# Vercel Sandbox

> Vercel Sandbox provides isolated, sandboxed virtual machine environments for safely executing untrusted code as part of AI agent workflows and application backends.

Vercel Sandbox is a product within Vercel's Agent Stack that provides isolated virtual machine environments for running code you didn't write — safely and at scale. It is designed for AI agent workflows, agentic applications, and any scenario where untrusted or dynamically generated code needs to execute without risk to the host environment. Sandbox is part of Vercel's broader platform alongside Workflows, AI Gateway, and the AI SDK.

## What It Is

Vercel Sandbox gives developers and AI agents a way to spin up short-lived, isolated compute environments on demand. Each sandbox runs in its own VM, keeping execution separate from the host application and from other sandboxes. The product is positioned as "the safest way to run code you didn't write," targeting use cases where AI-generated code, user-submitted scripts, or agent-driven execution needs a secure boundary. It integrates natively with the rest of Vercel's infrastructure, including Fluid Compute and Vercel Workflows.

## How It Fits in the Agent Stack

Vercel Sandbox is one of five components in Vercel's Agent Stack, alongside the AI SDK, AI Gateway, Workflows, and Eve. The stack is designed so that agents can reason, execute code in isolation, run for extended periods, and recover from failure. Sandbox specifically handles the "execute code in isolation" part of that loop, while Workflows handles durable orchestration and AI Gateway handles model routing. This modular design lets teams compose agentic systems from purpose-built primitives rather than building isolation infrastructure themselves.

## Execution Model and Limits

Sandboxes are billed based on active CPU time, provisioned memory, number of creations, network transfer, concurrent sandbox count, and sandbox storage. The Vercel pricing page lists specific included amounts per plan tier and per-unit overage rates. Concurrent sandbox limits differ significantly between plan tiers — the Hobby plan supports up to 10 concurrent sandboxes, while the Pro plan supports up to 2,000. Sandbox storage is also plan-dependent, with Hobby receiving a fixed allocation and Pro plans able to purchase additional GB-month storage.

## Deployment Model

Vercel Sandbox runs entirely on Vercel's managed infrastructure — there is no self-hosted option. Developers access it through Vercel's platform, and it is available to Hobby, Pro, and Enterprise plan users, with usage metered beyond included allocations. Because it is part of Vercel's cloud platform, it inherits Vercel's global network, security posture (SOC 2 Type 2, ISO 27001, PCI DSS), and observability tooling.

## Target Audience

The primary audience is developers building AI-powered applications and agentic systems that need to execute code dynamically — including coding agents, LLM-driven automation, and multi-step agent pipelines. It is also relevant to platform engineers who need to offer code execution as a feature within their own products without managing VM infrastructure directly. Vercel's homepage notes that the platform powers agent use cases at scale, citing Notion as running millions of agent conversations daily on Vercel (per Vercel's own marketing claims).

## Features
- Isolated sandboxed VM environments
- On-demand sandbox creation
- Safe execution of untrusted or AI-generated code
- Concurrent sandbox support (up to 2,000 on Pro)
- Active CPU-based billing
- Provisioned memory allocation
- Sandbox storage
- Network transfer included
- Integration with Vercel Workflows and AI Gateway
- Part of Vercel Agent Stack
- Global managed infrastructure
- SOC 2 Type 2 and ISO 27001 compliance

## Integrations
Vercel Workflows, Vercel AI Gateway, Vercel AI SDK, Vercel Fluid Compute, Vercel Functions, Next.js, Vercel Observability

## Platforms
WINDOWS, WEB, API, CLI

## Pricing
Freemium — Free tier available with paid upgrades

## Links
- Website: https://vercel.com/sandbox
- Documentation: https://vercel.com/docs
- Repository: https://github.com/vercel
- EveryDev.ai: https://www.everydev.ai/tools/vercel-sandbox