Hey all,
It is really not possible (or well organized) to respond to individual detailed questions on this thread. Would encourage you
Part of setting up a cloud agent is defining environment variables and secrets!
1 Like
After every action (click, type, scroll), the model gets a fresh screenshot of the screen, analyzes it, and figures out what to do next (take a new action, fix something in the code)
Once it’s confident everything is working through that continuous screenshot-based testing loop, it hits record and performs a clean walkthrough to demo the results (and if the demo unexpectedly fails, it discards the recording, tries to fix the issue, and re-records).
So the model doesn’t analyze the video! The video is a polished demo it records for you after it’s already validated that everything works.
1 Like
This is available for the Enterprise plan.
You should be looking at https://cursor.com/agents, or changing the following to “Cloud”.
Computer use in Cloud Agents is priced the same as a normal Cloud Agent session, and for now, there’s no extra charge for the VM. You’re only billed for the tokens the session consumes, at the same rates as any other agent task.
The key difference is that after each action (clicking, typing, scrolling), a screenshot is automatically captured and sent to the model. This means the model can see exactly what it produced and ideally catch and fix it without you having to describe what went wrong (or test it yourself).
Screenshots do add tokens to each step, so to get the most out of it, it helps to be specific and directive – “Navigate to /settings, toggle dark mode, and confirm there are no error modals” rather than “Check the settings page and see if anything looks wrong.”
We recommend using Skills and AGENTS.md to configure your agent. You can still use Rules!
If you’re facing a specific issue setting up Cloud Agents or think you’ve found a bug, please post over in Help or Bug Reports! Feel free to link here, but it’s not possible to troubleshoot many issues from one thread.
wait what? So not being on enterprise I can configure it personally but on the “Teams” plan i can’t create a shared allowlist for my “Team”? That is absurd…
How did I not know this! Thanks.
Presumably defined per project?
Thanks a lot for these new cloud agents, they’re amazing !!! Already started playing around, both for simple projects but also more intense, and they’re already super versatile !!!
I really want to leverage them at my main job company but for that the environment is missing cgroup memory or cpuset. My understanding is there would be no security aspects to enabling it, but it would allow better container management: Cloud agents missing cgroup memory and cpuset
Can be defined at a team level or scoped to specific repos!
I would really love for this to have multi-repo support - having multiple repost in workspace would allow to setup backend and frontend services and have agent to have full testing environments. Right now it’s a bit limiting for our case.
Also would be really great if API would allow to access shareable links to artefacts like videos so they could be pushed into pull request context. Right now I can’t seem to do that.
I want the local agent control by web. Same like Claude enabled it. Connect to machine running it from web.
I tried to use cloud agent today for the first time. To be honest - very confusing experience.
I used integration with Linear to assign one of issues to Cursor.
I saw that cloud agent was created and it did some work.
When I posted a follow-up question in issue comments, I think it created fresh VM and fresh agent from scratch? I was fully expecting that chat history would be kept, like with local agents. Without chat history preserved it feels like going in circles.
After agent finished work it failed to deploy some of its results - I didn’t know how to configure secrets, and honestly I was expecting Cursor to figure out and properly configure it for me - it has tons of info about my project - but no. So it failed at the end without all the secrets and Supabase access missing etc.
At this moment if would be very convenient to just continue chat with local agent using chat history and same branch. But seems its impossible, chat history can’t be handed-over to a local agent?
Will definitely continue to experiment but onboarding is rough and feature seems to be disconnected from rest of Cursor eg re-configuring same MCPs etc. Ideally I would like to click button and have ai figure things out and configure proper cloud env based on existing local env.
Closing this thread since it has been a while since this was announced. Any new questions, or loose ends from this thread should be filed as new reports in the appropriate forum category!