Cursor lets cloud agents run on machines you manage, with worker pools and sandbox partners

Cursor announced Self-Hosted Machines on 2 September 2026: cloud agents can execute on dynamically scheduled pools of machines inside a customer network. Teams manage the infrastructure; agents are still started and managed from Cursor. Jack Pertschuk writes that this is for control over where agents run and what they can reach—internal services and source control, custom hardware, or operating systems and build pipelines that are hard to package as a Cloud Agent build.

Cursor-hosted environments remain the default. Each default session still runs on a dedicated VM in Cursor’s cloud, with per-agent isolation, secret redaction, egress controls, and signed commits. Internally, Cursor says cloud agents now create more than 60% of the pull requests the company merges.

Only tool execution moves off Cursor’s VMs

With Self-Hosted Machines, only the execution environment moves. The agent loop, inference, and planning stay in the Cursor cloud. The customer machine holds the working copy of the repository, edits files, and runs commands. Tool outputs flow back to Cursor for inference and may contain code; agent transcripts may be processed and stored by Cursor. Teams can still reach cloud agents from the desktop app, cursor.com, mobile, Slack, GitHub, and Linear.

To register a machine, install the Cursor CLI and run agent worker start. That opens a long-lived outbound HTTPS connection to the Cursor cloud. When a session begins, Cursor’s harness handles inference and planning, then sends tool calls to a dedicated worker; the worker returns results for the next inference round. Cursor never initiates a connection into the customer network.

Workers come in two configurations. My Machines connects a single laptop or VM to an account for personal workflows. Pools are a named queue of workers for a team or enterprise. Capacity increases as requests arrive and decreases after workers disconnect.

Pools, hibernation, and sandbox partners

Team Pools are a named routing target: requests wait until an available worker claims them. A controller watches the queue and uses a team-supplied spawn script to start machines as needed. Teams can set an idle timeout so a machine resets and re-enters the pool, or preserve a workspace for follow-ups. Hibernation snapshots and stops an idle machine; if a follow-up arrives within the reconnect window, the snapshot is restored with the same worker ID. Pools are not tied to a single repository.

The docs repeat that the agent loop stays in Cursor’s cloud. Workers need outbound access only—no inbound ports, public IPs, or VPN tunnels. Screenshots and videos are uploaded so they can appear in pull requests and the dashboard.

Self-Hosted Machines does not require a custom sandbox layer. Cursor partners with AWS Lambda, Cloudflare, Coder, Daytona, E2B, Modal, Namespace, and Vercel so workers can start wherever a team’s sandboxes already run. Linux workers now support computer use alongside Macs: with dependencies including Chrome or Chromium, an agent can click, take screenshots, and control the browser. Users can watch the desktop or take control from Cursor.

Cloudflare sandboxes as one execution option

The same day, Cloudflare said Cursor Cloud Agents can run on Cloudflare Sandboxes. Cloudflare’s release matches Cursor’s split: Cursor still runs the agent loop—including inference, planning, and orchestration—while tool calls (terminal, filesystem, and browser) run in a customer-controlled sandbox. It also describes the outbound HTTPS worker model, My Machines versus named worker pools, and starting agents from the Cursor app, cursor.com, or the Cursor mobile app.

On Cursor’s post, AWS Lambda’s Ayush Kulkarni said developers can run agents in their own AWS account while Cursor orchestrates the work. Cloudflare’s Meagan Gamache said: “Developers should Have the flexibility to run coding agents on the platform that best supports their workflow, and companies should not have to compromise on control of where agents run and what they can access.”

Cursor points teams to its docs to connect a machine or configure a pool.

Source: Run cloud agents on machines you manage (Cursor, 2 September 2026). Also used: Team Pools; Cloudflare press release (2 September 2026).