# Cursor Cloud Agents can now run in Vercel Sandbox

Source: [Vercel](https://vercel.com/changelog/run-cursor-cloud-agents-vercel-sandbox)  
Feed7 permalink: https://feed7.dev/p/run-cursor-cloud-agents-vercel-sandbox-0l3bo4n  
Published: 2026-09-03T15:00:00.000Z  
Trust: Official Source (official_source)

## Why Included

Cursor Cloud Agents can execute inside per-request Vercel microVMs, giving teams control over isolation, credentials, retries, and worker lifecycle without operating persistent VMs.

## Source Summary

Cursor now lets Cloud Agents run on **Vercel Sandbox** through its Self-Hosted Machines APIs. Cursor retains the harness and inference loop, while each request gets an isolated **Firecracker microVM**; Vercel Functions and Workflow handle queues, provisioning, monitoring, retries, and cleanup.

## Practical Implication

This separates agent orchestration from execution infrastructure. Teams needing tighter runtime control can use scale-to-zero workers and short-lived, user-scoped credentials instead of maintaining a standing VM pool.

## Agent-Ready Context

Cursor now lets Cloud Agents run on **Vercel Sandbox** through its Self-Hosted Machines APIs. Cursor retains the harness and inference loop, while each request gets an isolated **Firecracker microVM**; Vercel Functions and Workflow handle queues, provisioning, monitoring, retries, and cleanup.

This separates agent orchestration from execution infrastructure. Teams needing tighter runtime control can use scale-to-zero workers and short-lived, user-scoped credentials instead of maintaining a standing VM pool.

Self-Hosted Machines requires a **Cursor Enterprise** plan, and the source describes a reference implementation rather than its operating cost or performance under load.

## Connected Context

Feed7 judgment across 691 accumulated Signals:

This turns the brain-versus-hands separation into a concrete Cursor deployment: Cursor keeps orchestration while Vercel supplies per-request isolated execution and lifecycle machinery. It reinforces per-agent sandboxing and user-scoped identity as practical controls, while narrowing applicability to Enterprise customers. The reference architecture still leaves load performance and operating economics unproven.

- [Scaling Managed Agents: Decoupling the brain from the hands](https://feed7.dev/p/managed-agents-00zpz67) — Both separate the agent harness from replaceable execution environments, independently reinforcing decoupled orchestration and sandbox infrastructure as a managed-agent architecture.
- [Give every agent in Herdr its own Vercel Sandbox](https://feed7.dev/p/give-every-agent-in-herdr-its-own-vercel-sandbox-1p4i553) — Both assign isolated Vercel sandboxes to individual agent work, but Herdr additionally returns reviewable Git patches, whereas the Cursor signal focuses on hosted provisioning and lifecycle operations.
- [Tethered: Our Agents Are Us — Shu Fang, Two Sigma](https://feed7.dev/p/tethered-our-agents-are-us-shu-fang-two-sigma-0c8u33i) — Cursor’s short-lived user-scoped credentials align with Two Sigma’s emphasis on preserving real-user attribution, though the candidate also adds trace and egress controls not established here.

## Context Map

- Layer: infra
- Domains: coding, security
- Topics: cloud-agents, sandboxing, coding-agents

## Uncertainty

- Self-Hosted Machines requires a **Cursor Enterprise** plan, and the source describes a reference implementation rather than its operating cost or performance under load.

## Agent Instruction

Use this item as source-backed context. Do not invent claims beyond the linked source. If this item conflicts with another source, call out the conflict.