diff --git a/.gitignore b/.gitignore
index f30fd66606..1f2f58534f 100644
--- a/.gitignore
+++ b/.gitignore
@@ -19,3 +19,5 @@ lychee.toml
 lychee-report.md
 .phpunit.result.cache
 code_samples/_inline_php/
+/build/
+*.egg-info/
diff --git a/hooks.py b/hooks.py
index dfdbe4ba70..a88502fba9 100644
--- a/hooks.py
+++ b/hooks.py
@@ -1,96 +1,5 @@
-"""
-MkDocs hooks for Ibexa developer documentation.
+"""MkDocs hooks entry point — delegates to the installable llms_txt package."""
 
-- Keeps the llmstxt plugin's ``sections`` config in sync with the ``nav``
-  defined in ``mkdocs.yml``.
-- Post-processes the Markdown generated by the llmstxt plugin.
+from llms_txt.hooks import on_config, on_page_content
 
-All content transformations live in ``llmstxt_preprocess.py``; this module is
-only MkDocs event glue.
-"""
-
-from __future__ import annotations
-
-import sys
-from pathlib import Path, PurePosixPath
-from typing import TYPE_CHECKING
-
-_here = Path(__file__).parent
-if str(_here) not in sys.path:
-    sys.path.insert(0, str(_here))
-
-from llmstxt_preprocess import (
-    absolutize_image_urls,
-    editions_from_frontmatter,
-    expand_macros,
-    inject_page_metadata,
-    renumber_ordered_lists,
-)
-from update_llmstxt_config import convert_nav_to_llmstxt_sections
-
-if TYPE_CHECKING:
-    from mkdocs.config.defaults import MkDocsConfig
-    from mkdocs.structure.pages import Page
-
-
-def on_config(config: "MkDocsConfig") -> None:
-    """Populate llmstxt sections from nav before the build starts.
-
-    The llmstxt plugin reads ``config.sections`` in its ``on_files`` event,
-    which fires after ``on_config``, so injecting here is the right place.
-    """
-    nav = config.get("nav")
-    if not nav:
-        return
-
-    llmstxt = config["plugins"].get("llmstxt")
-    if llmstxt is None:
-        return
-
-    docs_dir = Path(config["docs_dir"])
-    llmstxt.config.sections = convert_nav_to_llmstxt_sections(nav, docs_dir)
-
-
-def on_page_content(html: str, *, page: "Page", config: "MkDocsConfig", **kwargs) -> None:
-    """Reformat the Markdown content generated by the llmstxt plugin for this page.
-
-    Hooks run after plugins for every event, so at this point the llmstxt
-    plugin has already generated Markdown and stored it in ``_md_pages``.
-    We reformat here so the plugin writes the corrected content in its own
-    ``on_post_build`` (which also runs before ours, for the same reason).
-    """
-    llmstxt = config["plugins"].get("llmstxt")
-    if llmstxt is None:
-        return
-
-    src_uri = page.file.src_uri
-    page_info = llmstxt._md_pages.get(src_uri)
-    if page_info is None:
-        return
-
-    frontmatter = _read_frontmatter(page, config)
-    editions = editions_from_frontmatter(frontmatter)
-    description = expand_macros(str(frontmatter.get("description") or ""), config.get("extra") or {})
-    if "[[=" in description:
-        # Unresolved macros must not leak into the output.
-        description = ""
-    content = inject_page_metadata(page_info.content, description, editions)
-    content = renumber_ordered_lists(content)
-    # Same base URL and page directory the plugin uses for making link hrefs absolute.
-    page_dir = PurePosixPath(page.file.dest_uri).parent.as_posix()
-    content = absolutize_image_urls(content, llmstxt._base_url, page_dir)
-    if content != page_info.content:
-        llmstxt._md_pages[src_uri] = page_info._replace(content=content)
-
-
-def _read_frontmatter(page: "Page", config: "MkDocsConfig") -> dict:
-    """Read the page's YAML frontmatter from its source file."""
-    src_path = Path(config["docs_dir"]) / page.file.src_path
-    try:
-        from mkdocs.utils import meta as mkdocs_meta
-        with open(src_path, encoding="utf-8") as f:
-            raw = f.read()
-        _, frontmatter = mkdocs_meta.get_data(raw)
-        return frontmatter
-    except Exception:
-        return {}
+__all__ = ["on_config", "on_page_content"]
diff --git a/llmstxt_preprocess.py b/llmstxt_preprocess.py
index f0e3c7c6fb..e4d5b126c9 100644
--- a/llmstxt_preprocess.py
+++ b/llmstxt_preprocess.py
@@ -1,547 +1,5 @@
-"""
-Markdown conversion logic for the llmstxt plugin.
+"""mkdocs-llmstxt ``preprocess:`` entry point — delegates to the installable llms_txt package."""
 
-This module is the single place holding all content transformations used to
-produce the Markdown version of the documentation:
+from llms_txt.llmstxt_preprocess import preprocess
 
-- ``preprocess(soup, output)`` — HTML-level transforms invoked by the
-  mkdocs-llmstxt plugin (configured in ``plugins.yml``) before the HTML is
-  converted to Markdown.
-- Markdown post-processing helpers (``renumber_ordered_lists``,
-  ``inject_edition_badges``, ``editions_from_frontmatter``) applied by
-  ``hooks.py`` to the Markdown the plugin generated.
-
-Note: the plugin loads this file by path (``spec_from_file_location``) while
-``hooks.py`` imports it as a regular module, so two module objects may coexist.
-Keep this module stateless (constants and pure functions only).
-"""
-
-import html as html_module
-import re
-from urllib.parse import urljoin, urlparse
-
-from bs4 import BeautifulSoup as Soup, NavigableString
-
-PILL_CLASS_TO_EDITION = {
-    "pill--lts-update": "LTS Update",
-    "pill--experience": "Experience",
-    "pill--commerce": "Commerce",
-    "pill--headless": "Headless",
-    "pill--new-feature": "New feature",
-    "pill--first-release": "First release",
-}
-
-FRONTMATTER_EDITION_DISPLAY = {
-    "lts-update": "LTS Update",
-    "experience": "Experience",
-    "commerce": "Commerce",
-    "headless": "Headless",
-}
-
-
-def preprocess(soup: Soup, output: str) -> None:
-    """
-    Preprocess HTML to improve markdown conversion.
-
-    Runs with autoclean disabled so we can control the order:
-    1. Expand tabbed sets with labels before autoclean removes tabbed-labels.
-    2. Run autoclean-equivalent cleanup.
-    3. Replace inline edition badge spans with readable text.
-    4. Remove release notes filter UI.
-    5. Convert card macros to markdown lists.
-
-    Note: frontmatter edition injection is handled in hooks.py on_page_content,
-    where page.file.src_path is available directly.
-    """
-    _process_tabbed_sets(soup)
-    _autoclean(soup)
-    _process_inline_pills(soup)
-    _process_release_note_tags(soup)
-    _process_release_note_dates(soup)
-    _process_release_notes_filters(soup)
-    _process_cards(soup)
-    _process_info_tiles(soup)
-    _process_tables(soup)
-    _process_admonitions(soup)
-
-
-# ---------------------------------------------------------------------------
-# Tables
-# ---------------------------------------------------------------------------
-
-def _process_tables(soup: Soup) -> None:
-    """Replace ✔ (U+2714) with 'Yes' in table cells for readability."""
-    CHECK = "✔"
-    for cell in soup.find_all(["td", "th"]):
-        for text_node in cell.find_all(string=lambda s: CHECK in s):
-            text_node.replace_with(NavigableString(text_node.replace(CHECK, "Yes")))
-
-
-# ---------------------------------------------------------------------------
-# Tabbed sets
-# ---------------------------------------------------------------------------
-
-def _process_tabbed_sets(soup: Soup) -> None:
-    """Prepend each tab label as bold text before its content block."""
-    for tabbed_set in soup.find_all("div", class_="tabbed-set"):
-        labels_div = tabbed_set.find("div", class_="tabbed-labels")
-        content_div = tabbed_set.find("div", class_="tabbed-content")
-
-        if not labels_div or not content_div:
-            tabbed_set.unwrap()
-            continue
-
-        labels = [label.get_text(strip=True) for label in labels_div.find_all("label")]
-        blocks = content_div.find_all("div", class_="tabbed-block", recursive=False)
-
-        wrapper = soup.new_tag("div")
-        for i, block in enumerate(blocks):
-            if i < len(labels):
-                label_tag = soup.new_tag("p")
-                strong = soup.new_tag("strong")
-                strong.string = labels[i]
-                label_tag.append(strong)
-                wrapper.append(label_tag)
-            for child in list(block.children):
-                wrapper.append(child.extract())
-
-        tabbed_set.replace_with(wrapper)
-
-
-# ---------------------------------------------------------------------------
-# Autoclean equivalent (mirrors mkdocs-llmstxt autoclean, minus tabbed-labels)
-# ---------------------------------------------------------------------------
-
-def _autoclean(soup: Soup) -> None:
-    """Replicate the plugin's autoclean so we can run it after tab processing."""
-
-    # Unwrap links that wrap an image (e.g. lightbox links) so the image
-    # itself decides its fate below.
-    for link in soup.find_all("a"):
-        img = link.find("img")
-        if img:
-            link.replace_with(img.extract())
-
-    # Keep images that have alt text so markdownify emits ![alt](src)
-    # (URLs are made absolute later by absolutize_image_urls); drop the rest,
-    # since an image without a description tells an AI agent nothing.
-    # Done in its own pass: mutating the tree from inside a find_all() predicate
-    # makes the traversal skip the elements that follow the mutated one.
-    for img in soup.find_all("img"):
-        alt = (img.get("alt") or "").strip()
-        if not alt:
-            img.decompose()
-
-    def _should_remove(tag) -> bool:
-        if tag.name == "svg":
-            return True
-        classes = tag.get("class") or ()
-        if tag.name == "a" and "headerlink" in classes:
-            return True
-        if "twemoji" in classes:
-            return True
-        # tabbed-labels are already consumed by _process_tabbed_sets, but
-        # handle any stragglers defensively.
-        if "tabbed-labels" in classes:
-            return True
-        return False
-
-    for element in soup.find_all(_should_remove):
-        element.decompose()
-
-    for element in soup.find_all("autoref"):
-        element.replace_with(NavigableString(element.get_text()))
-
-    # Insert ", " between adjacent <code> elements in table cells.
-    # html.parser silently drops </br> (invalid void-closing tag), so adjacent
-    # <code> blocks have no separator and markdownify concatenates their backticks.
-    for td in soup.find_all(["td", "th"]):
-        children = list(td.children)
-        for i in range(len(children) - 1):
-            curr = children[i]
-            nxt = children[i + 1]
-            if getattr(curr, "name", None) == "code" and getattr(nxt, "name", None) == "code":
-                curr.insert_after(NavigableString(", "))
-
-    for element in soup.find_all("div", attrs={"class": "doc-md-description"}):
-        element.replace_with(NavigableString(element.get_text().strip()))
-
-    for element in soup.find_all("span", attrs={"class": "doc-labels"}):
-        element.decompose()
-
-    # Flatten line-numbered code blocks to a plain <pre>, keeping the
-    # language-* class (emitted thanks to pygments_lang_class) on the <pre>,
-    # where markdownify's code_language_callback looks for it.
-    for element in soup.find_all("table", attrs={"class": "highlighttable"}):
-        code_elem = element.find("code")
-        if code_elem:
-            classes = code_elem.get("class") or ()
-            language = next((c for c in classes if c.startswith("language-")), "")
-            attr = f' class="{language}"' if language else ""
-            element.replace_with(
-                Soup(f"<pre{attr}>{html_module.escape(code_elem.get_text())}</pre>", "html.parser")
-            )
-
-# ---------------------------------------------------------------------------
-# Inline edition badge spans (from snippet includes)
-# ---------------------------------------------------------------------------
-
-def _pill_edition(node) -> str:
-    """Return the edition name of an inline pill span, or '' if not one."""
-    if getattr(node, "name", None) != "span":
-        return ""
-    classes = node.get("class") or []
-    if "pill--inline" not in classes:
-        return ""
-    for pill_cls, edition_name in PILL_CLASS_TO_EDITION.items():
-        if pill_cls in classes:
-            return edition_name
-    return ""
-
-
-def _process_inline_pills(soup: Soup) -> None:
-    """Replace inline edition pill spans with readable text.
-
-    Consecutive pills (possibly separated by whitespace) are merged into a
-    single parenthetical, e.g. ' (Experience, Commerce)' instead of
-    ' (Experience) (Commerce)'.
-    """
-    for span in soup.find_all("span", class_="pill--inline"):
-        if span.parent is None:  # already consumed as part of a previous run
-            continue
-        edition = _pill_edition(span)
-        if not edition:
-            continue
-
-        # Collect the run of pills that follow, skipping whitespace between them.
-        editions = [edition]
-        consumed = []
-        node = span.next_sibling
-        pending_whitespace = []
-        while node is not None:
-            if isinstance(node, NavigableString) and not node.strip():
-                pending_whitespace.append(node)
-                node = node.next_sibling
-                continue
-            next_edition = _pill_edition(node)
-            if not next_edition:
-                break
-            editions.append(next_edition)
-            consumed += pending_whitespace + [node]
-            pending_whitespace = []
-            node = node.next_sibling
-
-        for extra_node in consumed:
-            extra_node.extract()
-        span.replace_with(soup.new_string(f" ({', '.join(editions)})"))
-
-
-def _process_release_note_tags(soup: Soup) -> None:
-    """Append edition labels from release-note__tags divs to their preceding heading.
-
-    Release notes use a <div class="release-note__tags"> block after each <h2>
-    containing empty <div class="pill pill--X"> elements rendered via CSS.
-    This converts them to a readable parenthetical on the heading, e.g.:
-      ## Google Gemini connector v5.0.7 (Headless, Experience, LTS Update, New feature)
-    """
-    for tags_div in soup.find_all("div", class_="release-note__tags"):
-        editions = []
-        for pill_div in tags_div.find_all("div"):
-            classes = pill_div.get("class", [])
-            for pill_cls, name in PILL_CLASS_TO_EDITION.items():
-                if pill_cls in classes:
-                    editions.append(name)
-                    break
-
-        heading = tags_div.find_previous_sibling(["h1", "h2", "h3", "h4"])
-        if heading and editions:
-            # Insert before the permalink anchor so it's part of the heading text
-            anchor = heading.find("a", class_="headerlink")
-            label = NavigableString(f" ({', '.join(editions)})")
-            if anchor:
-                anchor.insert_before(label)
-            else:
-                heading.append(label)
-
-        tags_div.decompose()
-
-
-def _process_release_note_dates(soup: Soup) -> None:
-    """Prefix release-note dates with 'Release date: ' so the bare date line stays understandable."""
-    for date_div in soup.find_all("div", class_="release-note__date"):
-        date_text = date_div.get_text(strip=True)
-        if date_text:
-            date_div.string = f"Release date: {date_text}"
-
-
-def _process_info_tiles(soup: Soup) -> None:
-    """Simplify info-tile links to a single clean link text.
-
-    Info tiles have a 'Details' label div and a separate content div.
-    After SVG removal, markdownify produces broken multi-line link text.
-    Replace the whole <a> content with just the meaningful text.
-    """
-    for tile in soup.find_all("a", class_="info-tile"):
-        # The label div ("Details") can be discarded
-        label = tile.find("div", class_="info-tile__details")
-        if label:
-            label.decompose()
-
-        # Flatten remaining content to plain text
-        text = tile.get_text(separator=" ", strip=True)
-        tile.clear()
-        tile.append(soup.new_string(text))
-
-
-# ---------------------------------------------------------------------------
-# Admonitions
-# ---------------------------------------------------------------------------
-
-def _process_admonitions(soup: Soup) -> None:
-    """Convert admonition divs to blockquotes with a bold 'Type: Title' heading.
-
-    Input:  <div class="admonition caution">
-                <p class="admonition-title">Recommended versions</p>
-                <p>Body text...</p>
-            </div>
-
-    Output: <blockquote>
-                <p><strong>Caution: Recommended versions</strong></p>
-                <p>Body text...</p>
-            </blockquote>
-    """
-    for admonition in soup.find_all("div", class_="admonition"):
-        classes = admonition.get("class", [])
-        admonition_type = next((c for c in classes if c != "admonition"), None)
-
-        title_elem = admonition.find("p", class_="admonition-title")
-        title_text = title_elem.get_text(strip=True) if title_elem else ""
-
-        blockquote = soup.new_tag("blockquote")
-
-        # Bold title paragraph
-        title_p = soup.new_tag("p")
-        strong = soup.new_tag("strong")
-        prefix = f"{admonition_type.capitalize()}: " if admonition_type else ""
-        strong.string = f"{prefix}{title_text}"
-        title_p.append(strong)
-        blockquote.append(title_p)
-
-        # Body: all children except the title
-        for child in list(admonition.children):
-            if child == title_elem:
-                continue
-            blockquote.append(child.extract())
-
-        admonition.replace_with(blockquote)
-
-
-def _process_release_notes_filters(soup: Soup) -> None:
-    """Remove interactive release-notes filter UI elements."""
-    for container in soup.find_all("div", class_="release-notes-filters"):
-        container.decompose()
-
-
-# ---------------------------------------------------------------------------
-# Card macros
-# ---------------------------------------------------------------------------
-
-def _process_cards(soup: Soup) -> None:
-    """Convert card macro HTML structures into markdown-friendly lists with links."""
-    for cards_div in soup.find_all("div", class_=lambda c: c and c.startswith("cards ")):
-        card_wrappers = cards_div.find_all("div", class_="card-wrapper")
-
-        if not card_wrappers:
-            continue
-
-        ul = soup.new_tag("ul")
-
-        for card_wrapper in card_wrappers:
-            link = card_wrapper.find("a", class_="card")
-            if not link:
-                continue
-
-            href = link.get("href", "")
-            if href.startswith("//"):
-                href = "https:" + href
-
-            title_elem = link.find("p", class_="title")
-            description_elem = link.find("p", class_="description")
-
-            if not title_elem:
-                continue
-
-            title = title_elem.get_text(strip=True)
-            description = description_elem.get_text(strip=True) if description_elem else ""
-
-            li = soup.new_tag("li")
-            link_tag = soup.new_tag("a", href=href)
-            link_tag.string = title
-            li.append(link_tag)
-
-            if description:
-                li.append(soup.new_string(": "))
-                li.append(soup.new_string(description))
-
-            ul.append(li)
-
-        cards_div.replace_with(ul)
-
-
-# ---------------------------------------------------------------------------
-# Markdown post-processing (applied by hooks.py to the generated Markdown)
-# ---------------------------------------------------------------------------
-
-def editions_from_frontmatter(frontmatter: dict) -> list:
-    """Map ``edition``/``editions`` frontmatter values to display names."""
-
-    def _to_list(value):
-        if isinstance(value, list):
-            return value
-        if isinstance(value, str):
-            return value.split()
-        return []
-
-    all_editions = _to_list(frontmatter.get("edition")) + _to_list(frontmatter.get("editions") or [])
-    return [FRONTMATTER_EDITION_DISPLAY.get(e, e) for e in all_editions if e]
-
-
-_MACRO_RE = re.compile(r"\[\[=\s*(\w+)\s*=\]\]")
-
-
-def expand_macros(text: str, variables: dict) -> str:
-    """Expand simple ``[[= name =]]`` macro variables (mkdocs-macros syntax).
-
-    Only plain scalar variables are substituted; unknown or complex macros are
-    left untouched so callers can detect and handle them.
-    """
-
-    def _substitute(match: re.Match) -> str:
-        value = variables.get(match.group(1))
-        return str(value) if isinstance(value, (str, int, float)) else match.group(0)
-
-    return _MACRO_RE.sub(_substitute, text)
-
-
-def inject_page_metadata(content: str, description: str = "", editions: list = ()) -> str:
-    """Insert the page description and an 'Editions: X, Y' line after the first h1 heading."""
-    metadata_lines = []
-    if description:
-        metadata_lines += ["", description]
-    if editions:
-        metadata_lines += ["", "Editions: " + ", ".join(editions)]
-    if not metadata_lines:
-        return content
-
-    lines = content.split("\n")
-    for i, line in enumerate(lines):
-        if line.startswith("# "):
-            lines[i + 1:i + 1] = metadata_lines
-            return "\n".join(lines)
-
-    return "\n".join(metadata_lines).lstrip("\n") + "\n\n" + content
-
-
-# ![alt](url) or ![alt](url "title")
-_IMAGE_RE = re.compile(r'!\[([^\]]*)\]\(([^)\s]+)((?:\s+"[^"]*")?)\)')
-
-
-def absolutize_image_urls(content: str, base_url: str, page_dir: str) -> str:
-    """Rewrite relative image URLs in Markdown to absolute ones.
-
-    The llmstxt plugin makes link hrefs absolute but not image srcs, so the
-    ``![alt](src)`` references generated by markdownify keep their
-    page-relative paths, which break in ``llms-full.txt``. This mirrors the
-    URL logic the plugin applies to links.
-
-    ``base_url`` is the site base URL (trailing slash), ``page_dir`` the
-    page's directory relative to the site root (e.g. ``"cdp/cdp"``).
-    """
-
-    def _absolutize(match: re.Match) -> str:
-        alt, url, title = match.groups()
-        if not url.startswith(("/", "#")) and not urlparse(url).scheme:
-            relative_base = urljoin(base_url, page_dir + "/") if page_dir else base_url
-            url = urljoin(relative_base, url)
-        return f"![{alt}]({url}{title})"
-
-    return _IMAGE_RE.sub(_absolutize, content)
-
-
-_ORDERED_MARKER_RE = re.compile(r"^(\s*)(\d+)\. (.*)$")
-_FENCE_OPEN_RE = re.compile(r"^(\s*)(`{3,}|~{3,})")
-# Width of an ordered-list marker ("1. ") — mdformat normalizes markers to
-# "1. " and indents item content by exactly this much.
-_MARKER_WIDTH = 3
-
-
-def renumber_ordered_lists(content: str) -> str:
-    """Replace repeated '1.' ordered-list markers with sequential numbers.
-
-    The llmstxt plugin runs mdformat on the converted Markdown, which keeps
-    the first marker of each ordered list (the list's start number) and
-    rewrites all following markers to "1." (its default numbering style).
-    This restores sequential numbers while keeping track of nesting:
-
-    - a stack of (marker indent, counter) pairs tracks open ordered lists;
-      the counter starts at the list's first marker number, so intentional
-      start values (<ol start="N">, or literal "2\\." paragraphs in the
-      source) are preserved,
-    - blank lines and lines indented to an item's content column (marker
-      indent + 3) are item continuation and don't interrupt numbering,
-    - a less-indented line closes the lists it's not a continuation of,
-    - fenced code blocks are passed through untouched.
-
-    Known limitations (acceptable for LLM-oriented output): markers of items
-    >= 10 are one character wider than the 3-space content indent, and lists
-    inside blockquotes keep their "1." markers.
-    """
-    lines = content.split("\n")
-    result = []
-    stack = []  # [marker_indent, counter] per open ordered list, outermost first
-    fence = None  # (fence_char, fence_length) while inside a fenced code block
-
-    def close_lists(indent: int) -> None:
-        # A line not indented to the content column of an open list closes it.
-        while stack and indent < stack[-1][0] + _MARKER_WIDTH:
-            stack.pop()
-
-    for line in lines:
-        if fence is not None:
-            stripped = line.strip()
-            if stripped and set(stripped) == {fence[0]} and len(stripped) >= fence[1]:
-                fence = None
-            result.append(line)
-            continue
-
-        fence_match = _FENCE_OPEN_RE.match(line)
-        if fence_match:
-            close_lists(len(fence_match.group(1)))
-            marker = fence_match.group(2)
-            fence = (marker[0], len(marker))
-            result.append(line)
-            continue
-
-        if not line.strip():
-            # Blank lines separate loose-list items and item paragraphs; they
-            # never terminate a list on their own in mdformat output.
-            result.append(line)
-            continue
-
-        marker_match = _ORDERED_MARKER_RE.match(line)
-        if marker_match:
-            indent = len(marker_match.group(1))
-            while stack and stack[-1][0] > indent:
-                stack.pop()
-            if stack and stack[-1][0] == indent:
-                stack[-1][1] += 1
-            else:
-                # A new list keeps its first marker number as the start.
-                stack.append([indent, int(marker_match.group(2))])
-            result.append(f"{marker_match.group(1)}{stack[-1][1]}. {marker_match.group(3)}")
-            continue
-
-        close_lists(len(line) - len(line.lstrip()))
-        result.append(line)
-
-    return "\n".join(result)
+__all__ = ["preprocess"]
diff --git a/pyproject.toml b/pyproject.toml
new file mode 100644
index 0000000000..0f32f11dfa
--- /dev/null
+++ b/pyproject.toml
@@ -0,0 +1,17 @@
+[build-system]
+requires = ["setuptools>=61.0"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "ibexa-llms-txt"
+version = "0.1.0"
+description = "Shared MkDocs hooks and llmstxt preprocessing logic for Ibexa documentation sites."
+requires-python = ">=3.9"
+dependencies = [
+    "beautifulsoup4",
+    "PyYAML",
+]
+
+[tool.setuptools]
+package-dir = {"" = "tools"}
+packages = ["llms_txt"]
diff --git a/requirements.txt b/requirements.txt
index eb13f097ff..8edaf77560 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -8,3 +8,4 @@ mkdocs-redirects==1.2.2
 mkdocs-autolinks-plugin==0.7.1
 Jinja2==3.1.6
 mkdocs-llmstxt
+-e .
diff --git a/tests/python/conftest.py b/tests/python/conftest.py
index b9018a2f6b..dd499f159d 100644
--- a/tests/python/conftest.py
+++ b/tests/python/conftest.py
@@ -1,5 +1,7 @@
 import sys
 from pathlib import Path
 
-# Make repo-root modules (llmstxt_preprocess.py) importable.
+# Make repo-root shim modules (llmstxt_preprocess.py, update_llmstxt_config.py)
+# importable; the llms_txt package itself is available via the editable
+# install (`-e .` in requirements-dev.txt).
 sys.path.insert(0, str(Path(__file__).resolve().parents[2]))
diff --git a/tests/python/test_image_urls.py b/tests/python/test_image_urls.py
index d81118c173..ad75a197b3 100644
--- a/tests/python/test_image_urls.py
+++ b/tests/python/test_image_urls.py
@@ -1,4 +1,4 @@
-from llmstxt_preprocess import absolutize_image_urls
+from llms_txt.llmstxt_preprocess import absolutize_image_urls
 
 BASE = "https://doc.ibexa.co/en/latest/"
 
diff --git a/tests/python/test_page_metadata.py b/tests/python/test_page_metadata.py
index 5029316e42..5900b5c363 100644
--- a/tests/python/test_page_metadata.py
+++ b/tests/python/test_page_metadata.py
@@ -1,4 +1,4 @@
-from llmstxt_preprocess import (
+from llms_txt.llmstxt_preprocess import (
     editions_from_frontmatter,
     expand_macros,
     inject_page_metadata,
diff --git a/tests/python/test_preprocess.py b/tests/python/test_preprocess.py
index d8ef5e96bf..6a02b9e846 100644
--- a/tests/python/test_preprocess.py
+++ b/tests/python/test_preprocess.py
@@ -2,8 +2,8 @@
 from bs4 import BeautifulSoup
 from mkdocs_llmstxt._internal.plugin import _converter
 
-import llmstxt_preprocess
-from llmstxt_preprocess import renumber_ordered_lists
+import llms_txt.llmstxt_preprocess as llmstxt_preprocess
+from llms_txt.llmstxt_preprocess import renumber_ordered_lists
 
 
 def to_markdown(html: str) -> str:
diff --git a/tests/python/test_renumber.py b/tests/python/test_renumber.py
index 12e1a403fe..aab58ce8ab 100644
--- a/tests/python/test_renumber.py
+++ b/tests/python/test_renumber.py
@@ -1,4 +1,4 @@
-from llmstxt_preprocess import renumber_ordered_lists
+from llms_txt.llmstxt_preprocess import renumber_ordered_lists
 
 
 def test_flat_list():
diff --git a/tools/llms_txt/__init__.py b/tools/llms_txt/__init__.py
new file mode 100644
index 0000000000..e69de29bb2
diff --git a/tools/llms_txt/hooks.py b/tools/llms_txt/hooks.py
new file mode 100644
index 0000000000..b6b3edd1c9
--- /dev/null
+++ b/tools/llms_txt/hooks.py
@@ -0,0 +1,92 @@
+"""
+MkDocs hooks for the llmstxt Markdown-conversion feature.
+
+- Keeps the llmstxt plugin's ``sections`` config in sync with the ``nav``
+  defined in ``mkdocs.yml``.
+- Post-processes the Markdown generated by the llmstxt plugin.
+
+All content transformations live in ``llmstxt_preprocess``; this module is
+only MkDocs event glue. Fully generic — no documentation-project-specific
+content — so it's shared across doc sites via this installable package.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path, PurePosixPath
+from typing import TYPE_CHECKING
+
+from llms_txt.llmstxt_preprocess import (
+    absolutize_image_urls,
+    editions_from_frontmatter,
+    expand_macros,
+    inject_page_metadata,
+    renumber_ordered_lists,
+)
+from llms_txt.update_llmstxt_config import convert_nav_to_llmstxt_sections
+
+if TYPE_CHECKING:
+    from mkdocs.config.defaults import MkDocsConfig
+    from mkdocs.structure.pages import Page
+
+
+def on_config(config: "MkDocsConfig") -> None:
+    """Populate llmstxt sections from nav before the build starts.
+
+    The llmstxt plugin reads ``config.sections`` in its ``on_files`` event,
+    which fires after ``on_config``, so injecting here is the right place.
+    """
+    nav = config.get("nav")
+    if not nav:
+        return
+
+    llmstxt = config["plugins"].get("llmstxt")
+    if llmstxt is None:
+        return
+
+    docs_dir = Path(config["docs_dir"])
+    llmstxt.config.sections = convert_nav_to_llmstxt_sections(nav, docs_dir)
+
+
+def on_page_content(html: str, *, page: "Page", config: "MkDocsConfig", **kwargs) -> None:
+    """Reformat the Markdown content generated by the llmstxt plugin for this page.
+
+    Hooks run after plugins for every event, so at this point the llmstxt
+    plugin has already generated Markdown and stored it in ``_md_pages``.
+    We reformat here so the plugin writes the corrected content in its own
+    ``on_post_build`` (which also runs before ours, for the same reason).
+    """
+    llmstxt = config["plugins"].get("llmstxt")
+    if llmstxt is None:
+        return
+
+    src_uri = page.file.src_uri
+    page_info = llmstxt._md_pages.get(src_uri)
+    if page_info is None:
+        return
+
+    frontmatter = _read_frontmatter(page, config)
+    editions = editions_from_frontmatter(frontmatter)
+    description = expand_macros(str(frontmatter.get("description") or ""), config.get("extra") or {})
+    if "[[=" in description:
+        # Unresolved macros must not leak into the output.
+        description = ""
+    content = inject_page_metadata(page_info.content, description, editions)
+    content = renumber_ordered_lists(content)
+    # Same base URL and page directory the plugin uses for making link hrefs absolute.
+    page_dir = PurePosixPath(page.file.dest_uri).parent.as_posix()
+    content = absolutize_image_urls(content, llmstxt._base_url, page_dir)
+    if content != page_info.content:
+        llmstxt._md_pages[src_uri] = page_info._replace(content=content)
+
+
+def _read_frontmatter(page: "Page", config: "MkDocsConfig") -> dict:
+    """Read the page's YAML frontmatter from its source file."""
+    src_path = Path(config["docs_dir"]) / page.file.src_path
+    try:
+        from mkdocs.utils import meta as mkdocs_meta
+        with open(src_path, encoding="utf-8") as f:
+            raw = f.read()
+        _, frontmatter = mkdocs_meta.get_data(raw)
+        return frontmatter
+    except Exception:
+        return {}
diff --git a/tools/llms_txt/llmstxt_preprocess.py b/tools/llms_txt/llmstxt_preprocess.py
new file mode 100644
index 0000000000..97d83e3711
--- /dev/null
+++ b/tools/llms_txt/llmstxt_preprocess.py
@@ -0,0 +1,550 @@
+"""
+Markdown conversion logic for the llmstxt plugin.
+
+This module is the single place holding all content transformations used to
+produce the Markdown version of the documentation:
+
+- ``preprocess(soup, output)`` — HTML-level transforms invoked by the
+  mkdocs-llmstxt plugin (configured in ``plugins.yml``) before the HTML is
+  converted to Markdown.
+- Markdown post-processing helpers (``renumber_ordered_lists``,
+  ``inject_edition_badges``, ``editions_from_frontmatter``) applied by
+  ``hooks.py`` to the Markdown the plugin generated.
+
+This package (``llms_txt``) is installed as a dependency by other Ibexa doc
+sites, which each keep a thin root-level ``llmstxt_preprocess.py`` shim
+(loaded by the mkdocs-llmstxt plugin via ``spec_from_file_location``, since
+its ``preprocess:`` config option requires a real file path) that re-exports
+``preprocess`` from here. Keep this module stateless (constants and pure
+functions only).
+"""
+
+import html as html_module
+import re
+from urllib.parse import urljoin, urlparse
+
+from bs4 import BeautifulSoup as Soup, NavigableString
+
+PILL_CLASS_TO_EDITION = {
+    "pill--lts-update": "LTS Update",
+    "pill--experience": "Experience",
+    "pill--commerce": "Commerce",
+    "pill--headless": "Headless",
+    "pill--new-feature": "New feature",
+    "pill--first-release": "First release",
+}
+
+FRONTMATTER_EDITION_DISPLAY = {
+    "lts-update": "LTS Update",
+    "experience": "Experience",
+    "commerce": "Commerce",
+    "headless": "Headless",
+}
+
+
+def preprocess(soup: Soup, output: str) -> None:
+    """
+    Preprocess HTML to improve markdown conversion.
+
+    Runs with autoclean disabled so we can control the order:
+    1. Expand tabbed sets with labels before autoclean removes tabbed-labels.
+    2. Run autoclean-equivalent cleanup.
+    3. Replace inline edition badge spans with readable text.
+    4. Remove release notes filter UI.
+    5. Convert card macros to markdown lists.
+
+    Note: frontmatter edition injection is handled in hooks.py on_page_content,
+    where page.file.src_path is available directly.
+    """
+    _process_tabbed_sets(soup)
+    _autoclean(soup)
+    _process_inline_pills(soup)
+    _process_release_note_tags(soup)
+    _process_release_note_dates(soup)
+    _process_release_notes_filters(soup)
+    _process_cards(soup)
+    _process_info_tiles(soup)
+    _process_tables(soup)
+    _process_admonitions(soup)
+
+
+# ---------------------------------------------------------------------------
+# Tables
+# ---------------------------------------------------------------------------
+
+def _process_tables(soup: Soup) -> None:
+    """Replace ✔ (U+2714) with 'Yes' in table cells for readability."""
+    CHECK = "✔"
+    for cell in soup.find_all(["td", "th"]):
+        for text_node in cell.find_all(string=lambda s: CHECK in s):
+            text_node.replace_with(NavigableString(text_node.replace(CHECK, "Yes")))
+
+
+# ---------------------------------------------------------------------------
+# Tabbed sets
+# ---------------------------------------------------------------------------
+
+def _process_tabbed_sets(soup: Soup) -> None:
+    """Prepend each tab label as bold text before its content block."""
+    for tabbed_set in soup.find_all("div", class_="tabbed-set"):
+        labels_div = tabbed_set.find("div", class_="tabbed-labels")
+        content_div = tabbed_set.find("div", class_="tabbed-content")
+
+        if not labels_div or not content_div:
+            tabbed_set.unwrap()
+            continue
+
+        labels = [label.get_text(strip=True) for label in labels_div.find_all("label")]
+        blocks = content_div.find_all("div", class_="tabbed-block", recursive=False)
+
+        wrapper = soup.new_tag("div")
+        for i, block in enumerate(blocks):
+            if i < len(labels):
+                label_tag = soup.new_tag("p")
+                strong = soup.new_tag("strong")
+                strong.string = labels[i]
+                label_tag.append(strong)
+                wrapper.append(label_tag)
+            for child in list(block.children):
+                wrapper.append(child.extract())
+
+        tabbed_set.replace_with(wrapper)
+
+
+# ---------------------------------------------------------------------------
+# Autoclean equivalent (mirrors mkdocs-llmstxt autoclean, minus tabbed-labels)
+# ---------------------------------------------------------------------------
+
+def _autoclean(soup: Soup) -> None:
+    """Replicate the plugin's autoclean so we can run it after tab processing."""
+
+    # Unwrap links that wrap an image (e.g. lightbox links) so the image
+    # itself decides its fate below.
+    for link in soup.find_all("a"):
+        img = link.find("img")
+        if img:
+            link.replace_with(img.extract())
+
+    # Keep images that have alt text so markdownify emits ![alt](src)
+    # (URLs are made absolute later by absolutize_image_urls); drop the rest,
+    # since an image without a description tells an AI agent nothing.
+    # Done in its own pass: mutating the tree from inside a find_all() predicate
+    # makes the traversal skip the elements that follow the mutated one.
+    for img in soup.find_all("img"):
+        alt = (img.get("alt") or "").strip()
+        if not alt:
+            img.decompose()
+
+    def _should_remove(tag) -> bool:
+        if tag.name == "svg":
+            return True
+        classes = tag.get("class") or ()
+        if tag.name == "a" and "headerlink" in classes:
+            return True
+        if "twemoji" in classes:
+            return True
+        # tabbed-labels are already consumed by _process_tabbed_sets, but
+        # handle any stragglers defensively.
+        if "tabbed-labels" in classes:
+            return True
+        return False
+
+    for element in soup.find_all(_should_remove):
+        element.decompose()
+
+    for element in soup.find_all("autoref"):
+        element.replace_with(NavigableString(element.get_text()))
+
+    # Insert ", " between adjacent <code> elements in table cells.
+    # html.parser silently drops </br> (invalid void-closing tag), so adjacent
+    # <code> blocks have no separator and markdownify concatenates their backticks.
+    for td in soup.find_all(["td", "th"]):
+        children = list(td.children)
+        for i in range(len(children) - 1):
+            curr = children[i]
+            nxt = children[i + 1]
+            if getattr(curr, "name", None) == "code" and getattr(nxt, "name", None) == "code":
+                curr.insert_after(NavigableString(", "))
+
+    for element in soup.find_all("div", attrs={"class": "doc-md-description"}):
+        element.replace_with(NavigableString(element.get_text().strip()))
+
+    for element in soup.find_all("span", attrs={"class": "doc-labels"}):
+        element.decompose()
+
+    # Flatten line-numbered code blocks to a plain <pre>, keeping the
+    # language-* class (emitted thanks to pygments_lang_class) on the <pre>,
+    # where markdownify's code_language_callback looks for it.
+    for element in soup.find_all("table", attrs={"class": "highlighttable"}):
+        code_elem = element.find("code")
+        if code_elem:
+            classes = code_elem.get("class") or ()
+            language = next((c for c in classes if c.startswith("language-")), "")
+            attr = f' class="{language}"' if language else ""
+            element.replace_with(
+                Soup(f"<pre{attr}>{html_module.escape(code_elem.get_text())}</pre>", "html.parser")
+            )
+
+# ---------------------------------------------------------------------------
+# Inline edition badge spans (from snippet includes)
+# ---------------------------------------------------------------------------
+
+def _pill_edition(node) -> str:
+    """Return the edition name of an inline pill span, or '' if not one."""
+    if getattr(node, "name", None) != "span":
+        return ""
+    classes = node.get("class") or []
+    if "pill--inline" not in classes:
+        return ""
+    for pill_cls, edition_name in PILL_CLASS_TO_EDITION.items():
+        if pill_cls in classes:
+            return edition_name
+    return ""
+
+
+def _process_inline_pills(soup: Soup) -> None:
+    """Replace inline edition pill spans with readable text.
+
+    Consecutive pills (possibly separated by whitespace) are merged into a
+    single parenthetical, e.g. ' (Experience, Commerce)' instead of
+    ' (Experience) (Commerce)'.
+    """
+    for span in soup.find_all("span", class_="pill--inline"):
+        if span.parent is None:  # already consumed as part of a previous run
+            continue
+        edition = _pill_edition(span)
+        if not edition:
+            continue
+
+        # Collect the run of pills that follow, skipping whitespace between them.
+        editions = [edition]
+        consumed = []
+        node = span.next_sibling
+        pending_whitespace = []
+        while node is not None:
+            if isinstance(node, NavigableString) and not node.strip():
+                pending_whitespace.append(node)
+                node = node.next_sibling
+                continue
+            next_edition = _pill_edition(node)
+            if not next_edition:
+                break
+            editions.append(next_edition)
+            consumed += pending_whitespace + [node]
+            pending_whitespace = []
+            node = node.next_sibling
+
+        for extra_node in consumed:
+            extra_node.extract()
+        span.replace_with(soup.new_string(f" ({', '.join(editions)})"))
+
+
+def _process_release_note_tags(soup: Soup) -> None:
+    """Append edition labels from release-note__tags divs to their preceding heading.
+
+    Release notes use a <div class="release-note__tags"> block after each <h2>
+    containing empty <div class="pill pill--X"> elements rendered via CSS.
+    This converts them to a readable parenthetical on the heading, e.g.:
+      ## Google Gemini connector v5.0.7 (Headless, Experience, LTS Update, New feature)
+    """
+    for tags_div in soup.find_all("div", class_="release-note__tags"):
+        editions = []
+        for pill_div in tags_div.find_all("div"):
+            classes = pill_div.get("class", [])
+            for pill_cls, name in PILL_CLASS_TO_EDITION.items():
+                if pill_cls in classes:
+                    editions.append(name)
+                    break
+
+        heading = tags_div.find_previous_sibling(["h1", "h2", "h3", "h4"])
+        if heading and editions:
+            # Insert before the permalink anchor so it's part of the heading text
+            anchor = heading.find("a", class_="headerlink")
+            label = NavigableString(f" ({', '.join(editions)})")
+            if anchor:
+                anchor.insert_before(label)
+            else:
+                heading.append(label)
+
+        tags_div.decompose()
+
+
+def _process_release_note_dates(soup: Soup) -> None:
+    """Prefix release-note dates with 'Release date: ' so the bare date line stays understandable."""
+    for date_div in soup.find_all("div", class_="release-note__date"):
+        date_text = date_div.get_text(strip=True)
+        if date_text:
+            date_div.string = f"Release date: {date_text}"
+
+
+def _process_info_tiles(soup: Soup) -> None:
+    """Simplify info-tile links to a single clean link text.
+
+    Info tiles have a 'Details' label div and a separate content div.
+    After SVG removal, markdownify produces broken multi-line link text.
+    Replace the whole <a> content with just the meaningful text.
+    """
+    for tile in soup.find_all("a", class_="info-tile"):
+        # The label div ("Details") can be discarded
+        label = tile.find("div", class_="info-tile__details")
+        if label:
+            label.decompose()
+
+        # Flatten remaining content to plain text
+        text = tile.get_text(separator=" ", strip=True)
+        tile.clear()
+        tile.append(soup.new_string(text))
+
+
+# ---------------------------------------------------------------------------
+# Admonitions
+# ---------------------------------------------------------------------------
+
+def _process_admonitions(soup: Soup) -> None:
+    """Convert admonition divs to blockquotes with a bold 'Type: Title' heading.
+
+    Input:  <div class="admonition caution">
+                <p class="admonition-title">Recommended versions</p>
+                <p>Body text...</p>
+            </div>
+
+    Output: <blockquote>
+                <p><strong>Caution: Recommended versions</strong></p>
+                <p>Body text...</p>
+            </blockquote>
+    """
+    for admonition in soup.find_all("div", class_="admonition"):
+        classes = admonition.get("class", [])
+        admonition_type = next((c for c in classes if c != "admonition"), None)
+
+        title_elem = admonition.find("p", class_="admonition-title")
+        title_text = title_elem.get_text(strip=True) if title_elem else ""
+
+        blockquote = soup.new_tag("blockquote")
+
+        # Bold title paragraph
+        title_p = soup.new_tag("p")
+        strong = soup.new_tag("strong")
+        prefix = f"{admonition_type.capitalize()}: " if admonition_type else ""
+        strong.string = f"{prefix}{title_text}"
+        title_p.append(strong)
+        blockquote.append(title_p)
+
+        # Body: all children except the title
+        for child in list(admonition.children):
+            if child == title_elem:
+                continue
+            blockquote.append(child.extract())
+
+        admonition.replace_with(blockquote)
+
+
+def _process_release_notes_filters(soup: Soup) -> None:
+    """Remove interactive release-notes filter UI elements."""
+    for container in soup.find_all("div", class_="release-notes-filters"):
+        container.decompose()
+
+
+# ---------------------------------------------------------------------------
+# Card macros
+# ---------------------------------------------------------------------------
+
+def _process_cards(soup: Soup) -> None:
+    """Convert card macro HTML structures into markdown-friendly lists with links."""
+    for cards_div in soup.find_all("div", class_=lambda c: c and c.startswith("cards ")):
+        card_wrappers = cards_div.find_all("div", class_="card-wrapper")
+
+        if not card_wrappers:
+            continue
+
+        ul = soup.new_tag("ul")
+
+        for card_wrapper in card_wrappers:
+            link = card_wrapper.find("a", class_="card")
+            if not link:
+                continue
+
+            href = link.get("href", "")
+            if href.startswith("//"):
+                href = "https:" + href
+
+            title_elem = link.find("p", class_="title")
+            description_elem = link.find("p", class_="description")
+
+            if not title_elem:
+                continue
+
+            title = title_elem.get_text(strip=True)
+            description = description_elem.get_text(strip=True) if description_elem else ""
+
+            li = soup.new_tag("li")
+            link_tag = soup.new_tag("a", href=href)
+            link_tag.string = title
+            li.append(link_tag)
+
+            if description:
+                li.append(soup.new_string(": "))
+                li.append(soup.new_string(description))
+
+            ul.append(li)
+
+        cards_div.replace_with(ul)
+
+
+# ---------------------------------------------------------------------------
+# Markdown post-processing (applied by hooks.py to the generated Markdown)
+# ---------------------------------------------------------------------------
+
+def editions_from_frontmatter(frontmatter: dict) -> list:
+    """Map ``edition``/``editions`` frontmatter values to display names."""
+
+    def _to_list(value):
+        if isinstance(value, list):
+            return value
+        if isinstance(value, str):
+            return value.split()
+        return []
+
+    all_editions = _to_list(frontmatter.get("edition")) + _to_list(frontmatter.get("editions") or [])
+    return [FRONTMATTER_EDITION_DISPLAY.get(e, e) for e in all_editions if e]
+
+
+_MACRO_RE = re.compile(r"\[\[=\s*(\w+)\s*=\]\]")
+
+
+def expand_macros(text: str, variables: dict) -> str:
+    """Expand simple ``[[= name =]]`` macro variables (mkdocs-macros syntax).
+
+    Only plain scalar variables are substituted; unknown or complex macros are
+    left untouched so callers can detect and handle them.
+    """
+
+    def _substitute(match: re.Match) -> str:
+        value = variables.get(match.group(1))
+        return str(value) if isinstance(value, (str, int, float)) else match.group(0)
+
+    return _MACRO_RE.sub(_substitute, text)
+
+
+def inject_page_metadata(content: str, description: str = "", editions: list = ()) -> str:
+    """Insert the page description and an 'Editions: X, Y' line after the first h1 heading."""
+    metadata_lines = []
+    if description:
+        metadata_lines += ["", description]
+    if editions:
+        metadata_lines += ["", "Editions: " + ", ".join(editions)]
+    if not metadata_lines:
+        return content
+
+    lines = content.split("\n")
+    for i, line in enumerate(lines):
+        if line.startswith("# "):
+            lines[i + 1:i + 1] = metadata_lines
+            return "\n".join(lines)
+
+    return "\n".join(metadata_lines).lstrip("\n") + "\n\n" + content
+
+
+# ![alt](url) or ![alt](url "title")
+_IMAGE_RE = re.compile(r'!\[([^\]]*)\]\(([^)\s]+)((?:\s+"[^"]*")?)\)')
+
+
+def absolutize_image_urls(content: str, base_url: str, page_dir: str) -> str:
+    """Rewrite relative image URLs in Markdown to absolute ones.
+
+    The llmstxt plugin makes link hrefs absolute but not image srcs, so the
+    ``![alt](src)`` references generated by markdownify keep their
+    page-relative paths, which break in ``llms-full.txt``. This mirrors the
+    URL logic the plugin applies to links.
+
+    ``base_url`` is the site base URL (trailing slash), ``page_dir`` the
+    page's directory relative to the site root (e.g. ``"cdp/cdp"``).
+    """
+
+    def _absolutize(match: re.Match) -> str:
+        alt, url, title = match.groups()
+        if not url.startswith(("/", "#")) and not urlparse(url).scheme:
+            relative_base = urljoin(base_url, page_dir + "/") if page_dir else base_url
+            url = urljoin(relative_base, url)
+        return f"![{alt}]({url}{title})"
+
+    return _IMAGE_RE.sub(_absolutize, content)
+
+
+_ORDERED_MARKER_RE = re.compile(r"^(\s*)(\d+)\. (.*)$")
+_FENCE_OPEN_RE = re.compile(r"^(\s*)(`{3,}|~{3,})")
+# Width of an ordered-list marker ("1. ") — mdformat normalizes markers to
+# "1. " and indents item content by exactly this much.
+_MARKER_WIDTH = 3
+
+
+def renumber_ordered_lists(content: str) -> str:
+    """Replace repeated '1.' ordered-list markers with sequential numbers.
+
+    The llmstxt plugin runs mdformat on the converted Markdown, which keeps
+    the first marker of each ordered list (the list's start number) and
+    rewrites all following markers to "1." (its default numbering style).
+    This restores sequential numbers while keeping track of nesting:
+
+    - a stack of (marker indent, counter) pairs tracks open ordered lists;
+      the counter starts at the list's first marker number, so intentional
+      start values (<ol start="N">, or literal "2\\." paragraphs in the
+      source) are preserved,
+    - blank lines and lines indented to an item's content column (marker
+      indent + 3) are item continuation and don't interrupt numbering,
+    - a less-indented line closes the lists it's not a continuation of,
+    - fenced code blocks are passed through untouched.
+
+    Known limitations (acceptable for LLM-oriented output): markers of items
+    >= 10 are one character wider than the 3-space content indent, and lists
+    inside blockquotes keep their "1." markers.
+    """
+    lines = content.split("\n")
+    result = []
+    stack = []  # [marker_indent, counter] per open ordered list, outermost first
+    fence = None  # (fence_char, fence_length) while inside a fenced code block
+
+    def close_lists(indent: int) -> None:
+        # A line not indented to the content column of an open list closes it.
+        while stack and indent < stack[-1][0] + _MARKER_WIDTH:
+            stack.pop()
+
+    for line in lines:
+        if fence is not None:
+            stripped = line.strip()
+            if stripped and set(stripped) == {fence[0]} and len(stripped) >= fence[1]:
+                fence = None
+            result.append(line)
+            continue
+
+        fence_match = _FENCE_OPEN_RE.match(line)
+        if fence_match:
+            close_lists(len(fence_match.group(1)))
+            marker = fence_match.group(2)
+            fence = (marker[0], len(marker))
+            result.append(line)
+            continue
+
+        if not line.strip():
+            # Blank lines separate loose-list items and item paragraphs; they
+            # never terminate a list on their own in mdformat output.
+            result.append(line)
+            continue
+
+        marker_match = _ORDERED_MARKER_RE.match(line)
+        if marker_match:
+            indent = len(marker_match.group(1))
+            while stack and stack[-1][0] > indent:
+                stack.pop()
+            if stack and stack[-1][0] == indent:
+                stack[-1][1] += 1
+            else:
+                # A new list keeps its first marker number as the start.
+                stack.append([indent, int(marker_match.group(2))])
+            result.append(f"{marker_match.group(1)}{stack[-1][1]}. {marker_match.group(3)}")
+            continue
+
+        close_lists(len(line) - len(line.lstrip()))
+        result.append(line)
+
+    return "\n".join(result)
diff --git a/tools/llms_txt/update_llmstxt_config.py b/tools/llms_txt/update_llmstxt_config.py
new file mode 100644
index 0000000000..b21d3ef261
--- /dev/null
+++ b/tools/llms_txt/update_llmstxt_config.py
@@ -0,0 +1,131 @@
+#!/usr/bin/env python3
+"""
+Update the llmstxt plugin configuration in plugins.yml based on mkdocs.yml nav structure.
+This script converts the mkdocs navigation into a format suitable for the llmstxt plugin.
+
+Files can be excluded from the llmstxt output by adding the following to their YAML frontmatter:
+
+    exclude_from_llmstxt: true
+
+This excludes the page from both llms.txt and llms-full.txt, skips generating its Markdown
+version, and hides the page action buttons. There is no per-file control for excluding a page
+from llms.txt only (both files are derived from the same `sections` config).
+Files without this property are included by default.
+"""
+
+import re
+import yaml
+from pathlib import Path
+
+
+def read_frontmatter(file_path):
+    """
+    Parse YAML frontmatter from a markdown file.
+    Returns a dict of frontmatter values, or an empty dict if none is found.
+    Handles files that begin with HTML comments before the frontmatter block.
+    """
+    try:
+        content = Path(file_path).read_text(encoding='utf-8')
+    except (OSError, UnicodeDecodeError):
+        return {}
+
+    # Strip leading HTML comments (e.g. <!-- vale ... -->)
+    content = re.sub(r'\A(\s*<!--.*?-->\s*)+', '', content, flags=re.DOTALL)
+
+    if not content.startswith('---'):
+        return {}
+
+    end = content.find('\n---', 3)
+    if end == -1:
+        return {}
+
+    try:
+        return yaml.safe_load(content[3:end]) or {}
+    except yaml.YAMLError:
+        return {}
+
+
+def is_excluded(file_path, docs_dir):
+    """
+    Return True if the file has 'exclude_from_llmstxt: true' in its frontmatter.
+    file_path is relative to docs_dir (as written in mkdocs nav).
+    """
+    full_path = Path(docs_dir) / file_path
+    fm = read_frontmatter(full_path)
+    return fm.get('exclude_from_llmstxt', False) is True
+
+
+def convert_nav_to_llmstxt_sections(nav_list, docs_dir):
+    """
+    Convert mkdocs nav list to llmstxt sections format.
+    Returns a dict mapping top-level section names to flat lists of file paths.
+    Files with 'exclude_from_llmstxt: true' in their frontmatter are skipped.
+    Sections that contain no remaining files are omitted.
+    """
+    sections = {}
+
+    def extract_files(item):
+        """Recursively extract included file paths from a nav item."""
+        files = []
+        if isinstance(item, str):
+            if not item.endswith('.html') and not is_excluded(item, docs_dir):
+                files.append(item)
+        elif isinstance(item, list):
+            for subitem in item:
+                files.extend(extract_files(subitem))
+        elif isinstance(item, dict):
+            for value in item.values():
+                if isinstance(value, str):
+                    if not value.endswith('.html') and not is_excluded(value, docs_dir):
+                        files.append(value)
+                elif isinstance(value, list):
+                    for subitem in value:
+                        files.extend(extract_files(subitem))
+        return files
+
+    for item in nav_list:
+        if isinstance(item, dict):
+            for section_name, section_content in item.items():
+                files = extract_files({section_name: section_content})
+                if files:
+                    sections[section_name] = files
+        elif isinstance(item, str):
+            if not item.endswith('.html') and not is_excluded(item, docs_dir):
+                sections.setdefault('Ibexa Developer Documentation', []).append(item)
+
+    return sections
+
+
+def update_plugins_yml(plugins_path, mkdocs_path):
+    """
+    Update the llmstxt plugin configuration in plugins.yml based on mkdocs.yml nav.
+    """
+    with open(plugins_path, 'r') as f:
+        plugins_data = yaml.safe_load(f)
+
+    with open(mkdocs_path, 'r') as f:
+        mkdocs_data = yaml.safe_load(f)
+
+    # docs/ directory is resolved relative to mkdocs.yml location
+    docs_dir = Path(mkdocs_path).parent / mkdocs_data.get('docs_dir', 'docs')
+
+    nav = mkdocs_data.get('nav', [])
+    new_sections = convert_nav_to_llmstxt_sections(nav, docs_dir)
+
+    plugins_list = plugins_data.get('plugins', [])
+    for plugin in plugins_list:
+        if isinstance(plugin, dict) and 'llmstxt' in plugin:
+            plugin['llmstxt']['sections'] = new_sections
+            print(f"✓ Updated llmstxt plugin configuration")
+            print(f"  Total sections: {len(new_sections)}")
+            break
+    else:
+        print("✗ llmstxt plugin not found in plugins.yml")
+        return False
+
+    with open(plugins_path, 'w') as f:
+        yaml.dump(plugins_data, f, default_flow_style=False, sort_keys=False,
+                  allow_unicode=True, width=120)
+
+    print(f"✓ Updated {plugins_path}")
+    return True
diff --git a/update_llmstxt_config.py b/update_llmstxt_config.py
index b232392c38..f2ad8b04a0 100644
--- a/update_llmstxt_config.py
+++ b/update_llmstxt_config.py
@@ -1,135 +1,14 @@
 #!/usr/bin/env python3
 """
-Update the llmstxt plugin configuration in plugins.yml based on mkdocs.yml nav structure.
-This script converts the mkdocs navigation into a format suitable for the llmstxt plugin.
+CLI wrapper to update the llmstxt plugin configuration in plugins.yml based on mkdocs.yml nav structure.
 
-Files can be excluded from the llmstxt output by adding the following to their YAML frontmatter:
-
-    exclude_from_llmstxt: true
-
-This excludes the page from both llms.txt and llms-full.txt, skips generating its Markdown
-version, and hides the page action buttons. There is no per-file control for excluding a page
-from llms.txt only (both files are derived from the same `sections` config).
-Files without this property are included by default.
+The actual conversion logic lives in the installable ``llms_txt`` package;
+this script just runs it against this repo's plugins.yml/mkdocs.yml.
 """
 
-import re
-import yaml
 from pathlib import Path
 
-
-def read_frontmatter(file_path):
-    """
-    Parse YAML frontmatter from a markdown file.
-    Returns a dict of frontmatter values, or an empty dict if none is found.
-    Handles files that begin with HTML comments before the frontmatter block.
-    """
-    try:
-        content = Path(file_path).read_text(encoding='utf-8')
-    except (OSError, UnicodeDecodeError):
-        return {}
-
-    # Strip leading HTML comments (e.g. <!-- vale ... -->)
-    content = re.sub(r'\A(\s*<!--.*?-->\s*)+', '', content, flags=re.DOTALL)
-
-    if not content.startswith('---'):
-        return {}
-
-    end = content.find('\n---', 3)
-    if end == -1:
-        return {}
-
-    try:
-        return yaml.safe_load(content[3:end]) or {}
-    except yaml.YAMLError:
-        return {}
-
-
-def is_excluded(file_path, docs_dir):
-    """
-    Return True if the file has 'exclude_from_llmstxt: true' in its frontmatter.
-    file_path is relative to docs_dir (as written in mkdocs nav).
-    """
-    full_path = Path(docs_dir) / file_path
-    fm = read_frontmatter(full_path)
-    return fm.get('exclude_from_llmstxt', False) is True
-
-
-def convert_nav_to_llmstxt_sections(nav_list, docs_dir):
-    """
-    Convert mkdocs nav list to llmstxt sections format.
-    Returns a dict mapping top-level section names to flat lists of file paths.
-    Files with 'exclude_from_llmstxt: true' in their frontmatter are skipped.
-    Sections that contain no remaining files are omitted.
-    """
-    sections = {}
-
-    def extract_files(item):
-        """Recursively extract included file paths from a nav item."""
-        files = []
-        if isinstance(item, str):
-            if not item.endswith('.html') and not is_excluded(item, docs_dir):
-                files.append(item)
-        elif isinstance(item, list):
-            for subitem in item:
-                files.extend(extract_files(subitem))
-        elif isinstance(item, dict):
-            for value in item.values():
-                if isinstance(value, str):
-                    if not value.endswith('.html') and not is_excluded(value, docs_dir):
-                        files.append(value)
-                elif isinstance(value, list):
-                    for subitem in value:
-                        files.extend(extract_files(subitem))
-        return files
-
-    for item in nav_list:
-        if isinstance(item, dict):
-            for section_name, section_content in item.items():
-                files = extract_files({section_name: section_content})
-                if files:
-                    sections[section_name] = files
-        elif isinstance(item, str):
-            if not item.endswith('.html') and not is_excluded(item, docs_dir):
-                sections.setdefault('Ibexa Developer Documentation', []).append(item)
-
-    return sections
-
-
-def update_plugins_yml(plugins_path, mkdocs_path):
-    """
-    Update the llmstxt plugin configuration in plugins.yml based on mkdocs.yml nav.
-    """
-    with open(plugins_path, 'r') as f:
-        plugins_data = yaml.safe_load(f)
-
-    with open(mkdocs_path, 'r') as f:
-        mkdocs_data = yaml.safe_load(f)
-
-    # docs/ directory is resolved relative to mkdocs.yml location
-    docs_dir = Path(mkdocs_path).parent / mkdocs_data.get('docs_dir', 'docs')
-
-    nav = mkdocs_data.get('nav', [])
-    new_sections = convert_nav_to_llmstxt_sections(nav, docs_dir)
-
-    plugins_list = plugins_data.get('plugins', [])
-    for plugin in plugins_list:
-        if isinstance(plugin, dict) and 'llmstxt' in plugin:
-            plugin['llmstxt']['sections'] = new_sections
-            print(f"✓ Updated llmstxt plugin configuration")
-            print(f"  Total sections: {len(new_sections)}")
-            break
-    else:
-        print("✗ llmstxt plugin not found in plugins.yml")
-        return False
-
-    with open(plugins_path, 'w') as f:
-        yaml.dump(plugins_data, f, default_flow_style=False, sort_keys=False,
-                  allow_unicode=True, width=120)
-
-    print(f"✓ Updated {plugins_path}")
-    return True
-
+from llms_txt.update_llmstxt_config import update_plugins_yml
 
 if __name__ == '__main__':
     script_dir = Path(__file__).parent