Vercel Sandbox Python SDK 1.0 is now available
Vercel Sandbox Python SDK 1.0 is out, so the public API is now stable and matches the TypeScript SDK's features. You can use it from your Python app to run AI-generated code, test changes, or analyze datasets in isolated Vercel sandboxes.
Async code works on asyncio and Trio through AnyIO, and context managers clean up the sandbox for you. If your code is synchronous, use vercel.sandbox.sync. Vercel's example creates a sandbox from a Git branch, runs uv run pytest -q inside it, then tears it down when the block ends. Install it with uv add 'vercel-sandbox>=1.0'.
The 1.0.0 release notes also add SandboxManagedCredentials(npm=True), which lets a sandbox install private npm packages with your team's Vercel-managed credentials without exposing tokens inside it. Retries and timeouts now follow the TypeScript SDK, and you can set an explicit request_timeout.
Coming from 0.x, there are two breaking changes. Command input and output now use bytes, so you encode and decode text yourself, and TextReader is replaced by ByteReader. Python 3.11 or newer is also required, and 3.15 is supported.