Sign InOpen Brain
CursorEngineering PostOfficial Source

Run cloud agents on machines you manage

Cursor cloud agents can now execute tools on machines and elastic pools inside your network while Cursor retains planning and inference. This unlocks internal access and custom hardware, not full self-hosting.

Cursor · Sep 2, 2026
Open Source Open MarkdownOpen JSON
Source Summary

Cursor’s **Self-Hosted Machines** move repository access, file edits, and commands onto customer-managed workers. A worker opens outbound HTTPS to Cursor, receives tool calls, and returns results; pools can scale from a queue and serve multiple repositories.

Practical Implication

Use this when agents need internal services, private source control, GPUs, Macs, Kubernetes, or unusual build pipelines. Linux workers now support **computer use**, and idle machines can reset, preserve a workspace, or resume from a hibernated snapshot.

Agent-Ready Context
Cursor’s **Self-Hosted Machines** move repository access, file edits, and commands onto customer-managed workers. A worker opens outbound HTTPS to Cursor, receives tool calls, and returns results; pools can scale from a queue and serve multiple repositories.

Use this when agents need internal services, private source control, GPUs, Macs, Kubernetes, or unusual build pipelines. Linux workers now support **computer use**, and idle machines can reset, preserve a workspace, or resume from a hibernated snapshot.

Only execution moves. **Inference, planning, and the agent loop remain in Cursor’s cloud**, while tool output can include code and transcripts may be processed or stored there. Teams still need to evaluate that boundary, worker isolation, capacity controls, and recovery behavior.
Connected Context · Feed7 Judgment

This expands Cursor Cloud Agents into environments that managed sandboxes may not reach—private services, GPUs, Macs, Kubernetes, and custom build systems—without moving the agent loop out of Cursor’s cloud. It therefore changes execution control more than data residency: teams gain worker and workspace lifecycle choices but still must assess tool-output exposure, isolation, capacity, and recovery.

Cursor Cloud Agents can now run in Vercel SandboxThese are two execution backends for the same Cursor orchestration boundary: customer-managed workers maximize environment access, while Vercel supplies per-request microVM isolation and managed lifecycle.What we’ve learned building cloud agentsThe emphasis on complete development environments explains the value of customer-managed workers, while the reported reliability work makes queueing, recovery, and workspace lifecycle concrete evaluation requirements.Expanding Managed Agents in Gemini API: background tasks, remote MCP and moreBoth support long-running agents with persistent execution state, but Cursor places that state on customer workers while Gemini manages sandbox continuity around asynchronous tasks and credential rotation.Introducing organizations for Cursor EnterpriseSelf-hosted pools introduce capacity and multi-repository allocation decisions that complement Cursor Enterprise’s existing team-level budget, model-access, and usage controls.
Context Map
toolscoding#cloud-agents#computer-use#enterprise
Uncertainty
Only execution moves. **Inference, planning, and the agent loop remain in Cursor’s cloud**, while tool output can include code and transcripts may be processed or stored there. Teams still need to evaluate that boundary, worker isolation, capacity controls, and recovery behavior.