Cursor cloud agents can run on your machines
Cursor now lets its cloud coding agents execute on machines inside a customer's own network, keeping planning and inference in Cursor's cloud while tool calls run next to internal services, source control, and custom hardware.
Cursor's Self-Hosted Machines feature moves the execution environment for its cloud agents out of Cursor's cloud and onto machines a customer runs. The agent loop, inference, and planning stay in Cursor's cloud; only tool calls are dispatched to a worker in the customer's environment, which holds the repository working copy, edits files, and runs commands.
Registration is via the Cursor CLI: agent worker start opens a long-lived outbound HTTPS connection, and Cursor never initiates a connection into the customer network. Workers can be attached individually ("My Machines") or grouped into pools — named queues that scale up as requests arrive via a controller watching the queue and a team-supplied spawn script, and scale down as workers disconnect. Pools are not tied to a repository, so one pool can serve many. Idle machines can be hibernated by snapshot and restored with the same worker ID if a follow-up arrives inside the reconnect window.
Cursor names AWS Lambda, Cloudflare, Coder, Daytona, E2B, Modal, Namespace, and Vercel as sandbox partners. Linux workers now support computer use alongside Macs, given Chrome or Chromium and the required dependencies; Namespace Devboxes provide Apple silicon Macs for iOS and macOS builds.
Cursor-hosted VMs remain the default. The company says tool outputs flow back for inference and may contain code, and that agent transcripts may be processed and stored by Cursor. It also states that cloud agents now create more than 60% of the pull requests it merges internally. Pricing was not disclosed in the announcement.