Agents should work for you, not around you.
ControlClaw is a managed way to run OpenClaw agents for yourself or your team: each agent on its own machine, behind an AI firewall, with approvals and a live audit trail. It is built and operated by Waldos LLC.
The agent got the keys. Nobody kept a copy.
OpenClaw made it easy to give an AI agent an inbox, a browser, a calendar and a terminal. That is exactly what makes it useful, and exactly what makes it risky: an agent that can read your email can also leak it, and one that can click “Pay” can click it twice.
We wanted to keep running OpenClaw as it is, with the plugins and skills we already had, but with the controls a company would expect: isolation between agents, a gate on every outbound channel, a human in the loop for anything irreversible, and a record of what happened.
So we built the missing layer rather than a different agent. ControlClaw provisions standard OpenClaw on isolated machines, routes their traffic through a dedicated firewall, and gives you a console to approve, block and review. Everything underneath stays plain files you can export.
Three things we won't trade away.
Isolation by default
One agent, one machine. A compromised agent should not be able to reach another agent's memory, secrets or sessions.
Humans decide the irreversible
Sending, paying, deleting and logging in pause for a person. Everything else runs, and is logged so you can check later.
No lock-in, ever
Standard OpenClaw, your own model keys, your own OAuth apps. Leave any time with your config, skills and memory intact.
Waldos LLC
A small, independent software company. ControlClaw is our main product and is developed in the open.