Packages

This page lists the packages in the artifact set, with their versions and a line on what works. The limitations page has what does not.

Every package here is pinned to one version. The set carries one version of each package, built from its released source, not a choice of versions. Compiled code is linked into python.wasm as builtin modules: modules compiled into the interpreter itself, so nothing is loaded from a shared library at run time. The Python half of each package is in the stdlib zip.

The interpreter

CPython 3.15.0, built for wasm32-wasip1. python.wasm -V prints the version, then a line of the platform's facts:

(wasm32-wasip1: numpy's default integer is 32-bit; no threads; no ctypes, mmap, sockets, ssl, zstd or scipy.odr; compute about 1.3 to 2.3 times slower than native)

zlib, bz2, lzma and sqlite3 are built in, over zlib, bzip2 1.0.8, liblzma from xz 5.8.4 and SQLite 3.54.1. bz2, lzma and tarfile's bz2 and xz modes write the same bytes as native Python. sqlite3 has native's feature set: FTS3/4/5, R-tree, Geopoly, math functions, dbstat and percentile.

Compiled packages

package	version	what works
numpy	2.5.2	Built in. Linear algebra, dot and matmul run on the set's OpenBLAS 0.3.34, as native numpy's do. The default integer is 32-bit.
pandas	3.0.6	Built in, including read_html, read_xml and to_xml through lxml, .xlsx through openpyxl and xlsxwriter, and the Styler.
scipy	1.18.1	Everything but scipy.odr: linalg, sparse, optimize (HiGHS's linprog and milp), integrate, interpolate, signal, fft, special, stats, spatial, ndimage, cluster, io and the rest. It runs on one thread.
scikit-learn	1.9.1	Every public submodule, and its six bundled datasets. It runs serially, with the same answers.
statsmodels	0.15.0	Every public submodule, statsmodels.api and statsmodels.formula.api included, with formulaic 1.2.2 (with interface-meta 2.0.1 and wrapt 2.5.1) as a second formula engine. Sixteen public-domain datasets ship.
matplotlib	3.11.2	The Agg backend. PNG, PDF, PS, EPS and SVG save. Contours through contourpy 1.4.0, and Delaunay triangulations through qhull 8.0.2. Text through FreeType 2.14.3, HarfBuzz 14.1.0, SheenBidi 3.0.0 and libraqm 0.10.5. See matplotlib.
Pillow	13.0.0	Its core, with zlib only: PNG works; JPEG and the other codecs do not.
kiwisolver	1.5.1	Built in, as one of matplotlib's dependencies.
contourpy	1.4.0	Contours for matplotlib. Its threaded algorithm runs on one thread.
lxml	6.1.3	lxml.etree and lxml.objectify over libxml2 2.15.4 and libxslt 1.1.45 with libexslt: XPath, XSLT, XInclude, C14N, DTD, XML Schema, RELAX NG and Schematron validation, the HTML parser and iterparse.
regex	2026.9.29	Fuzzy matching, full Unicode properties, graphemes, variable-length lookbehind, captures(), with native regex's answers.

A golden job is a fixed program whose output the project compares with native Python's at the same pins. numpy, pandas, scipy, scikit-learn, statsmodels, matplotlib, lxml and regex each have golden jobs, and each gives native's answers.

Pure-Python packages

matplotlib brings fontTools 4.67.0, cycler, pyparsing and packaging. The zip also carries python-dateutil, tzdata and six. Besides these, the stdlib zip carries thirty pure-Python packages, plus seaborn and reportlab:

family	distribution and version	what it is for
Excel	openpyxl 3.1.5, et-xmlfile 2.0.0	read and write .xlsx; pandas' .xlsx reader
	xlsxwriter 3.2.9	write .xlsx (pandas' default writer), in memory too
	xlrd 2.0.2	read old .xls files
plotting	seaborn 0.13.2 (beside matplotlib)	statistical plots over pandas frames
PDF	pypdf 6.21.1	read, extract text, merge, split, encrypt (RC4)
	reportlab 5.0.1	write PDFs: text, tables, charts, Unicode text, images
HTML and XML	beautifulsoup4 4.15.0, soupsieve 3.0.3	parse messy HTML, CSS selectors
	html5lib 1.1, webencodings 0.6.1	a browser's HTML parsing; bs4's html5lib parser
	defusedxml 0.7.1	parse untrusted XML, refusing entity tricks
	xmltodict 1.0.4	XML to dicts and back
text formats	pyyaml 6.0.3	YAML
	tomli-w 1.2.0	write TOML (the stdlib's tomllib reads it)
	jinja2 3.1.6, markupsafe 3.0.4	templates, HTML escaping
	tabulate 0.10.0	text tables; df.to_markdown()
	charset-normalizer 3.5.2	guess a file's encoding
modelling	networkx 3.7	graphs and their algorithms
	patsy 1.0.3	model formulas (y ~ x + C(g)) to design matrices
symbolic	sympy 1.14.0, mpmath 1.3.0	symbolic maths; arbitrary precision
utilities	attrs 26.1.0, typing-extensions 4.16.0	classes without boilerplate; typing backports
	more-itertools 11.2.1, toolz 1.2.0	iteration and functional helpers
	tqdm 4.70.1	progress bars
	joblib 1.6.0, cloudpickle 3.1.2	caching, Parallel, pickling closures
	threadpoolctl 3.7.0	thread-pool limits (there are none to limit here)
	narwhals 2.27.1	dataframe-agnostic code over pandas

pyyaml and markupsafe come from their source releases and run as pure Python: pyyaml has no libyaml, and yaml.CSafeLoader is absent. Each package's importlib.metadata answers work, so df.to_markdown() finds tabulate.

Each family has a golden job whose output equals native Python's at the same pins.

What is not carried

These are not in the set, so each raises its ImportError, as it does natively without it:

• pytz: use zoneinfo, which reads the stdlib zip's tzdata;
• cryptography, which pypdf needs for AES;
• antlr4 and lark, for sympy's parse_latex;
• odfpy, for pandas' engine="odf", and pyxlsb, for .xlsb files;
• cssselect and lxml_html_clean, for lxml.cssselect and lxml.html.clean.

reportlab's bitmap renderer needs rlPyCairo, which is not carried either; it raises RenderPMError. Draw reportlab charts to PDF or SVG instead.

Adding a package of your own

You cannot add a compiled package: the guest has no dlopen and loads no shared library. A pure-Python package can come from a read-only grant named in import_path=, as on the guest filesystem page. It comes after the runtime on the import path, so a module the set already has is always the set's.

There is no pip inside the guest.

To build a package into the artifact set itself, see Adding a package.

In one line

CPython 3.15.0 with numpy, pandas, scipy, scikit-learn, statsmodels, matplotlib, lxml and regex compiled in, and thirty pure-Python packages plus seaborn and reportlab in the zip, each at one pinned version.