One control plane runs your whole business as a fleet of agents — automated assistants that carry out tasks for you — on a server you own. Each module is built on a proven open-source tool and adds agents on top; the kernel coordinates them, sends every task to the cheapest capable AI model, and pauses anything risky for your one-click approval. Everything below is running live on demo data for a fictional tenant, Summit Roofing Co.
A simple config file (no coding needed) that lists every module, the agents it runs, and which actions must pause for your approval.
The coordinator: starts the modules, gives each its agents, runs them on a schedule, and drives workflows that span several modules.
Sends every task to the cheapest model that can do it well — a model on your own hardware first, a premium one only for the hard 5%.
The shared memory of your business — the profile, customers, and policies every agent works from.
The human-in-the-loop gate: anything that moves money, touches compliance, or changes infrastructure pauses here for your one-click sign-off, and every decision is logged.
Each module's core is the open-source tool it is built on.