Cursor's self-hosted machines run cloud agents inside your network

Keep code and secrets on your own AWS, Cloudflare, Modal or Vercel boxes; team pools autoscale and hibernate; agents get computer-use on Linux and Mac.

Nowline SEP 7 2:00 PM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • Your codebase never leaves your network

    Cursor's self-hosted machines run the agent's tool calls — clones, builds, tests — on infrastructure you control, so source, build outputs, and secrets stay in-network instead of on Cursor's cloud.

  • Run it on compute you already pay for

    Supported targets include AWS Lambda, Cloudflare, Modal, Vercel, Coder, Daytona, Namespace, and E2B. 'My Machines' connects a single laptop or VM; 'Team Pools' are named worker queues that grow as requests arrive and shrink when they stop.

  • Idle pools hibernate, so you stop paying to stay warm

    Pool workers can hibernate when idle and restore within a reconnect window when the next prompt lands — no keeping expensive VMs or GPUs warm just for the next turn.

  • Agents get computer-use on your own Linux and Mac

    Self-hosted workers can now click, type, take screenshots, and drive a browser on Linux and Mac, and you can watch or take manual control from inside Cursor — handy for end-to-end tests and browser flows that need your real, logged-in environment.

  • Elsewhere: the Vercel AI SDK can recover mid-stream

    The AI SDK added a streamRetries option: when a provider errors after streamText has already begun sending tokens, you can catch it and retry just that leg instead of dropping the whole response — one less way a flaky provider breaks a live UI.