diff --git a/assets/javascripts/news.json b/assets/javascripts/news.json
index 06e4411595..b369dd257b 100644
--- a/assets/javascripts/news.json
+++ b/assets/javascripts/news.json
@@ -1,4 +1,8 @@
 [
+  [
+    "2026-09-04",
+    "New documentation: <a href=\"/valibot/\">Valibot</a>"
+  ],
   [
     "2026-08-15",
     "New documentations: <a href=\"/rack/\">Rack</a>, <a href=\"/pytest/\">pytest</a>"
diff --git a/lib/docs/scrapers/valibot.rb b/lib/docs/scrapers/valibot.rb
new file mode 100644
index 0000000000..ef2af1e295
--- /dev/null
+++ b/lib/docs/scrapers/valibot.rb
@@ -0,0 +1,218 @@
+require 'pathname'
+
+module Docs
+  class Valibot < UrlScraper
+    self.name = 'Valibot'
+    self.slug = 'valibot'
+    self.type = 'simple'
+    self.release = '1.4.2'
+    self.base_url = 'https://valibot.dev/'
+    self.links = {
+      home: 'https://valibot.dev/',
+      code: 'https://github.com/open-circle/valibot'
+    }
+
+    # https://github.com/open-circle/valibot/blob/main/LICENSE.md
+    options[:attribution] = <<-HTML
+      &copy; Fabian Hiller<br>
+      Licensed under the MIT License.
+    HTML
+
+    def get_latest_version(opts)
+      get_npm_version('valibot', opts)
+    end
+
+    # valibot.dev serves a Markdown version of every page at the same URL with a
+    # `.md` extension, and indexes all of them in llms.txt. Scraping those rather
+    # than the HTML pages saves us from stripping the framework noise Qwik leaves
+    # in the server-rendered markup (`<!--qv-->` comments and `q:*` attributes on
+    # nearly every node), and llms.txt also tells us which section a page belongs
+    # to, which is what the entry types are built from.
+    INDEX_PATH = 'llms.txt'
+
+    # The llms.txt sections to scrape. The blog is left out: those posts are
+    # release announcements and design write-ups, not documentation.
+    SECTIONS = ['Guides', 'API reference'].freeze
+
+    # The guide that becomes the documentation's root page.
+    ROOT_SOURCE_PATH = 'guides/introduction'
+
+    # A documentation page as listed in llms.txt. `source_path` is the path on
+    # valibot.dev, `path` the one used inside DevDocs (see #index_page).
+    Page = Struct.new(:name, :type, :path, :source_path)
+
+    # `path` is either a DevDocs path or the path of the page on valibot.dev,
+    # which only differ for the API reference types (see #index_page).
+    def build_page(path)
+      path = path.delete_prefix('/')
+      page = pages_by_path[path] || pages_by_source_path[path]
+      raise "#{self.class.name}: #{path.inspect} isn't listed in #{INDEX_PATH}" if page.nil?
+
+      response = request_one(source_url_for(page))
+      result = render(page, body_of(response)) if process_response?(response)
+      yield result if block_given?
+      result
+    end
+
+    def build_pages
+      pages_by_url = pages_by_source_path.values.index_by { |page| source_url_for(page) }
+      instrument 'running.scraper', urls: pages_by_url.keys
+
+      request_all pages_by_url.keys do |response|
+        if process_response?(response)
+          yield render(pages_by_url[response.url.to_s], body_of(response))
+        else
+          instrument 'ignore_response.scraper', response: response
+        end
+        nil # returning an Array would queue its values as additional URLs
+      end
+    end
+
+    private
+
+    # The pages are fetched as Markdown (`text/markdown`), which the inherited
+    # implementation rejects because it isn't an HTML content type.
+    def process_response?(response)
+      raise "Error status code (#{response.code}): #{response.url}" if response.error?
+      raise "Empty response body: #{response.url}" if response.blank?
+      response.success?
+    end
+
+    # Typhoeus hands back binary strings, whereas the documentation is UTF-8.
+    def body_of(response)
+      response.body.dup.force_encoding(Encoding::UTF_8)
+    end
+
+    def index_pages
+      @index_pages ||= parse_index(body_of(request_one(url_for(INDEX_PATH))))
+    end
+
+    def pages_by_path
+      @pages_by_path ||= index_pages.index_by(&:path)
+    end
+
+    def pages_by_source_path
+      @pages_by_source_path ||= index_pages.index_by(&:source_path)
+    end
+
+    def source_url_for(page)
+      url_for "#{page.source_path}.md"
+    end
+
+    # llms.txt is a flat Markdown document: `## Section`, `### Subsection` and
+    # `- [Name](url.md)` lines, in that order.
+    INDEX_ENTRY_REGEXP = /\A- \[(?<name>.+)\]\((?<url>\S+\.md)\)/
+
+    def parse_index(body)
+      section = subsection = nil
+
+      body.each_line.filter_map do |line|
+        case line
+        when /\A## (.+)/
+          section, subsection = $1.strip, nil
+          nil
+        when /\A### (.+)/
+          subsection = $1.strip
+          nil
+        when INDEX_ENTRY_REGEXP
+          name, url = $~[:name], $~[:url]
+          next unless SECTIONS.include?(section) && url.start_with?(base_url.to_s)
+          index_page name, section, subsection, url[base_url.to_s.length..].delete_suffix('.md')
+        end
+      end
+    end
+
+    def index_page(name, section, subsection, source_path)
+      if section == 'Guides'
+        path = source_path == ROOT_SOURCE_PATH ? 'index' : source_path
+        # The guide subsections overlap with the API reference ones ("Schemas"),
+        # so they are prefixed to keep the two apart in the sidebar.
+        type = "Guides: #{subsection}"
+      else
+        # DevDocs paths are case-insensitive (see NormalizePathsFilter), but
+        # Valibot names most of its types after the function they belong to
+        # (`Enum` vs `enum`). Types therefore get their own subdirectory.
+        basename = File.basename(source_path).downcase
+        path = subsection == 'Types' ? "api/types/#{basename}" : "api/#{basename}"
+        type = subsection
+      end
+
+      Page.new(name, type, path, source_path)
+    end
+
+    def render(page, body)
+      output = markdown_renderer.render(strip_markdown_note(body))
+      output = fix_code_blocks(output)
+      output = fix_links(output, page)
+      output << attribution_for(page)
+
+      { path: page.path, store_path: "#{page.path}.html", output: output, entries: [entry_for(page)] }
+    end
+
+    def entry_for(page)
+      # The root page is the documentation itself and has no name or type.
+      page.path == 'index' ? Entry.new(nil, 'index', nil) : Entry.new(page.name, page.path, page.type)
+    end
+
+    # Every page opens with a note pointing at its HTML version and at llms.txt,
+    # which is meaningless inside DevDocs.
+    MARKDOWN_NOTE_REGEXP = /^> This document is the Markdown version of [^\n]+\n\n?/
+
+    def strip_markdown_note(body)
+      body.sub MARKDOWN_NOTE_REGEXP, ''
+    end
+
+    # DevDocs highlights <pre> elements based on their data-language attribute.
+    def fix_code_blocks(html)
+      html.gsub(%r{<pre><code class="([^"]+)">}) { %(<pre data-language="#{$1}"><code>) }
+    end
+
+    def fix_links(html, page)
+      html.gsub(/href="(\/[^"]*)"/) { %(href="#{fix_link($1, page)}") }
+    end
+
+    def fix_link(href, page)
+      path, _, fragment = href.partition('#')
+      path = path.delete_prefix('/')
+
+      if (target = pages_by_source_path[path.delete_suffix('.md')])
+        path = relative_path_from(page.path, target.path)
+      else
+        # Anything that isn't scraped (the blog, the playground, the thesis PDF)
+        # is linked to the website instead.
+        path = File.join(base_url.to_s, path.sub(/\.md\z/, '/'))
+      end
+
+      fragment.empty? ? path : "#{path}##{fragment}"
+    end
+
+    def relative_path_from(path, other_path)
+      Pathname.new(other_path).relative_path_from(File.dirname(path)).to_s
+    end
+
+    # Mirrors AttributionFilter, which isn't in the pipeline because these pages
+    # never go through it. The link points at the HTML version of the page.
+    def attribution_for(page)
+      url = File.join(base_url.to_s, page.source_path, '/')
+      <<-HTML.strip_heredoc
+      <div class="_attribution">
+        <p class="_attribution-p">
+          #{options[:attribution].strip_heredoc.delete "\n"}<br>
+          <a href="#{url}" class="_attribution-link">#{url}</a>
+        </p>
+      </div>
+      HTML
+    end
+
+    def markdown_renderer
+      @markdown_renderer ||= Redcarpet::Markdown.new(
+        Redcarpet::Render::HTML.new(with_toc_data: true),
+        autolink: true,
+        fenced_code_blocks: true,
+        no_intra_emphasis: true,
+        strikethrough: true,
+        tables: true
+      )
+    end
+  end
+end
diff --git a/public/icons/docs/valibot/16.png b/public/icons/docs/valibot/16.png
new file mode 100644
index 0000000000..4b26b54465
Binary files /dev/null and b/public/icons/docs/valibot/16.png differ
diff --git a/public/icons/docs/valibot/16@2x.png b/public/icons/docs/valibot/16@2x.png
new file mode 100644
index 0000000000..fc7c53fe7d
Binary files /dev/null and b/public/icons/docs/valibot/16@2x.png differ
diff --git a/public/icons/docs/valibot/SOURCE b/public/icons/docs/valibot/SOURCE
new file mode 100644
index 0000000000..22b4aab9c5
--- /dev/null
+++ b/public/icons/docs/valibot/SOURCE
@@ -0,0 +1 @@
+https://valibot.dev/icon-32px.png