Wasmer SDK for Python

Run Wasmer packages in lightweight, composable sandboxes from Python. The public API is handwritten and typed; a Python-independent UniFFI library calls the Rust wasmer-sdk underneath.

Install

pip install wasmer-sdk

Published wheels support macOS and Linux on arm64 and x86_64. One wheel works across supported Python 3 versions on the same platform because the native boundary does not use the CPython ABI.

Load or create a package from Wasm

Load a single raw WASI/WASIX module with pkg = await wasmer.packages.load(Path("hello.wasm").read_bytes()). Byte sources detect Wasm or WEBC by their contents. Raw modules must export _start; their single command and entrypoint are named main automatically. Run with sandbox.command(pkg) or sandbox.command("main"). WEBC packages retain their declared commands and entrypoint.

Create a reusable package from module bytes and optional bundled files:

from pathlib import Path
from wasmer_sdk import PackageCommandDefinition, PackageDefinition, Wasmer

async def run_module():
    async with Wasmer() as wasmer:
        pkg = await wasmer.packages.create(PackageDefinition(
            modules={"app": Path("hello.wasm").read_bytes()},
            commands={"hello": PackageCommandDefinition(module="app")},
            files={"/data/config.json": '{"debug":true}'},
        ))
        async with await wasmer.sandboxes.create(packages=[pkg]) as sandbox:
            print((await sandbox.command(pkg, ["--help"]).run()).text())

Creation copies module and file data into an in-memory package without a WEBC archive. Command modules must export _start and run with the existing WASI/WASIX runner. One command becomes the entrypoint automatically; with multiple commands, set entrypoint="hello" or select pkg.command("hello"). File keys are absolute guest paths without . or .. segments. Package file writes use private execution overlays; use the sandbox workspace for persistent data. The result also works with sandbox.install_package(pkg).

Run Python inside Wasmer

import asyncio

from wasmer_sdk import Wasmer


async def main() -> None:
    wasmer = Wasmer()
    sandbox = await wasmer.sandboxes.create(
        packages=["python/python@=3.13.20"],
        files={
            "main.py": "print(sum(n * n for n in range(10)))",
        },
    )
    output = await sandbox.command(
        "python", ["/workspace/main.py"]
    ).run()
    print(output.text())


asyncio.run(main())

The @= package-version syntax is an exact pin. Use it when a sandbox must resolve the same Wasmer package version on every run.

Constructing Wasmer() is synchronous. Operations that perform I/O are awaited. A wasmer instance can be reused across multiple sandboxes. For long-lived applications and tests, use async with for deterministic cleanup:

async def main() -> None:
    async with Wasmer() as wasmer:
        sandbox = await wasmer.sandboxes.create(
            packages=["python/python@=3.13.20"],
        )
        async with sandbox:
            output = await sandbox.command(
                "python", ["--version"]
            ).run()
            print(output.text())


asyncio.run(main())

run() raises ProcessExitError for a non-zero exit, termination, or timeout by default. Pass check=False when the outcome is expected and should be inspected as an Output. Spawned-process wait() remains unchecked by default.

Live processes expose asynchronous stdin, stdout, and stderr streams. process.stdout.lines() is an async iterator suitable for servers and agent loops.

Networking and caching

Network access is an explicit sandbox capability:

sandbox = await wasmer.sandboxes.create(
    packages=["wasmer/edgejs@0.2.0"],
    network="host",
)

Packages and native compiled artifacts are cached in .wasmer by default:

wasmer = Wasmer(cache_root=".cache/wasmer")

The registry and downloaded package layout is shared with Node.js and Rust. Compiled artifacts remain target-specific.

Examples

Run the same guest programs used by the JavaScript and Rust SDK examples:

python3 python/examples/python.py
python3 python/examples/multiple_runtimes.py
python3 python/examples/edgejs_http.py
python3 python/examples/postgres_psql.py

The PostgreSQL example requires psql on PATH; pass --psql /path/to/psql otherwise. The common guest inputs live in ../fixtures/.

Build and test locally

From the repository root, install Rust 1.95 and build the UniFFI module:

rustup toolchain install 1.95.0 --profile minimal
python3 python/scripts/build.py --release

Binary releases contain Python 3 wheels for Linux x86_64/ARM64 and macOS Intel/Apple Silicon. Each wheel has a py3-none ABI tag; a separate wheel per Python minor version is unnecessary. The exact wheels published to PyPI are also attached to wasmer-sdk-python-v<version> GitHub releases with checksums.

The builder enables native V8/Node-API integration on Linux x86_64 and macOS Apple Silicon. The pinned V8 build does not support Linux ARM64 or macOS Intel; those wheels use the sys backend for Wasm/WASI/WASIX and do not support native N-API guests such as Edge.js. Use --backend sys or --backend napi-v8 to select the backend explicitly when building locally.

Run the Python suite against that local module:

PYTHONPATH=python/src \
  python3 -m unittest discover -s python/tests -v

Run an individual example against the local build:

PYTHONPATH=python/src python3 python/examples/python.py

The PostgreSQL tests require native psql. Set PSQL=/path/to/psql when it is not on PATH.

The generated _native.py module and platform library are private implementation details. Applications should import only from wasmer_sdk.