Summary
On September 2, 2026, Cursor released self-hosted machines for Cloud Agents, letting teams keep tool execution entirely inside their own network. The feature ships with dynamic pool scheduling for team workloads, multi-provider support (AWS Lambda, Vercel, Modal, and others), and computer-use capabilities on both Linux and Mac hosts.
What changed
Cursor added a self-hosted execution mode for Cloud Agents. Instead of running agent tool calls in Cursor-managed infrastructure, teams can route them to pools of their own machines across AWS Lambda, Vercel, Modal, and other providers, with a dynamic scheduler balancing workloads across the pool. Computer-use style automation now works from those self-hosted Linux and Mac hosts.
Why it matters
The two biggest reasons enterprises won't let Cloud Agents anywhere near production repos are code egress and secrets exposure. Pinning agent tool execution to customer-controlled hosts removes both objections in one shot, and multi-provider pool scheduling means teams don't have to standardize on one runtime. It also matches how Anthropic (Enterprise Frontier Safeguards) and Cloudflare (Identity-Aware AI Gateway) have moved: give the customer the trust boundary and let the AI vendor operate inside it.
Evidence excerpt
Self-hosted machines: Cursor introduced support for self-hosted infrastructure, allowing teams to keep tool execution entirely in your own network. Key features include dynamic pool scheduling for team workloads, support for multiple cloud providers (AWS Lambda, Vercel, Modal, etc.), and computer use capabilities on Linux and Mac systems.