Cursor can now run Cloud Agent tool execution on machines you manage. That is what Self-Hosted Machines (My Machines) actually are.
Inference still lives in Cursor's cloud. The agent loop still lives in Cursor's cloud. Your repo, file edits, and terminal commands can run on your box.
That is the whole product. Most SaaS teams do not need it yet.
Who should care
You should care if Cloud Agents need to touch code that cannot leave your network: customer data in fixtures, private packages, or a locked-down CI image.
You should also care if legal already asked where agent commands execute.
Who should ignore it
If your team is still leaking secrets into the client, skipping RLS, or running Cloud Agents against a public repo with no review, this feature is a distraction.
Self-hosted machines do not fix a leaked service_role key. They do not write your policies. They do not review the PR.
What changes for SaaS teams
- Tool execution can sit next to your existing runners.
- Model calls still leave the building.
- You now have another machine to patch, observe, and rotate credentials on.
Treat it like adding a CI runner that can write to the repo, not like going air-gapped.
I wrote the longer version for SaaS teams deciding whether this is a 2026 infrastructure change or just a changelog item: Cursor Self-Hosted Machines.
If the bigger risk is the database, start here: Supabase RLS for vibe-coded SaaS.
Top comments (0)