Limits
Every organization has limits — how many accounts, clusters, servers, people and running operations — and agents have an hourly budget of tokens. Most of them grow as you do.
What is limited
- Cloud accounts, clusters, servers and members. Servers are the machines Rockslab runs for you, counted per request: a build is its control plane plus its workers, and an autoscaled pool counts at its ceiling. Machines you made yourself in your Hetzner project do not count.
- Running operations. How many builds, installs and syncs may run at once (two by default). One more is allowed for every two clusters you run.
- Agent tokens. What scripts and agents may spend in an hour — see Agent tokens below.
Billing & plan › Usage & limits shows each limit, what is used against it and where its value comes from. Taking things away is never refused.
When you hit one
The cabinet says what is used, what the limit is and what the request would add — then the one next step that applies to you: a larger plan, or asking us for more.
When every operation slot is busy, a new operation is not refused: it queues and starts by itself when one of yours finishes. It shows as queued with its place in line.
Ask for more
Billing & plan › Usage & limits, next to the limit, press Ask for more…. Owners and admins can ask. Say how much and why; a person at Rockslab decides — sometimes for more than you asked, never silently.
- One open request per limit at a time.
- A limit you keep running near for a day may be raised automatically, within your plan's reach.
- A granted value is a floor: as you grow, the formula can lift it further.
Agent tokens
Every call a script or agent makes with an API token — through the API or the MCP server — costs agent tokens. People clicking in the cabinet spend none.
| Call | Example | Tokens |
|---|---|---|
| free | remove, stop, cancel, revoke | 0 |
| read | anything that changes nothing | 1 |
| change | rename, labels, settings | 5 |
| operation | anything that starts a job | 50 |
| heavy | build a cluster, add or grow a pool, deploy a bundle | 500 |
Each call is paid from two rolling hourly budgets: yours, and the organization's, shared by every agent in it. Both grow with the size of your infrastructure. Spent tokens drip back over the hour rather than all at once on the hour. The cabinet sidebar shows both and when they refill.
A call refused by the platform's limits costs nothing. A call refused because the request itself was wrong costs a small attempt fee from your budget only, so a runaway loop hurts you, not your colleagues. Need a bigger personal budget? Ask support.