Which agent program runs on it?+
OpenClaw, an open-source program that runs an AI agent. It is what we have run our own agent on since February 2026, and the version on your machine is pinned, so an update never arrives unannounced. We are rebuilding our own machine into the image customers get. The program can be swapped later if something better arrives; what you are paying for is the machine, the backups and the person on the ticket.
Whose model key is it?+
Yours. You pick the provider and pay it directly. Before you connect the key you set a monthly spend limit in that provider's console, and we do it with you on the handover call.
What does the model itself cost?+
That is your provider's bill, not ours, and we never resell it. A light workload for a small team runs about 10 to 80 dollars a month on a mid-range model, and can pass 140 dollars a month on the most expensive tier. The spend limit is how you cap it.
Why not a cheaper server with the agent installed on it?+
You can run it yourself. The software is open source and the install is a weekend. This plan is the patching after a snapshot, the rollback when an update fails, the backups, and a person who answers when it stops working. That is the part that costs money.
I read that these agents get hacked. What is different here?+
Most of those stories are an agent left open to the internet on somebody's cloud account. Ours has no public control port: it listens only on the machine itself. The instruction files it runs from are locked, so a message cannot rewrite them. The machine is backed up nightly, so a bad day is a restore. It is not magic. An agent you give write access to can still do what it was asked to do, which is why writes wait for your approval.
What counts as a ticket?+
A scoped request about the machine, the agent program, the backups or a rollback, worked by our desk with the logs open. Your prompts, your skills and what the model says are outside the plan; those are a setup package or a maintenance ticket.