better-profanity-fast

Fast better_profanity-compatible profanity filter. Matching runs in Rust; Python is only the public API.

There is no pure-Python fallback. Install a wheel for your OS/CPU, or build from the sdist with a Rust toolchain.

from better_profanity_fast import Profanity

filt = Profanity()  # default wordlist; construct once per process
filt.contains_profanity("he is a m0th3rf*cker")  # True
filt.censor("Dude, I hate shit.")  # 'Dude, I hate ****.'

In Django, create one Profanity() at import / AppConfig ready time. Do not build a new filter on every request.

Benchmark vs better_profanity 0.7

Same process, same 916-word list (wordlist.txt), CPython 3.14 on macOS arm64. Median of 3+ rounds. Speedup is 0.7 time / this package (higher means faster).

0.7 walks a VaryingString list per token, so a ~14k-character paragraph takes ~2.8 seconds. This package hashes leet variants and scans in Rust.

contains_profanity on dirty text early-exits here; 0.7 still scans. Censor always walks the full string on both sides, so the long-censor rows (~3700x) are the fairest like-for-like comparison.

Construct the filter once. Init is slower than 0.7 because we pre-expand leet hashes (~500k) at load time.

Matching is not identical to 0.7: this package also catches stretched spellings (peenis) and long infixes (ThisRetard). The table still uses the same inputs on both libraries.

Install

pip install better-profanity-fast

Requires CPython 3.9+. Wheels are abi3: one Linux/macOS/Windows wheel covers every CPython version from 3.9 through 3.14. Alpine needs the musllinux wheel (also published).

The default wordlist is wordlist.txt in the Python package, not inside the native library. Swap that file, or set BETTER_PROFANITY_WORDLIST to another path, then construct a new Profanity() — no Rust rebuild.

export BETTER_PROFANITY_WORDLIST=/etc/my-app/wordlist.txt

A dictionary letter may consume at most 8 identical input characters (peenis still matches; 9+ repeats of the same letter do not).

Digit-only tokens (1011, years, ports, order IDs) are never treated as leetspeak. Phrases that happen to contain digits (2 girls 1 cup) still match.

Local development

pip install maturin
maturin develop --release
cargo test --release --lib
python -m unittest discover -s tests -v

License

MIT. Copyright (c) 2026 Eric Carmichael. The default wordlist is adapted from better_profanity (MIT).