Hi! I’ve noticed that my Automations sometimes trigger models different from the one configured, probably through subagents or some internal process.
All of my active Automations use Grok 4.7 or Composer, so I expect them to consume Cursor Models. However, in Usage I can see calls to Claude and Gemini, which are consuming my Other Models quota.
Today I got a message saying I had hit my usage limit and Cursor granted me extra Cloud Agent credits (thank you!!), but what confused me is that Other Models is at 100% while Cursor Models is only at 22%.
My Automations and runs are showing Grok 4.7, so I’d like to understand if this behavior is expected and if there’s a way to make sure Automations use only Cursor Models.
Yes, this is expected right now. Your automations do run on Grok 4.7, but when an automation tests its own work (opening a browser, taking screenshots or recording a demo), it hands that step to a built-in computer use helper that runs on Claude, and reviewing the recording runs on Gemini. Those helpers don’t follow the automation’s model yet, which is why you see Claude and Gemini rows with the cloud icon. We’re working on moving these workloads to first-party models when possible!
If you want your automations to stay on Cursor Models:
Open the Cursor dashboard and go to Cloud Agents, then Environments.
Open the environment your automations use.
Turn off Testing (on some dashboards it’s in the … menu as “Disable Testing”).
With Testing off, agents in that environment won’t open a browser, take screenshots or record videos, so those Claude and Gemini helpers aren’t used. The trade-off is that you lose those screenshots and demo videos on your runs.
