etrafilatura¶

Web page extraction at Rust speed, with trafilatura's syntax. The Python binding of trafilatura-rs, a pure-Rust engine by nchapman.

pip install etrafilatura

import etrafilatura as trafilatura   # drop-in syntax

downloaded = trafilatura.fetch_url("https://example.org/article")
text = trafilatura.extract(downloaded, output_format="markdown", with_metadata=True)

Same verbs, keyword arguments, Document and lxml trees, renderers and settings.cfg as trafilatura, on a compiled engine and with three dependencies (lxml, urllib3, charset-normalizer) instead of seven. On the synthetic test pages the output is byte-identical in 64 of 84 format cases; on real pages the engine selects slightly different content → Compatibility.

Call	Does
`extract(html)`	Page → string in the chosen output format
`bare_extraction(html)`	Page → `Document` with `.text`, `.body`, `.comments`, metadata
`extract_with_metadata(html)`	Page → `Document`, metadata always included
`extract_metadata(html)`	Page → metadata-only `Document`
`fetch_url(url)` / `fetch_response(url)`	Download, decode, SSRF-guarded
`load_html(source)`	str/bytes/Response → lxml `HtmlElement`
`baseline(html)` / `html2txt(html)`	Last-resort extraction / tag stripping

Why¶

Speed¶

The extraction pipeline is Rust: html5ever parsing, scoring and fallbacks run compiled, with the GIL released.

Light¶

lxml, urllib3 and charset-normalizer are the only Python dependencies — no htmldate, courlan, justext or certifi, and no native toolchain at install time: abi3 wheels for Linux, macOS and Windows.

Compatible¶

Import it as trafilatura and keep your call sites: same signatures, output formats and configuration keys. Extensions are keyword-only, such as input_format="auto" to extract straight from a URL or a file.

Format by format → Outputs · every call → API · what can fail → Errors · gap-by-gap drop-in map → Compatibility

trafilatura-rs

Extraction quality, heuristics and performance belong to trafilatura-rs (itself a port of go-trafilatura). Extraction issues are upstream issues; anything about the Python API is ours.