pyhtml2text

Fast HTML to plain text and to Markdown for Python, written in Rust.

Relative links can be resolved against the page's URL, so the output's links work on their own.

Markdown

from pyhtml2text import html2markdown

html = """<h2>Before you start</h2>
<p>Read the <a href="setup.html">setup guide</a> and <b>back up</b> your data.</p>
<img src="/img/diagram.png" alt="How it fits together">
<table><tr><th>Plan</th><th>Price</th></tr><tr><td>Basic</td><td>Free</td></tr></table>"""

print(html2markdown(html, base_url="https://example.com/docs/intro.html"))

Output:

## Before you start

Read the [setup guide](https://example.com/docs/setup.html) and **back up** your data.

![How it fits together](https://example.com/img/diagram.png)

| Plan | Price |
| --- | --- |
| Basic | Free |

html2markdown(html, *, base_url=None) writes CommonMark with GitHub's tables and strikethrough, in the usual conventions:

• # headings, - bullets, _emphasis_ and **strong**, ~~strikethrough~~.
• Inline links and images, [text](url) and ![alt](src), made absolute when base_url is given. A link whose only content is an icon uses its aria-label or title as its text.
• Fenced code blocks, tagged with the language when the page names one (class="language-python").
• Pipe tables. Cells spanning rows or columns become empty cells, a table without a header row gets an empty one, and tables used for page layout are written as plain paragraphs.
• Characters are escaped only where Markdown would otherwise read them as syntax, so snake_case, 5 * 3 and $5 stay as they are.
• Lines aren't wrapped. Hidden content (hidden, aria-hidden, display: none), scripts, styles and form controls are left out.

The output is checked by rendering it back with a CommonMark parser and comparing every word, link and image with the original page, on real sites and on tens of thousands of random documents (tests/markdown_check.py).

Plain text

from pyhtml2text import html2text

html = '<p>Please <a href="/login">sign in</a> first.</p>'

print(html2text(html, base_url="https://example.com/docs/page.html"))
# Please [sign in][1] first.
#
# [1]: https://example.com/login

html2text(html, width=80, *, ...)

option	default
base_url	None	resolve relative hrefs against this URL (RFC 3986). Absolute and mailto: etc. links are left alone
decorate	True	markdown-ish *em*, **strong**, `code`
link_footnotes	True	list link targets as [1]: ... footnotes
table_borders	True	draw table borders
wrap_links	True	allow long links to wrap
max_wrap_width	None	wrap paragraphs narrower than width
raw	False	flatten tables so cell text stays together

Both functions take a str or UTF-8 bytes (e.g. response.content). Requires Python 3.10 or later.

Performance

Median time on an Apple Silicon Mac. To Markdown:

	Wikipedia article (1 MB)	BBC News front page (370 KB)	Python docs page (245 KB)
markdownify	184 ms	28 ms	97 ms
html2text (pure Python)	73 ms	13 ms	37 ms
html-to-markdown	44 ms	5.6 ms	15 ms
pyhtml2text	14 ms	1.7 ms	4.5 ms

To plain text:

	Wikipedia article (780 KB)	BBC News front page (370 KB)
html2text (pure Python)	71.5 ms	14.9 ms
html2text-rs	15.2 ms	2.5 ms
pyhtml2text	7.0 ms	1.3 ms

• The GIL is released while converting, so threads scale across cores. On free-threaded Python (3.14t and 3.15t) the module runs without re-enabling the GIL.
• Wheels for Linux (x86_64, aarch64), macOS and Windows x64 are built with profile-guided optimisation, which makes them about 25% faster. Other platforms get the same code without it.
• Deeply nested HTML is parsed in linear time, so hostile input can't stall a worker: 40,000 nested lists take 0.02 s, against 18 s with stock html5ever, and 100,000 levels of nesting followed by unmatched end tags take under 0.1 s.

Credits

HTML is parsed by a copy of html5ever (MIT/Apache-2.0) with servo/html5ever#800 by Josef Guenther applied, which makes parsing deeply nested HTML linear (see vendor/). Text is measured with unicode-width (MIT/Apache-2.0) and Markdown output is checked with pulldown-cmark (MIT). This package isn't affiliated with or endorsed by any of these projects, so please report issues here rather than upstream.