diff --git a/.gitignore b/.gitignore
index 894d169b79..6a00c44d4b 100644
--- a/.gitignore
+++ b/.gitignore
@@ -13,6 +13,8 @@
 # Ignore the build directory
 /build
 
+/repo-docs
+
 # Ignore cache
 /.sass-cache
 /.cache
diff --git a/Gemfile b/Gemfile
index 1e853a52f9..8c7bfcf1a7 100644
--- a/Gemfile
+++ b/Gemfile
@@ -17,6 +17,7 @@ gem "middleman-search_engine_sitemap"
 gem "git"
 gem "html-pipeline"
 gem "mdl"
+gem "open3"
 
 gem "govuk_publishing_components"
 gem "govuk_schemas"
diff --git a/Gemfile.lock b/Gemfile.lock
index 87bd5e5630..bbd33b7bb8 100644
--- a/Gemfile.lock
+++ b/Gemfile.lock
@@ -321,6 +321,7 @@ GEM
     octokit (6.1.1)
       faraday (>= 1, < 3)
       sawyer (~> 0.9)
+    open3 (0.1.2)
     openapi3_parser (0.9.2)
       commonmarker (~> 0.17)
     padrino-helpers (0.15.3)
@@ -497,6 +498,7 @@ DEPENDENCIES
   middleman
   middleman-search_engine_sitemap
   octokit
+  open3
   rake
   rspec
   rubocop-govuk
diff --git a/app/github_repo_fetcher.rb b/app/github_repo_fetcher.rb
index c49d2d7715..69e3e78b58 100644
--- a/app/github_repo_fetcher.rb
+++ b/app/github_repo_fetcher.rb
@@ -1,4 +1,5 @@
 require "octokit"
+require "open3"
 
 class GitHubRepoFetcher
   # The cache is only used locally, as GitHub Pages rebuilds the site
@@ -12,6 +13,7 @@ class GitHubRepoFetcher
   # TODO: we should supply a command line option to automate the
   # cache clearing step above.
   LOCAL_CACHE_DURATION = 1.week
+  REPO_DIR = "#{Bundler.root}/repo-docs/".freeze
 
   def self.instance
     @instance ||= new
@@ -39,11 +41,24 @@ def readme(repo_name)
   def docs(repo_name)
     return nil if repo(repo_name).private_repo?
 
-    CACHE.fetch("alphagov/#{repo_name} docs", expires_in: LOCAL_CACHE_DURATION) do
-      recursively_fetch_files(repo_name, "docs")
-    rescue Octokit::NotFound
-      nil
+    repo_root = File.join(REPO_DIR, repo_name)
+
+    unless File.exist?(repo_root)
+      puts "Cloning #{repo_name} docs"
+      git_commands = [
+        ["git", "clone", "-n", "--depth=1", "--filter=tree:0", "https://github.com/alphagov/#{repo_name}", repo_root],
+        ["git", "sparse-checkout", "set", "--no-cone", "docs", { chdir: repo_root }],
+        ["git", "checkout", { chdir: repo_root }],
+      ]
+      git_commands.each do |command|
+        _stdout_str, stderr_str, status = Open3.capture3(*command)
+        raise "Repo docs clone failed with error: #{stderr_str}" unless status.success?
+      end
     end
+
+    return nil unless File.exist?(File.join(repo_root, "docs"))
+
+    recursively_fetch_files(repo_name, %w[docs])
   end
 
 private
@@ -62,28 +77,35 @@ def latest_commit(repo_name, path)
     }
   end
 
-  def recursively_fetch_files(repo_name, path)
-    docs = client.contents("alphagov/#{repo_name}", path:)
-    top_level_files = docs.select { |doc| doc.path.end_with?(".md") }.map do |doc|
-      data_for_github_doc(doc, repo_name)
+  def recursively_fetch_files(repo_name, path_stack)
+    repo_dir = Dir["#{File.join(REPO_DIR, repo_name, *path_stack)}/*"]
+    top_level_files = repo_dir.select { |file_path| File.file?(file_path) && file_path.end_with?(".md") }.map do |doc_path|
+      data_for_github_doc(doc_path, repo_name)
     end
-    docs.select { |doc| doc.type == "dir" }.each_with_object(top_level_files) do |dir, files|
-      files.concat(recursively_fetch_files(repo_name, dir.path))
+    repo_dir.select { |file_path| File.directory?(file_path) }.each_with_object(top_level_files) do |dir, files|
+      files.concat(recursively_fetch_files(repo_name, path_stack << File.basename(dir)))
     end
   end
 
-  def data_for_github_doc(doc, repo_name)
-    contents = HTTP.get(doc.download_url)
-    docs_path = doc.path.sub("docs/", "").match(/(.+)\..+$/)[1]
-    filename = docs_path.split("/")[-1]
-    title = ExternalDoc.title(contents) || filename
+  def data_for_github_doc(doc_path, repo_name)
+    contents = File.read(doc_path)
+
+    path = Pathname.new(doc_path)
+    docs_path_with_extension = path.each_filename.to_a.drop_while { |f| f != "docs" }.drop(1).join("/")
+    docs_path_without_extension = docs_path_with_extension.chomp(".md")
+    relative = path.each_filename.to_a.drop_while { |f| f != "docs" }.join("/")
+
+    basename_without_extension = File.basename(doc_path, ".md")
+    title = ExternalDoc.title(contents) || basename_without_extension
+
+    url = "https://github.com/alphagov/#{repo_name}/blob/main/#{relative}"
     {
-      path: "/repos/#{repo_name}/#{docs_path}.html",
+      path: "/repos/#{repo_name}/#{docs_path_without_extension}.html",
       title: title.to_s.force_encoding("UTF-8"),
       markdown: contents.to_s.force_encoding("UTF-8"),
-      relative_path: doc.path,
-      source_url: doc.html_url,
-      latest_commit: latest_commit(repo_name, doc.path),
+      relative_path: relative,
+      source_url: url,
+      latest_commit: latest_commit(repo_name, relative),
     }
   end
 
diff --git a/spec/app/github_repo_fetcher_spec.rb b/spec/app/github_repo_fetcher_spec.rb
index 425f119502..327d199be9 100644
--- a/spec/app/github_repo_fetcher_spec.rb
+++ b/spec/app/github_repo_fetcher_spec.rb
@@ -1,5 +1,6 @@
 RSpec.describe GitHubRepoFetcher do
   before :each do
+    stub_const("GitHubRepoFetcher::REPO_DIR", "#{Bundler.root}/spec/fixtures/repo-docs/".freeze)
     stub_request(:get, "https://api.github.com/users/alphagov/repos?per_page=100")
       .to_return(
         body: "[ { \"name\": \"some-repo\", \"default_branch\": \"main\" } ]",
@@ -116,13 +117,9 @@ def readme_url
   end
 
   describe "#docs" do
-    let(:repo_name) { SecureRandom.uuid }
+    let(:repo_name) { "some-repo" }
     let(:commit) { { sha: SecureRandom.hex(40), timestamp: Time.now.utc.to_s } }
 
-    def docs_url(repo_name)
-      "https://api.github.com/repos/alphagov/#{repo_name}/contents/docs"
-    end
-
     def github_repo_fetcher_returning(repo)
       instance = GitHubRepoFetcher.new
       allow(instance).to receive(:repo).with(repo_name).and_return(repo)
@@ -133,96 +130,46 @@ def github_repo_fetcher_returning(repo)
     context "the repo contains a reachable docs/ folder" do
       let(:expected_hash_structure) { hash_including(:title, :markdown, :path, :relative_path, :source_url, :latest_commit) }
 
-      def with_stubbed_client(temporary_client, instance)
-        before_client = instance.instance_variable_get(:@client)
-        instance.instance_variable_set(:@client, temporary_client)
-        yield
-        instance.instance_variable_set(:@client, before_client)
-      end
-
-      def stub_doc(contents: "arbitrary contents", path: "docs/foo.md")
-        doc = double("doc", type: "file", download_url: "foo_url", path:, html_url: "foo_html_url")
-        allow(HTTP).to receive(:get).with(doc.download_url).and_return(contents)
-        doc
-      end
-
-      it "returns an array of hashes" do
-        instance = github_repo_fetcher_returning(public_repo)
-
-        with_stubbed_client(double("Octokit::Client", contents: [stub_doc]), instance) do
-          expect(instance.docs(repo_name)).to match([expected_hash_structure])
-        end
-      end
-
       it "derives each document title from its markdown" do
         instance = github_repo_fetcher_returning(public_repo)
-        doc = stub_doc(contents: "# title \n Some document")
-
-        with_stubbed_client(double("Octokit::Client", contents: [doc]), instance) do
-          doc = instance.docs(repo_name).first
-          expect(doc[:title]).to eq("title")
-        end
+        allow(File).to receive(:read).and_return("# title \n Some document")
+        expect(instance.docs(repo_name).first[:title]).to eq("title")
       end
 
       it "derives document title from its filename if not present in markdown" do
         instance = github_repo_fetcher_returning(public_repo)
-        doc = stub_doc(contents: "bar \n Some document")
-
-        with_stubbed_client(double("Octokit::Client", contents: [doc]), instance) do
-          doc = instance.docs(repo_name).first
-          expect(doc[:title]).to eq("foo")
-        end
+        allow(File).to receive(:read).and_return("bar \n Some document")
+        expect(instance.docs(repo_name).first[:title]).to eq("foo")
       end
 
       it "maintains the original directory structure" do
         instance = github_repo_fetcher_returning(public_repo)
-        doc = stub_doc(path: "docs/subdir/foo.md")
-
-        with_stubbed_client(double("Octokit::Client", contents: [doc]), instance) do
-          doc = instance.docs(repo_name).first
-          expect(doc[:path]).to eq("/repos/#{repo_name}/subdir/foo.html")
-          expect(doc[:relative_path]).to eq("docs/subdir/foo.md")
-        end
+        doc = instance.docs(repo_name).second
+        expect(doc[:path]).to eq("/repos/#{repo_name}/foo/bar.html")
+        expect(doc[:relative_path]).to eq("docs/foo/bar.md")
       end
 
       it "retrieves documents recursively" do
-        dir = double("dir", type: "dir", path: "docs/foo")
-        nested_doc = stub_doc(path: "docs/foo/bar.md")
-
         instance = github_repo_fetcher_returning(public_repo)
-        allow(HTTP).to receive(:get).with(nested_doc.download_url).and_return("some contents")
-        stubbed_client = double("Octokit::Client")
-        allow(stubbed_client).to receive(:contents).with("alphagov/#{repo_name}", path: "docs")
-          .and_return([dir])
-        allow(stubbed_client).to receive(:contents).with("alphagov/#{repo_name}", path: "docs/foo")
-          .and_return([nested_doc])
-
-        with_stubbed_client(stubbed_client, instance) do
-          expect(instance.docs(repo_name)).to match([expected_hash_structure])
-        end
+        expect(instance.docs(repo_name)).to include(expected_hash_structure)
       end
 
       it "skips over any non-markdown files" do
         instance = github_repo_fetcher_returning(public_repo)
-        non_markdown_file = double("non markdown file", type: "file", path: "docs/digests.png")
-
-        with_stubbed_client(double("Octokit::Client", contents: [non_markdown_file]), instance) do
-          expect(instance.docs(repo_name)).to eq([])
-        end
+        allow(Dir).to receive(:[]).and_return(["#{Bundler.root}/spec/fixtures/repo-docs/some-repo/docs/digests.png"])
+        expect(instance.docs(repo_name)).to eq([])
       end
     end
 
     it "returns nil if no docs folder exists" do
       instance = github_repo_fetcher_returning(public_repo)
-      stub_request(:get, docs_url(repo_name))
-        .to_return(status: 404, body: "{}", headers: { content_type: "application/json" })
-
-      expect(instance.docs(repo_name)).to be_nil
+      allow(File).to receive(:exist?).and_call_original
+      allow(File).to receive(:exist?).with("#{Bundler.root}/spec/fixtures/repo-docs/some-repo/docs").and_return(false)
+      expect(instance.docs(repo_name)).to eq(nil)
     end
 
     it "returns nil if the repo is private" do
       instance = github_repo_fetcher_returning(private_repo)
-
       expect(instance.docs(repo_name)).to eq(nil)
     end
   end
diff --git a/spec/fixtures/repo-docs/some-repo/docs/foo.md b/spec/fixtures/repo-docs/some-repo/docs/foo.md
new file mode 100644
index 0000000000..e074f2fb7b
--- /dev/null
+++ b/spec/fixtures/repo-docs/some-repo/docs/foo.md
@@ -0,0 +1,32 @@
+# Lorem ipsum
+
+## tl;dr
+
+Lorem ipsum dolor sit amet, consectetur adipiscing elit. Nam efficitur lacus in
+neque consequat porta, [Suspendisse iaculis](#suspendisse-iaculis). Sed et
+[inline link](./inline-link.md), imperdiet eros ut, [aliased link].
+[Absolute link](https://nam.com/eget/dui/absolute-link.md)
+
+[localhost](localhost:999)
+
+[Relative docs link with period](./docs/prefixed.md)
+
+[Relative docs link without period](docs/no-prefix.md)
+
+[Link relative to root](/public/json_examples/requests/foo.json)
+
+[Subfolder](docs/some-subfolder/foo.md)
+
+Visit “http://localhost:3108”
+
+## Suspendisse iaculis
+
+![Suspendisse iaculis](suspendisse_iaculis.png)
+
+[aliased link]: ./lib/aliased_link.rb
+
+## Other headings to test
+
+### data.gov.uk
+
+### Patterns & Style Guides
diff --git a/spec/fixtures/repo-docs/some-repo/docs/foo/bar.md b/spec/fixtures/repo-docs/some-repo/docs/foo/bar.md
new file mode 100644
index 0000000000..e074f2fb7b
--- /dev/null
+++ b/spec/fixtures/repo-docs/some-repo/docs/foo/bar.md
@@ -0,0 +1,32 @@
+# Lorem ipsum
+
+## tl;dr
+
+Lorem ipsum dolor sit amet, consectetur adipiscing elit. Nam efficitur lacus in
+neque consequat porta, [Suspendisse iaculis](#suspendisse-iaculis). Sed et
+[inline link](./inline-link.md), imperdiet eros ut, [aliased link].
+[Absolute link](https://nam.com/eget/dui/absolute-link.md)
+
+[localhost](localhost:999)
+
+[Relative docs link with period](./docs/prefixed.md)
+
+[Relative docs link without period](docs/no-prefix.md)
+
+[Link relative to root](/public/json_examples/requests/foo.json)
+
+[Subfolder](docs/some-subfolder/foo.md)
+
+Visit “http://localhost:3108”
+
+## Suspendisse iaculis
+
+![Suspendisse iaculis](suspendisse_iaculis.png)
+
+[aliased link]: ./lib/aliased_link.rb
+
+## Other headings to test
+
+### data.gov.uk
+
+### Patterns & Style Guides