Vercel Sandbox
Code execution
Vercel Sandbox runs untrusted or agent-generated code in isolated Linux microVMs. Each sandbox is a Firecracker microVM with its own filesystem and network. Developers create and manage sandboxes through a JavaScript and TypeScript SDK, a Python SDK or a sandbox command line tool, authenticating with Vercel OIDC tokens or access tokens. The default image includes the current Node.js LTS and Python, and persistent sandboxes save state when stopped. Usage is metered, with a free Hobby allotment.
Billing uses Active CPU time, so time spent waiting on model calls or network responses is not charged as CPU.

Pinout
- MCP server: not checked yet
- HTTP API: not checked yet
- command line: supported, verified
- Python SDK: supported, verified
- TypeScript SDK: supported, verified
- OpenAPI description: not checked yet
| Interface | Status | Source we read | Read on |
|---|---|---|---|
| MCP server | Not checked yet | — | — |
| HTTP API | Not checked yet | — | — |
| Command line | Supported, verified | Vercel docs, Sandbox, Using Vercel Sandbox | |
| Python SDK | Supported, verified | Vercel docs, Sandbox, Using Vercel Sandbox | |
| TypeScript SDK | Supported, verified | Vercel docs, Sandbox, Using Vercel Sandbox | |
| OpenAPI description | Not checked yet | — | — |
Facts
| Purpose | Run untrusted or agent-generated code in isolated Linux microVMs. Vercel docs, Sandbox, read . |
|---|---|
| Isolation | Each sandbox runs in a Firecracker microVM with its own filesystem and network. Vercel docs, Sandbox, Features, read . |
| Interfaces | @vercel/sandbox for JavaScript and TypeScript, the vercel-sandbox Python package and the sandbox CLI. Vercel docs, Sandbox, Using Vercel Sandbox, read . |
| Authentication | Vercel OIDC tokens (recommended) or access tokens. Vercel docs, Sandbox, Authentication, read . |
| Default image | vercel/sandbox/universal, including the current Node.js LTS, Python 3.14 and coding agents. Vercel docs, Sandbox, System specifications, read . |
| Persistence | Persistence is the default: sandboxes auto-save state on stop and resume. Vercel docs, Sandbox, Features, read . |
| Pricing, hobby allotment | Hobby includes 5 hours per month of Active CPU, 420 GB-hours per month of provisioned memory and 5,000 creations per month. Vercel docs, Sandbox pricing and quotas, read . |
| Pro rate | Active CPU is $0.128 per hour on Pro in the default iad1 region; rates vary by region. Vercel docs, Sandbox pricing and quotas, read . |
| Session limits | Maximum session duration is 45 minutes on Hobby and 24 hours on Pro and Enterprise. Vercel docs, Sandbox pricing and quotas, read . |
| Regions | Available in 19 regions; the default is iad1. Vercel docs, Sandbox pricing and quotas, read . |
| SDK and CLI repository license | Apache-2.0 (github.com/vercel/sandbox, which holds the SDK and CLI codebase; the hosted service is not covered). Repository LICENSE file, read . |
| Last commit | . GitHub API, latest commit on the default branch of vercel/sandbox, read . |
Call example
sandbox run echo "Hello Sandbox!"Limits
- Hobby sandbox creation pauses once the included quota is used.
- Rates for several meters differ by region.
- The repository license covers the SDK and CLI, not the hosted service.
Strengths
- JavaScript, Python and command line interfaces are documented.
- Active CPU billing excludes time spent waiting on I/O.
- Sandboxes can be persistent and snapshotted.
Pricing
As written on Vercel Sandbox's own page on the read date. Prices change: check there before you pay.
Alternatives to Vercel Sandbox
All code execution tools for agents
Run untrusted or agent-written code in isolated Linux sandboxes on Workers.

Isolated Linux sandboxes for running agent code through Python or JavaScript SDKs.

Hardware-isolated microVM sandboxes for agent code, local-first.

Sandboxes for agent code, reachable by API, CLI, MCP and SDKs.
Last check: . Method: public repository and docs read automatically, page and day written down. Next check due within 90 days. How we check
Is this your tool? Write to hello@agent-tools.directory from an address on its domain to correct a pin or a fact.