ai-sdk-tool-code-execution weekly downloads statistics

AI SDK Code Execution Tool

A TypeScript package that provides a code execution tool for the AI SDK. Execute Python code in a sandboxed environment using Vercel Sandbox.

Installation

Prerequisites

• A Vercel account with access to Vercel Sandbox (available in Beta on all plans) • Vercel CLI installed • A Vercel OIDC token for authentication

Setup

1. Link your Vercel project

From your project directory, link to a new or existing Vercel project:

2. Pull environment variables

Download your Vercel OIDC token:

This creates a .env.local file with your VERCEL_OIDC_TOKEN that the SDK uses to authenticate with Vercel Sandbox.

Note: Development tokens expire after 12 hours. Run vercel env pull again when your token expires.

3. Add your AI provider credentials

If using Vercel AI Gateway, add your API key to .env.local:

Usage

The executeCode tool allows your AI agent to run Python 3.13 code in a Vercel Sandbox environment. The agent can perform calculations, data processing, and other computational tasks safely in an isolated environment.

Options

Configure the tool with optional parameters:

Example with debug enabled:

When debug is enabled, you'll see detailed logs of code execution in your terminal.

How it works

This package uses Vercel Sandbox to execute Python code in an ephemeral, isolated environment. Each code execution:

1. Creates a new sandbox with Python 3.13 runtime 2. Executes your code using python3 -c 3. Captures stdout, stderr, and exit codes 4. Automatically stops the sandbox after execution

Important notes

• Python 3.13 runtime: Code runs in Vercel Sandbox's python3.13 image • Not a REPL: You must use print() to see output. Bare expressions produce no output • Isolated execution: Each sandbox runs in a secure, ephemeral environment on Amazon Linux 2023 • Authentication required: Requires a valid Vercel OIDC token • Resource limits: See Vercel Sandbox pricing and limits

Alternative authentication

If you cannot use VERCEL_OIDC_TOKEN, you can authenticate with access tokens. Set these environment variables:

Find your team ID, project ID, and create an access token in your Vercel dashboard.

Development

Testing locally

Test the tool with the included test script:

Building

Build the package:

Publishing

Update the version in package.json, then publish:

The package automatically builds before publishing.

Project structure

Monitoring usage

Track your sandbox usage in the Vercel dashboard:

1. Go to your project 2. Click the AI tab 3. Click Sandboxes to view execution history and URLs

View compute usage across all projects in the Usage tab of your dashboard.

Learn more

• Vercel Sandbox documentation • Vercel Sandbox examples • Vercel Sandbox SDK reference

License

ISC

Description from the npm listing.