vercel-sandbox

You can safely run untrusted code by deploying it inside ephemeral Firecracker microVMs that offer millisecond startup and network isolation. This sandbox approach prevents harmful operations from affecting your host system during execution.

Isolating AI agent code execution is best achieved using ephemeral Firecracker microVMs with strict network and I/O policy controls. This ensures deterministic, secure execution by preventing unauthorized access while running generated scripts.

Supported runtimes in a Firecracker sandbox include node24, node22, and python3.13. These environments enable safe experimentation for code generation tools and developer playgrounds within an isolated working directory.

You need Firecracker microVMs for a developer playground if you require strict isolation of untrusted code and precise network policy controls. They provide ephemeral, secure environments suitable for safe experimentation and tutorials.

Yes, you can control network and file I/O access for untrusted code by applying policy controls within the microVM API surface. This ensures safe execution and deterministic isolation for your AI agent workloads.

Ephemeral microVMs for code execution are limited by their temporary nature, meaning state is lost after execution ends. They are designed for safe experiments and tutorials rather than long-running persistent processes.