Pyodide vs a server sandbox: can Pyodide run pip packages?

Pyodide installs pure-Python wheels and packages built for it, without threads or subprocesses; a server sandbox runs any pip package.

A Runtime sandbox runs ordinary CPython on a real Linux machine, so pip install works for any wheel on PyPI, native extensions included. A new sandbox ran its first Python command 351 ms after the create request at the median, over the public internet, on 24 September 2026 (speed).

What Pyodide is

Pyodide is "a Python distribution for the browser and Node.js based on WebAssembly": CPython compiled to WebAssembly with Emscripten. Version 314.0.0, released 9 June 2026, moved it to Python 3.14.2; the current release on 25 September 2026 is 314.0.7, from 14 September 2026. When it runs in a browser, the project says, "Python has full access to the Web APIs."

It is a good fit for Python that should run on the user's own device: notebooks in the browser, teaching tools, and data work where the data never leaves the page.

Can Pyodide run pip packages?

Some of them. Pyodide installs packages with micropip, and after Pyodide loads "only the Python standard library is available." micropip installs:

• Pure-Python wheels from PyPI. The home page says "any pure Python package with a wheel available on PyPi is supported." A wheel ending in py3-none-any.whl is the sign.
• Packages built for Pyodide, the wasm32 Emscripten wheels. The project has ported many packages with C, C++ and Rust extensions. Its list for the current release includes NumPy 2.4.6, pandas 3.0.2, SciPy 1.18.0, Matplotlib 3.10.8, scikit-learn 1.8.0, DuckDB 1.5.1 and lxml 6.1.3.

Anything else fails. The FAQ says a package with "binary extensions (e.g. C, Fortran or Rust)" that nobody has built for Pyodide "needs to be cross-compiled for Pyodide." A wheel fetched from your own server needs CORS headers, or the browser blocks it.

What else does not work in Pyodide?

From Pyodide's own pages on WebAssembly constraints and its FAQ:

• No threads, processes or subprocesses. "Fork and pthreads do not work in Pyodide." Starting a thread, a multiprocessing worker or a subprocess raises RuntimeError. Code that calls git, ffmpeg or a compiler cannot run.
• Sockets, only experimentally, and only in Node.js. "By default, Pyodide does not support sockets in the browser." Node.js can enable an experimental socket API with useNodeSockFS().
• HTTP through the browser. Requests are "subject to the same limitations as any JavaScript network call": "very little or no control over certificates, timeouts, proxies", and CORS can block them.
• Missing standard modules. venv, ensurepip, termios, tkinter, curses, fcntl, resource and others are removed; ssl is "a stub implementation that does not use OpenSSL."

Side by side

Is Pyodide safe for untrusted code?

It depends where it runs. In a browser, Pyodide code runs in the page with "full access to the Web APIs", so model-written code shares the page's origin and whatever the page can reach. In Node.js, it runs inside your server's process. A server sandbox keeps the code off your machines and your users' pages entirely.

When each fits

• Pick Pyodide when the code should run on the user's device: a browser notebook, a teaching exercise, a chart from data the user dropped in, with pure-Python or already-ported packages.
• Pick a server sandbox when the code comes from a model or an agent and may need any package, a subprocess, a compiler, a database driver, the network or more memory than a browser tab, or must not run where your users' sessions live.

Run the same Python in a Runtime sandbox

Allow the package index, install what the code needs, then switch the network off before the model's code runs:

TypeScriptimport { Sandbox } from "withruntime";await using sbx = await Sandbox.create({  network: { internet: true, allow: ["pypi.org", "*.pythonhosted.org"] },  timeoutSeconds: 600,  onLeaseEnd: "stop",});await sbx.exec("pip install polars", { check: true, timeoutMs: 180_000 });await sbx.network.set({ internet: false });const run = await sbx.exec([  "python3",  "-c",  "import polars, subprocess; print(polars.__version__, subprocess.run(['nproc'], capture_output=True, text=True).stdout)",]);console.log(run.stdout);

Pythonfrom withruntime import Sandboxwith Sandbox.create(    network={"internet": True, "allow": ["pypi.org", "*.pythonhosted.org"]},    timeout_seconds=600,    on_lease_end="stop",) as sbx:    sbx.exec("pip install polars", check=True, timeout_ms=180_000)    sbx.network.set(internet=False)    run = sbx.exec([        "python3",        "-c",        "import polars, subprocess; print(polars.__version__, subprocess.run(['nproc'], capture_output=True, text=True).stdout)",    ])    print(run.stdout)

For a chat that runs Python turn after turn, the code interpreter keeps variables between cells and returns charts as PNG and data frames as tables (code interpreter for chatbots). NumPy, pandas and matplotlib are in the default image.

New accounts get 50 free sandbox hours, no card:

Terminalnpx withruntime sandbox run --trial -- python3 -c 'import pandas; print(pandas.__version__)'

More: WebAssembly vs microVM, what is a code interpreter?, data analysis agent, Python SDK.

Sources

Checked 25 September 2026.

• Pyodide, changelog, loading packages, packages built in Pyodide
• WebAssembly constraints, FAQ, using sockets

Facts on this page were checked on 25 September 2026.