# Vercel Sandbox

> Ephemeral Firecracker microVMs on Vercel for untrusted and AI-generated code — millisecond startup, Node and Python runtimes, persistent by default.

Vercel Sandbox (GA January 2026) runs untrusted and AI-generated code in ephemeral Firecracker microVMs: millisecond startup, Node and Python runtimes with sudo, sandboxes persistent by default via automatic filesystem snapshots, up to 2,000 concurrent on Pro. The SDK and CLI are open-source Apache-2.0; Hobby gets a real free monthly allotment, Pro is usage-billed.

Website: https://vercel.com/docs/sandbox

Vercel Sandbox is the platform answer to the agent-code-execution problem: if your stack already lives on Vercel — AI SDK apps, v0 output, Next.js products — the sandbox is *right there*, with the same OIDC auth, billing, and SDK ergonomics as everything else you deploy.

## Highlights

- **Firecracker isolation** — each sandbox is a microVM with its own filesystem and network; sandboxed code can't touch your environment, data, or cloud resources.
- **Real runtimes with root** — Node 26/24/22 and Python 3.13 on Amazon Linux, sudo included: package installs, Docker-in-sandbox, even VPN clients and FUSE.
- **Persistent by default** — automatic filesystem snapshots on stop; resume by name and skip the reinstall; explicit snapshots and beta Drives for attachable storage.
- **Serious ceilings** — millisecond startup, timeouts to 5 hours, 32 vCPUs/64GB at the top tier, 2,000 concurrent sandboxes on Pro.
- **Open SDK + CLI** — `@vercel/sandbox` (and a Python SDK) open-sourced Apache-2.0 at GA, with a CLI for scripting fleets.
- **Honest free tier** — Hobby includes monthly Active-CPU hours, creations, and storage at no charge (it pauses rather than bills when exhausted).

## In an AI-assisted workflow

```bash
npm i @vercel/sandbox     # auth via your linked project's OIDC: vercel link && vercel env pull
# const sandbox = await Sandbox.create(); await sandbox.runCommand("python", ["analyze.py"])
```

The canonical loop: your agent (likely on the [AI SDK](/tools/vercel-ai-sdk)) generates code → executes it in a sandbox → reads results as observations. Billing nuance worth knowing: I/O wait isn't billed as Active CPU, so long-running-but-idle agent sessions cost less than wall-clock suggests.

> [!NOTE]
> Two setup quirks: it currently runs in a single region (`iad1`), and auth wants a linked Vercel project even if you deploy nothing. And remember persistence-by-default means snapshots accrue storage — clean up or opt out for throwaways.

## Good to know

Beta June 2025, **GA January 30, 2026**, with v0, Blackbox AI, and Roo Code cited in production. The ecosystem gravity is the real differentiator — outside Vercel, [E2B](/tools/e2b) (code-interpreter ergonomics, open infra), [Daytona](/tools/daytona) (speed, multi-OS), and [Modal](/tools/modal) (sandboxes inside a GPU platform) each pull differently: [Sandboxing AI-Generated Code](/guides/advanced/sandboxing-ai-generated-code) maps the choice.

---

_Source: https://agentscamp.com/tools/vercel-sandbox — Tool on AgentsCamp._