Vercel Sandbox Python SDK 1.0 reaches TypeScript parity
Vercel's Python package for running code in isolated sandboxes hits a stable 1.0, with a breaking bytes change for anyone on 0.x.
Vercel Sandbox Python SDK 1.0 arrives
On 16 November 2026 Vercel published a changelog post announcing that the Vercel Sandbox Python SDK has reached 1.0. The entry is credited to Scott Trinh, a software engineer at the company, and runs to about a minute of reading. Its two headline claims: the public API is now stable, and the Python package has feature parity with the TypeScript SDK. Vercel frames the SDK's purpose as running AI-generated code, testing changes and analysing datasets in isolated sandboxes, all driven from a Python application. The full changelog entry carries the sample code and links to the docs.
How the SDK is used
Most of the post covers two mechanics.
Sandboxes are opened with a context manager. The example creates one from a Git source, passing a repository URL and a revision, which in the sample is a pull request branch. Inside the block, the script runs a command with arguments, using pytest in quiet mode and setting check to true so a failure raises. When the block ends, the sandbox is cleaned up automatically. For a job that clones a branch, runs a test suite and then disappears, that is the whole lifecycle in a few lines.
The second mechanic is sync and async. Async support is built on AnyIO, which the post says covers asyncio and Trio. Synchronous code uses vercel.sandbox.sync instead. Installation goes through uv with the requirement vercel-sandbox>=1.0, and authentication has to be configured before any of this works.
What changes for Python teams
The post says 1.0 brings feature parity with the TypeScript SDK, which implies earlier Python versions lagged it. The practical effect is that a Python backend can reach for the same primitive the TypeScript package exposes: an isolated sandbox, a command, and a result.
The stability promise matters for anything long-lived. A 1.0 label generally means the maintainers intend to stop moving the API underneath you. Vercel does not say how long it will support 1.0, or what happens to 0.x releases from here.
Limits and open questions
This is a changelog, not a reference page. It does not mention pricing, how long a sandbox can live, concurrency or rate limits, which runtimes are available beyond the uv and pytest shown in the example, or which regions sandboxes run in. Builders who need those numbers will have to read the documentation.
Upgrading has one named hazard. The post tells 0.x users to review the release notes for breaking changes and calls out a switch to bytes for command input and output. That suggests existing code which passes strings into commands, or reads command output as text, will need adjusting. The post does not say how many other breaks exist.
What this likely means
Everything in this section is inference, not something the post states. Feature parity with the TypeScript SDK likely lowers the cost of adopting Vercel Sandbox for Python-heavy teams, since the Python package can now be the entry point rather than something secondary. A 1.0 release also suggests Vercel expects the API to be depended on rather than trialled. The changelog says nothing about adoption, performance or reliability, so none of that is confirmed.
For people building apps with AI
Executing model-generated code safely has become an ordinary part of shipping an AI feature, and a sandbox is one answer to it. Draftship, the AI app builder, is not affiliated with Vercel, and nothing in the announcement says Draftship supports the Sandbox Python SDK. If you are working on the app side of an AI feature, Draftship's guide to adding AI to your app covers the layer that sits above code execution.
Primary sources
- Vercel Sandbox Python SDK 1.0 is now available · Vercel, Nov 16, 2026
This article was drafted with AI assistance from the primary sources listed below and checked automatically against them. An editor hasn't reviewed it yet.