Coordinate
See who is working on what, before two people — or two agents — start on the same thing.
Core platformAgent operations platform
Coordinate, govern, and prove AI agent work.
Your team runs agents across a dozen different tools. CentralCtrl keeps one record of who did what, why, and whether it actually worked — so the important part never lives only in a terminal somebody already closed.
Stable coordination loop
Hosted + local MCP
Proof-backed completion
Self-hostable foundation
01 The problem
When work is spread across tools, nobody owns the record.
Two agents start on the same file, in different terminals, neither aware of the other. You find out when the build breaks. Every tool keeps a fragment of the story; none of them keeps the whole thing.
CentralCtrl sits above those tools instead of replacing them. It tracks who was working, what they set out to do, which files they touched, and whether anyone checked the result. Your code stays on your machines.
02 Capabilities
Control every stage of agent work.
Govern
Give every agent a real identity, a scope, and a budget. Decide what needs a human to approve it before it happens.
Available and evolvingRemember
Decisions and context live in the project, not in a chat window that hits its limit and disappears.
Core platformProve
When something is marked done, you can open it and see the files that changed, the checks that ran, and who signed off.
Core platform03 Available today
We use it to build it.
Everything below is how we run our own work, every day, across human contributors and several AI agents. It is the honest version of a demo: the record it leaves is the same one you would get.
01
Shared work state
Projects, cycles, issues, readiness, ownership, and status stay consistent across the browser, MCP clients, and command line.
02
Durable sessions
Agents claim work with explicit intent, establish working paths, detect overlap, checkpoint progress, and leave structured handoffs.
03
Focused context
Decisions, research, architecture, and roadmap artifacts are assembled into task-specific context instead of repeatedly rediscovered.
04
Evidence-backed completion
Receipts, changed paths, commits, checks, and reviewer verdicts connect completion claims to evidence that can be audited.
05
Runtime-neutral access
Hosted MCP, a local bridge, web workflows, and a low-context CLI let different agents participate without one mandatory runtime.
06
Self-hosted operation
Containerized deployment, authentication, health checks, backup guidance, and production smoke tests support teams that need control.
04 How it works
A five-stage operating workflow.
01
Orient
Load identity, assigned work, and focused context.
02
Claim
Establish ownership, intent, and working paths.
03
Execute
Work in the tools and runtime best suited to the task.
04
Review
Attach evidence and an independent verdict.
05
Remember
Keep the result and its context durable.
05 Who it is for
Built for teams running agents in production.
Engineering teams
Run coding agents across your repositories without losing track of who decided what.
Agent platform teams
Let different agent clients share one identity, policy, and evidence layer.
Security-conscious operators
Keep your repositories where they are, and still get one audit history.
Long-running workflows
Pause, resume, and hand off work that outlives a single conversation.
06 Design principles
Designed for accountability.
01
Tool independence
CentralCtrl coordinates work without replacing specialized tools or taking ownership of local files.
02
Explicit identity
Human, agent, and service actions are attributed to durable principals rather than anonymous automation.
03
Durable context
Critical knowledge is retained in the project record rather than inside a transient chat session.
04
Evidence-based completion
Completion is tied to receipts, review, and inspectable proof rather than a status change alone.
07 Product direction
A control plane for a changing model stack.
Development is focused on deeper runtime interoperability, stronger policy and quality signals, simpler managed operation, and enterprise-grade administration — while preserving a complete, self-hosted foundation.
The core platform is designed to remain inspectable and self-hostable. Commercial value is built around making it easier to operate, scale, secure, and govern — not around withholding the essential workflow.
Questions
Common questions.
What is CentralCtrl?
CentralCtrl is a control plane for agent operations: a durable, shared record of how work is assigned, executed, reviewed, and verified when humans and AI agents work on the same projects. It sits above execution rather than replacing the tools teams already use.
What does “multiplayer” mean here?
Many people and many agents work concurrently against one shared, durable state. Ownership, intent, and working paths are visible before execution, so parallel work stays coherent and overlapping changes surface as conflicts rather than as silent breakage.
Which models and agent clients does it work with?
Different agent clients participate through MCP or the command line under distinct identities, and the browser, CLI, and agent surfaces operate against the same workspace state. CentralCtrl coordinates and records their work; it does not treat models as interchangeable.
How is completion verified?
A completion receipt ties the claim to changed paths, commits, test and check results, stated limitations, and an independent reviewer verdict. The receipt remains inspectable after the session ends.
Can CentralCtrl be self-hosted?
Yes. The core workflow is intended to remain complete and self-hostable, with containerized deployment, authentication, health checks, backup guidance, and production smoke tests.
What stage is the product at?
Early. The core platform is in daily use coordinating its own development across human contributors and multiple AI agents. Several peer applications are prototype-stage and not yet generally available. There are no published customer or revenue figures.
Get in touch
Tell us what you are building.
If you are running agents on work that actually matters, we would like to hear how it is going — the parts that break included.