Another AI approves what your agent does
Claude Code and Cursor come set up so that a classifier model decides which actions go ahead, instead of asking you each time.
Claude Code · Cursor
What has changed
AI now works on your computer and in the cloud. See what it can do without asking your permission.
AI agents no longer just answer. They now read your files, run programs and act in services you use, like your email, your documents or GitHub.
Some do it without asking your permission.
I want to understand itWhat each agent does, explained without jargonI want to check my computerFree · Windows, Mac and Linux
It's open source. RosmarOps only reads your configuration, and nothing leaves your computer unless you approve it.
Is this for me?
If you use any of these agents, yes.
They are agents that work inside your computer or your code editor. RosmarOps recognizes them and shows you what they can do there.
On your computerClaude CodeCursorHermes AgentCodexGrok BuildAntigravity CLIGitHub Copilot in VS CodeCline in VS CodeGemini CLI
In the cloud, from your spaceClaudeChatGPTGeminiGrokCodexGitHub CopilotComet
Do you use ChatGPT, Claude or Gemini in the browser? This is for you too. In your space you can go through their settings with a few simple questions: which of your services they have connected, whether they can act on their own and whether they train on your conversations. It gives you a score, calculated in your browser. Go to your space
Is it true?
What they do as they come installed
We checked them one by one, on the same reference computer, reading their code and their official documentation. It changes with every version, and we review it with each of ours.
One file in your project gives instructions to several agents at once
It's called AGENTS.md, and many agents read it. If one changes it, it changes what the others do.
Cursor · Codex · GitHub Copilot · Cline · Grok Build · Antigravity CLI
Claude Code and Hermes Agent also read it in some cases.
They search the internet without asking your permission
Web search comes turned on by default. A search can send out part of what the agent has read.
Gemini CLI · Grok Build
Some keep their permissions where they can't be reviewed
Their permissions live inside the app, not in a file. Neither you nor RosmarOps can check them on disk, so RosmarOps lists them but doesn't score them, so as not to guess.
Claude Desktop · Antigravity app · Copilot in JetBrains
And on my computer?
What you see above is how they come. Yours may be different.
It depends on how you have set them up, on the keys stored on your computer and on the other agents you use. RosmarOps gives you a score for each agent, the first thing worth changing and how much it improves if you do.
Free and open source (AGPL-3.0). How to check it's authentic
Windows may warn that RosmarOps is an unrecognised app: it is new and not signed yet. Click «More info» and then «Run anyway».
Example · made-up computer
Antigravity CLI · ExposedFrom 0 to 100. The higher, the less exposed.
The first thing worth changing
Your Stripe key is saved in git
With the exact change, ready to paste into your agent.
If you change it, the score goes up from 11 to 13
See the full report
How does it work?
Three steps, a few minutes
Download it and open it
There's nothing to install and no account to create.
It reads your configuration
It looks at what your agents can do on your computer, without changing anything.
It tells you what to change
A score for each agent and the exact change, in order of importance.
Can I trust it?
Three promises you can check
It only reads
It doesn't change your files or your agents. It tells you what to change, and you decide.
Nothing leaves without your permission
Without your permission it doesn't connect to anything. With it, it only sends your results, encrypted, to your space.
Open source
You can read the code. Every version comes with its checksum and an attestation of how it was built, so you can check that it's ours.
And after that?
Follow how your score changes, across all your computers
Your space keeps every check and shows you what has improved, what is new and what is worth doing next. It's free, in open beta.
Example · made-up computers
- a3f91c02Windows64
- 7be04d19Windows71
- c25e8a60macOS88
Start with your computer
How we calculate the score: methodology. What it doesn't look at: limits.