Limits and usage
Every workspace has limits on what it can use in a day and at once. They keep one workspace from using up capacity that everyone shares, and they keep a runaway agent from running up costs.
Why there are limits
Groundcrew's servers run builds, agents, and language model calls for every workspace. Limits make sure a busy or misbehaving workspace slows down on its own instead of slowing down everyone else. They also cap what an agent can spend before a person looks at it.
Workspace limits
| Limit | Default | Notes |
|---|---|---|
| Runs going at once | 2 | Runs a person started: deploys, discovery, and investigations of reported incidents. The watcher's investigations do not count. |
| Deploy builds | 10 a day | Each deploy that reaches the build stage counts once, however many attempts Builder makes. |
| Longest build attempt | 15 minutes | An attempt that takes longer is stopped, and its build sandbox with it. |
| Web research | 10 a day | Medic's web searches during investigations. |
| Connections | 10 | All kinds together. |
| Resources per connection | 2,000 | Discovery keeps the first 2,000 and says so. |
Daily limits reset at 00:00 UTC.
Model spend
Agents call NVIDIA Nemotron models through Nebius Token Factory: Scout on Nano, Builder and Deployer on Super, and Medic's root cause on Ultra. Every model call shows its model, tokens, and cost on the run, and each deploy and incident shows its total.
Each workspace has a daily cap on model spend. By default it is the platform's daily cap. An owner can set a lower cap for the workspace in Settings, Approvals and limits, but cannot raise it above the platform's. Once the workspace reaches its cap, agents in it stop calling models for the rest of the UTC day and say why.
The platform also has overall caps. If one is reached, agents everywhere stop with a message saying so, until it resets.
Seeing your usage
Settings, Usage shows today's model spend against the cap, and each limit with what the workspace used of it. The Overview page shows model usage for the day as well.
When a limit is reached
Groundcrew refuses the new run or connection with a message that names the limit and when it resets, for example: This workspace reached its limit of 10 deploy builds today. It resets at 00:00 UTC.
- A deploy that reaches its build stage after the build limit is used up stops there with that message. Nothing is built or pushed.
- The incident watcher keeps opening incidents, so you still hear about outages.
- When web research is used up, Medic investigates without it and the incident says research was skipped.
Request limits
Per person, the service also limits:
- 300 requests a minute.
- 10 live run streams open at once.
- Request bodies of 1 MB. A kubeconfig or env var list larger than that is refused.
New sign-ups are limited per hour and per address, and can be paused for a while if needed. People invited to a workspace can always join.
Need more?
Limits can be raised for one workspace by the people who run Groundcrew. Ask through the contact on the Security page, with your workspace name and which limit you need raised. When you run your own Groundcrew server, every limit is a setting; see the repository's self-hosting guide.