Limitations

PyMode runs upstream CPython 3.13 compiled to wasm32-wasi inside Cloudflare Workers Durable Objects. This works well for most Python code, but the WASM/WASI/DO environment has inherent constraints.

Platform Constraints

Constraint	Details
No filesystem persistence	The WASM linear memory is reset per request. Use KV/R2/D1 for storage.
No network sockets	Standard socket module is unavailable. Use pymode.tcp and pymode.http instead.
No threads	threading is disabled (--disable-threads). Use pymode.parallel for real parallelism via child DOs.
No subprocess	os.system, subprocess, os.fork are not available in WASI.
No signals	Signal handling is emulated (-lwasi-emulated-signal), limited functionality.
30s CPU limit	Per-request CPU time is capped at 30 seconds (CF Workers limit).
128MB memory	Per-DO memory limit. Large data processing should use pymode.parallel to distribute across DOs.
10MB bundle size	CF Workers’ paid-tier limit on compressed deploy. A typical pymode deploy produces a 6-8 MB gzipped wasm depending on which packages are AOT-preimported; the remainder is your project sources + site-packages. Heavy native deps (numpy/pandas) consume more of the budget.

Bundled Stdlib Modules

PyMode bundles a subset of CPython’s stdlib as string constants in the worker. These are the modules available in the WASM runtime:

Always Available (bundled in stdlib-fs)

Category	Modules
Encodings	encodings, encodings.aliases, encodings.utf_8, encodings.ascii, encodings.latin_1
JSON	json
Regex	re
Collections	collections, collections.abc, functools, operator
Types	enum, types, typing, abc
Core	warnings, contextlib, dataclasses, copy, copyreg, weakref, _weakrefset
Text	string, textwrap
Crypto	base64, hashlib, hmac, secrets
Numbers	random, bisect, heapq, numbers, fractions, decimal
Date/Time	datetime, calendar
Path	fnmatch, glob, pathlib
URL	urllib, urllib.parse, ipaddress
Import	importlib, importlib.abc, importlib.machinery, importlib.util
Serialization	pickle, _compat_pickle, struct
I/O	io, _pyio
Misc	keyword, reprlib, traceback, linecache, tokenize, token, csv, pprint

Built-in C Modules (compiled into python.wasm)

These are C extension modules linked directly into the WASM binary:

array, _asyncio, _bisect, _contextvars, _csv, _heapq, _json, _pickle, _random, _struct, _zoneinfo, math, cmath, _statistics, _datetime, _decimal, _md5, _sha1, _sha2, _sha3, _blake2, _codecs, _collections, errno, _io, itertools, _sre, _thread, time, _typing, _weakref, _abc, _functools, _locale, _operator, _stat, _symtable, unicodedata, pyexpat, _elementtree

Polyfilled Modules

Module	Implementation	Notes
binascii	Pure Python (lib/polyfills/binascii.py)	Replaces the C extension. Covers hex, base64, crc32, uu encoding.
socket	Pure Python (lib/polyfills/socket.py)	Routes through pymode.tcp host imports. Provides constants, getaddrinfo, create_connection.
threading	Pure Python (lib/polyfills/threading.py)	No-op locks within a DO. Thread.start() spawns child DOs via pymode.parallel for real parallelism.
logging	Pure Python (lib/polyfills/logging/)	Threading-free logging that writes to stderr. Full API: getLogger, basicConfig, handlers.
multiprocessing	Pure Python (lib/polyfills/multiprocessing/)	cpu_count() returns 1, exception classes. Use pymode.parallel for real parallelism.
ssl	Pure Python (lib/polyfills/ssl.py)	TLS termination handled by CF’s edge — Python code doesn’t need raw TLS.
zlib	Pure Python (lib/polyfills/zlib.py)	Minimal compression polyfill for packages that import zlib at module level.

Not Available

These stdlib modules require OS features that WASI doesn’t provide:

Module	Reason
socket	Polyfilled — routes through pymode.tcp host imports.
ssl	Polyfilled — TLS termination handled by CF’s edge.
subprocess	No process spawning in WASI.
multiprocessing	Polyfilled — cpu_count() returns 1. Use pymode.parallel for real parallelism via child DOs.
threading	Polyfilled — no-op locks within a DO. Thread.start() spawns child DOs via pymode.parallel.
sqlite3	C extension not compiled for WASM. Use D1 binding instead.
ctypes	No dynamic linking for arbitrary shared libraries.
tkinter	No GUI in Workers.
curses	No terminal in Workers.
readline	No interactive terminal.
mmap	Not available in WASI.
fcntl, termios, grp, pwd	Unix-specific, not in WASI.
select, selectors	Polyfilled — selectors bundled in stdlib.
signal	Emulated only, limited functionality.
resource	No resource limits API in WASI.

Now Bundled (previously missing)

These modules have been added to the stdlib bundle since initial release:

Module	Added for
html, html.parser, html.entities	beautifulsoup4, web scraping
email (full package)	requests, urllib3, http.client
http, http.client, http.cookiejar	requests, httpx, urllib3
xml, xml.sax, xml.etree, xml.parsers	langchain, defusedxml
asyncio (full package)	pydantic, typing_extensions
argparse	numpy, click
locale	httpx, distro
tempfile, shutil	many packages
importlib.metadata, importlib.resources	package metadata discovery
concurrent.futures	tenacity, langsmith
zoneinfo	pydantic datetime
tomllib	pydantic, project configs
configparser	various packages
sysconfig	pydantic, setuptools
unittest	pyparsing

Requesting a module: If you need a stdlib module that’s not bundled, open an issue or add it to the bundle list in scripts/generate-stdlib-fs.ts.

Python Language Support

Full CPython 3.13 language support including:

• All syntax features (match/case, walrus operator, f-strings, etc.)
• Generators, async generators, coroutines
• Decorators, metaclasses, descriptors
• Exception groups, ExceptionGroup
• Type hints, typing module

Not supported:

• async/await at the handler level (use Asyncify for I/O instead)
• asyncio event loop (no I/O multiplexing in WASI)

Package Support

C-extension packages: current state

pymode deploy auto-detects when your project depends on a C-extension package and, if a variant wasm exists, uses that variant directly (fast deploy — no AOT rebuild). The numpy variant is wired and verified end-to-end as of this writing.

Package	How it’s served	Verified deployable today
numpy	python-numpy.wasm variant	✅ (linalg stubbed)
Pillow	python-pillow.wasm variant	✅
pydantic_core	python-pydantic-core.wasm variant	✅
pyyaml	pure-Python fallback (uv pip)	✅
multidict / yarl / frozenlist / propcache	pure-Python fallback (uv pip)	likely (same pattern as pyyaml)
msgpack	pure-Python fallback (uv pip)	likely
markupsafe	statically linked	yes
regex / xxhash	statically linked	yes
pandas	needs numpy + Cython compile pass	no
scipy	needs Fortran cross-compile	no
scikit-learn	depends on scipy	no
lxml	needs libxml2 + libxslt cross-compile	no
cryptography	needs OpenSSL cross-compile	no
orjson / ujson	Rust/C-only (no pure-Python fallback)	no
cffi	runtime C-compile model incompatible with WASM	no

numpy caveats: the variant ships _multiarray_umath (array core + random + FFT) but does not ship _umath_linalg. Linear algebra functions in numpy.linalg are stubbed and raise ImportError if called. Everything else (arrays, math, FFT, random) works.

For projects needing pandas / scipy / scikit-learn / lxml / cryptography, Cloudflare’s official Python Workers ships pre-built wheels for those via Pyodide — that’s the working answer today. Closing pymode’s catalog gap is a multi-month engineering project; we do it package-by-package as needed.

Tested & Working

These packages have conformance tests running in the actual workerd runtime:

Package	Type	Tests	Notes
pydantic	Rust ext (compiled to WASM)	4 passing	python-pydantic-core.wasm variant
fastapi	Pure Python (ASGI)	1 passing	Request validation, typed APIs
langchain-core	Pure Python	5 passing	Messages, documents, utils, serialization
langgraph	Pure Python	5 passing	StateGraph, compile, invoke, conditional edges
instructor	Pure Python	1 passing	Structured LLM output via pydantic
openai SDK	Pure Python	1 passing	Client types importable
numpy	C ext (compiled to WASM)	8 passing	python-numpy.wasm variant, arrays, FFT, random
jinja2	Pure Python	3 passing	Templates, loops, filters, inheritance
requests	Pure Python	1 passing	Session building, request construction
httpx	Pure Python	1 passing	Request building, URL parsing
beautifulsoup4	Pure Python	2 passing	HTML parsing, CSS selectors
pyyaml	Pure Python fallback	2 passing	YAML parse/dump
click	Pure Python	1 passing	CLI command creation
attrs	Pure Python	1 passing	Validators, asdict
starlette	Pure Python	1 passing	ASGI routing

The vitest suite covers these against the actual workerd runtime. Run npm test to see the current count and pass/fail status for your local environment.

Polyfilled C/Rust Extensions

These packages have native extensions replaced with pure-Python polyfills:

Package	Original	Polyfill
uuid_utils	Rust (_uuid_utils)	stdlib uuid + time + os.urandom
xxhash	C (_xxhash)	hashlib.sha256 as stand-in
ormsgpack	Rust (ormsgpack)	Wraps pure-Python msgpack
jiter	Rust (jiter.jiter)	stdlib json
markupsafe	C (_speedups)	Pure Python fallback (built-in)
multiprocessing	C (process model)	Polyfill: cpu_count()=1, exception classes

Memory budget

A single Durable Object has a 128 MB memory budget. Most pure-Python packages run comfortably; CPython itself plus a typical handler costs ~20-40 MB depending on imports. Packages that allocate large arrays (pandas, numpy) reach the limit faster.

The pymode.parallel API spawns child DOs for compute parallelism — each child has its own memory budget. This pattern can in principle fan out work that’s too large for one DO, but isn’t automatic; you opt in by calling spawn() / gather() from your handler.

Won’t Work in WASM (any DO)

Package	Reason
tensorflow	Needs GPU, ~500MB, C++/CUDA runtime
torch	Needs GPU, ~2GB, C++/CUDA runtime
scipy	Needs BLAS/LAPACK (Fortran), complex C

For ML inference, use Workers AI (env.AI.run()) instead.