login (Claude)
provider
Worker · build-01
Linux
api
project
Codex
agent
OpenAI API
provider
Worker · desktop
Windows
mobile-app
project
OpenCode
agent
Ollama (local)
provider
The problem
Coding agents are powerful. Running them is the hard part.
Using one agent in one terminal works. Keeping several agents busy across projects, machines and model limits — and trusting what they produce — needs an orchestration layer.
Usage limits stop work mid-task
A subscription or API limit ends the session, and the task waits until someone notices.
Agents live on one laptop
Your repositories, machines and build servers are elsewhere; agents only run where you started them.
Context runs out
Long tasks exhaust the context window and lose track of what was done.
Agents crash and hang
A process dies or stalls, and nobody restarts it from where it was.
Nobody checks the result
"Done" is the agent’s claim. Tests, types and the build are run by hand, if at all.
No shared queue or history
Tasks, decisions and outcomes are scattered across terminals and chat windows.
How it works
Every task takes the same path
You describe the change. The platform decides where and with what it runs, keeps it going, checks it, and delivers it the way your Git policy says.
- 01TaskWhat to change and how to check it
- 02ProjectOne or more repositories
- 03WorkerA machine with the code
- 04AgentClaude Code, Codex, Gemini…
- 05ProviderOwn login or add-on model
- 06CapabilitiesSkills, MCP, plugins
- 07ExecutionCheckpointed session
- 08VerificationYour tests, types, build
- 09RecoveryLimits, crashes, lost workers
- 10GitCommit, push or pull request
Features
Everything between “do this” and a verified branch
Task orchestration
Queue, prioritise and schedule coding tasks onto the right machine.
Worker fleet
Run agents on your laptops, desktops and build servers from one control plane.
Multiple agents
Claude Code, Codex, Gemini CLI, OpenCode and Aider behind one agent-neutral core.
Multiple providers
Keep tasks running on add-on models when an agent reaches its usage limit.
Automatic recovery
Survive usage limits, exhausted context, crashes, hangs and lost machines.
Capabilities
Skills, MCP servers and plugins, governed by permissions and approvals.
Verification
A task is complete when your checks pass — not when the agent says so.
Git automation
Commit, push or open a pull request — by policy, and never destructively.
Projects and integrations
Projects with one or many repositories, fed by GitHub, GitLab and webhooks.
AI project manager
Break a large goal into dependent, verifiable tasks — then create them with one click.
From the terminal
Create, plan and review from where you work
The dashboard, the agentctl CLI, the VS Code extension, the mobile app and the REST API all drive the same queue.
- Sign in with any method through the browser
- Plans that become dependent tasks
- Reviews of any branch or pull request
- API tokens for CI and scripts
Terminal
$ agentctl task create --project prj_api --title "Validate emails" \ --prompt "Reject invalid emails on sign-up and show the error. Add tests."Created tsk_… QUEUED$ agentctl task create --project prj_api --plan --title "Plan: multi-currency" \ --prompt "Support EUR and GBP in checkout, invoices and reports."$ agentctl task create --project prj_api --title "Review payments" \ --review-head feature/payments --review-base main --prompt "Focus on idempotency."$ agentctl logs tsk_… Verification passed · Git commit on ao/validate-emails-… · Completed and verified
Two ways to run it
The same platform, operated by you or by us
Self-hosting is the complete product, not a trial. The managed cloud sells operations: hosting, backups, updates and monitoring.
Self-hosted
Your infrastructure, full control
Docker Compose, Kubernetes or plain Node.js. No license server, no telemetry, no limits. Your credentials and capability registry never leave your network.
Managed cloud
We run the control plane
Sign up and pair your machines. We operate the control plane, database, backups, updates and monitoring. Your workers, code and provider keys stay with you.
Ecosystem
Agents, providers and capabilities, documented
Directories of what the platform works with — each entry with compatibility, installation, permissions and a manifest you can register in your own installation.
Who it’s for
From one developer to an engineering organization
Individual developers
Keep your agents working across your own machines, and come back to verified branches.
Software teams
One queue, shared workers, roles, automatic reviews and a history of every task.
Agencies
Many clients and repositories, each in its own project with its own policies.
Enterprises
Self-host on private infrastructure with SSO, audit logs, approvals and policies.
Questions
Straight answers
- Is Agent Orchestration an AI coding agent?
- No. It orchestrates the agents you already use — Claude Code, Codex, Gemini CLI, OpenCode and Aider — on your own machines, and adds scheduling, recovery, verification and Git on top.
- Does my code leave my machines?
- Agents run on your workers, in your project folders. The control plane never clones or stores your repositories; it stores task metadata, timelines, reports and artifacts. Agents talk to the model providers you configure.
- Do I need API keys?
- No. By default every agent uses its own login, such as a Claude subscription. API keys are only needed for optional add-on models.
- Is the self-hosted version limited?
- No. Self-hosted installations have no limits on tasks, workers, projects or members, make no calls to vendor infrastructure, and include every feature. The managed cloud sells operations, not features.
Put your agents to work — and trust what they deliver.
Start with the managed cloud, or self-host the complete platform in minutes.