Fix: satisfy pyrefly in test_pdf_ops and clarify pdf_ops docstrings

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Chore: drop stale type-check baseline entries for bulk_edit
2026-10-07 16:47:13 +00:00 · 2026-10-03 15:03:31 -07:00 · 2026-10-03 14:59:01 -07:00 · 2026-10-03 14:58:36 -07:00 · 2026-10-03 14:54:15 -07:00 · 2026-10-03 14:47:53 -07:00
60 changed files with 9121 additions and 16408 deletions
@@ -0,0 +1,53 @@
+name: Install apt packages
+description: >
+  Installs apt packages with a per-attempt timeout and retries. The hosted runner's Azure apt mirror intermittently serves archives at a crawl, which apt's own timeout does not catch. A stuck attempt is killed and retried, which usually lands on a healthy mirror node.
+
+inputs:
+  packages:
+    description: Space separated list of packages to install
+    required: true
+  attempts:
+    description: Maximum number of attempts
+    default: "3"
+  attempt-timeout:
+    description: Seconds before a single attempt is killed
+    default: "180"
+runs:
+  using: composite
+  steps:
+    - name: Install ${{ inputs.packages }}
+      shell: bash
+      env:
+        PACKAGES: ${{ inputs.packages }}
+        ATTEMPTS: ${{ inputs.attempts }}
+        ATTEMPT_TIMEOUT: ${{ inputs.attempt-timeout }}
+      run: |
+        SUDO=""
+        if [ "$(id -u)" -ne 0 ]; then
+          SUDO="sudo"
+        fi
+
+        attempt_install() {
+          $SUDO apt-get update -q
+          # Intentionally unquoted, this is a list of package names
+          $SUDO apt-get install -q --yes --no-install-recommends $PACKAGES
+        }
+        export -f attempt_install
+        export SUDO PACKAGES
+
+        for attempt in $(seq 1 "${ATTEMPTS}"); do
+          echo "::group::apt install attempt ${attempt}/${ATTEMPTS}"
+          start=$(date +%s)
+          if timeout --kill-after=10 "${ATTEMPT_TIMEOUT}" bash -c attempt_install; then
+            echo "::endgroup::"
+            echo "Installed in $(( $(date +%s) - start ))s"
+            exit 0
+          fi
+          echo "::endgroup::"
+          echo "::warning::apt install attempt ${attempt}/${ATTEMPTS} failed or exceeded ${ATTEMPT_TIMEOUT}s"
+          # A killed attempt can leave dpkg half configured
+          $SUDO dpkg --configure -a || true
+        done
+
+        echo "::error::apt install failed after ${ATTEMPTS} attempts"
+        exit 1
@@ -59,7 +59,6 @@ updates:
          - "drf-*"
          - "djangorestframework"
          - "whitenoise"
-          - "bleach"
          - "jinja2"
      # Async, Task Queuing & Caching
      async-tasks:
@@ -102,17 +102,16 @@ jobs:
        with:
          python-version: "${{ matrix.python-version }}"
      - name: Install uv
-        uses: astral-sh/setup-uv@bec219d24cd3e171d82865faccec33120bb574f4 # v10.1.0
+        uses: astral-sh/setup-uv@c18668ad3cf93ea998bef934396af7bb5c839dc7 # v10.2.0
        with:
          version: ${{ env.DEFAULT_UV_VERSION }}
          enable-cache: true
          python-version: ${{ steps.setup-python.outputs.python-version }}
      - name: Install system dependencies
-        timeout-minutes: 10
-        run: |
-          sudo apt-get update -qq
-          sudo apt-get install -qq --no-install-recommends \
-            unpaper tesseract-ocr imagemagick ghostscript poppler-utils
+        timeout-minutes: 12
+        uses: $/.github/actions/apt-install
+        with:
+          packages: unpaper tesseract-ocr imagemagick ghostscript poppler-utils
      - name: Configure ImageMagick
        run: |
          sudo cp docker/rootfs/etc/ImageMagick-6/paperless-policy.xml /etc/ImageMagick-6/policy.xml
@@ -139,13 +138,13 @@ jobs:
            pytest
      - name: Upload test results to Codecov
        if: always()
-        uses: codecov/codecov-action@fb8b3582c8e4def4969c97caa2f19720cb33a72f # v7.0.0
+        uses: codecov/codecov-action@303a32d7a59b442fa8d48b6a1cc6825c09c847a5 # v7.1.1
        with:
          flags: backend-python-${{ matrix.python-version }}
          files: junit.xml
          report_type: test_results
      - name: Upload coverage to Codecov
-        uses: codecov/codecov-action@fb8b3582c8e4def4969c97caa2f19720cb33a72f # v7.0.0
+        uses: codecov/codecov-action@303a32d7a59b442fa8d48b6a1cc6825c09c847a5 # v7.1.1
        with:
          flags: backend-python-${{ matrix.python-version }}
          files: coverage.xml
@@ -176,7 +175,7 @@ jobs:
        with:
          python-version: "${{ env.DEFAULT_PYTHON }}"
      - name: Install uv
-        uses: astral-sh/setup-uv@bec219d24cd3e171d82865faccec33120bb574f4 # v10.1.0
+        uses: astral-sh/setup-uv@c18668ad3cf93ea998bef934396af7bb5c839dc7 # v10.2.0
        with:
          version: ${{ env.DEFAULT_UV_VERSION }}
          enable-cache: true
@@ -106,7 +106,7 @@ jobs:
          echo "repository=${repo_name}"
          echo "name=${repo_name}" >> $GITHUB_OUTPUT
      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@37fe631027851001ddb9b187196cc803df7f5f0e # v4.3.0
+        uses: docker/setup-buildx-action@f87e5991a6d7451dcb8d9637bfbc97413f497069 # v4.4.1
      - name: Login to GitHub Container Registry
        uses: docker/login-action@dbcb813823bdd20940b903addbd779551569679f # v4.6.0
        with:
@@ -132,7 +132,7 @@ jobs:
            type=semver,pattern={{major}}.{{minor}}
      - name: Build and push by digest
        id: build
-        uses: docker/build-push-action@53b7df96c91f9c12dcc8a07bcb9ccacbed38856a # v7.3.0
+        uses: docker/build-push-action@c3c9e263c25d99ce0380d002d59b67737d91b0dc # v7.4.0
        with:
          context: .
          file: ./Dockerfile
@@ -182,7 +182,7 @@ jobs:
          echo "Downloaded digests:"
          ls -la /tmp/digests/
      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@37fe631027851001ddb9b187196cc803df7f5f0e # v4.3.0
+        uses: docker/setup-buildx-action@f87e5991a6d7451dcb8d9637bfbc97413f497069 # v4.4.1
      - name: Login to GitHub Container Registry
        uses: docker/login-action@dbcb813823bdd20940b903addbd779551569679f # v4.6.0
        with:
@@ -78,7 +78,7 @@ jobs:
        with:
          python-version: ${{ env.DEFAULT_PYTHON_VERSION }}
      - name: Install uv
-        uses: astral-sh/setup-uv@bec219d24cd3e171d82865faccec33120bb574f4 # v10.1.0
+        uses: astral-sh/setup-uv@c18668ad3cf93ea998bef934396af7bb5c839dc7 # v10.2.0
        with:
          version: ${{ env.DEFAULT_UV_VERSION }}
          enable-cache: true
@@ -174,13 +174,13 @@ jobs:
        run: cd src-ui && pnpm run test --max-workers=2 --shard=${{ matrix.shard-index }}/${{ matrix.shard-count }}
      - name: Upload test results to Codecov
        if: always()
-        uses: codecov/codecov-action@fb8b3582c8e4def4969c97caa2f19720cb33a72f # v7.0.0
+        uses: codecov/codecov-action@303a32d7a59b442fa8d48b6a1cc6825c09c847a5 # v7.1.1
        with:
          flags: frontend-node-${{ matrix.node-version }}
          directory: src-ui/
          report_type: test_results
      - name: Upload coverage to Codecov
-        uses: codecov/codecov-action@fb8b3582c8e4def4969c97caa2f19720cb33a72f # v7.0.0
+        uses: codecov/codecov-action@303a32d7a59b442fa8d48b6a1cc6825c09c847a5 # v7.1.1
        with:
          flags: frontend-node-${{ matrix.node-version }}
          directory: src-ui/coverage/
@@ -216,7 +216,7 @@ jobs:
        with:
          python-version: '3.12'
      - name: Install uv
-        uses: astral-sh/setup-uv@bec219d24cd3e171d82865faccec33120bb574f4 # v10.1.0
+        uses: astral-sh/setup-uv@c18668ad3cf93ea998bef934396af7bb5c839dc7 # v10.2.0
        with:
          version: '0.12.x'
          enable-cache: false
@@ -231,9 +231,10 @@ jobs:
      - name: Install dependencies
        run: cd src-ui && pnpm install --frozen-lockfile
      - name: Install backend system dependencies
-        run: |
-          apt-get update
-          apt-get install --yes --quiet --no-install-recommends libmagic1
+        timeout-minutes: 12
+        uses: $/.github/actions/apt-install
+        with:
+          packages: libmagic1
      - name: Install backend dependencies
        env:
          PYTHON_VERSION: ${{ steps.setup-python.outputs.python-version }}
@@ -59,7 +59,7 @@ jobs:
        with:
          python-version: ${{ env.DEFAULT_PYTHON_VERSION }}
      - name: Install uv
-        uses: astral-sh/setup-uv@bec219d24cd3e171d82865faccec33120bb574f4 # v10.1.0
+        uses: astral-sh/setup-uv@c18668ad3cf93ea998bef934396af7bb5c839dc7 # v10.2.0
        with:
          version: ${{ env.DEFAULT_UV_VERSION }}
          enable-cache: false
@@ -70,9 +70,10 @@ jobs:
        run: |
          uv sync --python "${PYTHON_VERSION}" --dev --frozen
      - name: Install system dependencies
-        run: |
-          sudo apt-get update -qq
-          sudo apt-get install -qq --no-install-recommends gettext liblept5
+        timeout-minutes: 12
+        uses: $/.github/actions/apt-install
+        with:
+          packages: gettext liblept5
      # ---- Build Documentation ----
      - name: Build documentation
        env:
@@ -212,7 +213,7 @@ jobs:
        with:
          python-version: ${{ env.DEFAULT_PYTHON_VERSION }}
      - name: Install uv
-        uses: astral-sh/setup-uv@bec219d24cd3e171d82865faccec33120bb574f4 # v10.1.0
+        uses: astral-sh/setup-uv@c18668ad3cf93ea998bef934396af7bb5c839dc7 # v10.2.0
        with:
          version: ${{ env.DEFAULT_UV_VERSION }}
          enable-cache: false
@@ -44,7 +44,7 @@ jobs:
      - name: Run Semgrep
        run: semgrep scan --config auto --sarif-output results.sarif
      - name: Upload results to GitHub code scanning
-        uses: github/codeql-action/upload-sarif@b96794f015dfd88f77b49b1c93e0fa7110f94c63 # v4.38.0
+        uses: github/codeql-action/upload-sarif@2892aa5e19bbd11bc0cff5427e3b750a04d9e3c2 # v4.38.2
        if: always()
        with:
          sarif_file: results.sarif
@@ -29,7 +29,7 @@ jobs:
    steps:
      - name: Clean temporary images
        if: "${{ env.TOKEN != '' }}"
-        uses: stumpylog/image-cleaner-action/ephemeral@4fe057d991d63b8f6d5d22c40f17c1bca2226537 # v0.12.0
+        uses: stumpylog/image-cleaner-action/ephemeral@21f875bab2376314e0525c614e923c669434f0e4 # v0.13.0
        with:
          token: "${{ env.TOKEN }}"
          owner: "${{ github.repository_owner }}"
@@ -56,7 +56,7 @@ jobs:
    steps:
      - name: Clean untagged images
        if: "${{ env.TOKEN != '' }}"
-        uses: stumpylog/image-cleaner-action/untagged@4fe057d991d63b8f6d5d22c40f17c1bca2226537 # v0.12.0
+        uses: stumpylog/image-cleaner-action/untagged@21f875bab2376314e0525c614e923c669434f0e4 # v0.13.0
        with:
          token: "${{ env.TOKEN }}"
          owner: "${{ github.repository_owner }}"
@@ -39,7 +39,7 @@ jobs:
          persist-credentials: false
      # Initializes the CodeQL tools for scanning.
      - name: Initialize CodeQL
-        uses: github/codeql-action/init@b96794f015dfd88f77b49b1c93e0fa7110f94c63 # v4.38.0
+        uses: github/codeql-action/init@2892aa5e19bbd11bc0cff5427e3b750a04d9e3c2 # v4.38.2
        with:
          languages: ${{ matrix.language }}
          # If you wish to specify custom queries, you can do so here or in a config file.
@@ -47,4 +47,4 @@ jobs:
          # Prefix the list here with "+" to use these queries and those in the config file.
          # queries: ./path/to/local/query, your-org/your-repo/queries@main
      - name: Perform CodeQL Analysis
-        uses: github/codeql-action/analyze@b96794f015dfd88f77b49b1c93e0fa7110f94c63 # v4.38.0
+        uses: github/codeql-action/analyze@2892aa5e19bbd11bc0cff5427e3b750a04d9e3c2 # v4.38.2
@@ -22,7 +22,7 @@ jobs:
          token: ${{ secrets.PNGX_BOT_PAT }}
          persist-credentials: false
      - name: crowdin action
-        uses: crowdin/github-action@0d5670f539973aea2f01abce61a8989934df0025 # v3.0.2
+        uses: crowdin/github-action@9c23991700c0ec5256fd41089b9d9d7d540e424e # v3.3.0
        with:
          upload_translations: false
          download_translations: true
@@ -25,11 +25,12 @@ jobs:
        id: setup-python
        uses: actions/setup-python@5fda3b95a4ea91299a34e894583c3862153e4b97 # v7.0.0
      - name: Install system dependencies
-        run: |
-          sudo apt-get update -qq
-          sudo apt-get install -qq --no-install-recommends gettext
+        timeout-minutes: 12
+        uses: $/.github/actions/apt-install
+        with:
+          packages: gettext
      - name: Install uv
-        uses: astral-sh/setup-uv@bec219d24cd3e171d82865faccec33120bb574f4 # v10.1.0
+        uses: astral-sh/setup-uv@c18668ad3cf93ea998bef934396af7bb5c839dc7 # v10.2.0
        with:
          version: ${{ env.DEFAULT_UV_VERSION }}
          enable-cache: true
@@ -38,7 +38,7 @@ repos:
          - json
  # See https://github.com/prettier/prettier/issues/15742 for the fork reason
  - repo: https://github.com/rbubley/mirrors-prettier
-    rev: 'v3.9.6'
+    rev: 'v3.9.9'
    hooks:
      - id: prettier
        types_or:
@@ -46,11 +46,11 @@ repos:
          - ts
          - markdown
        additional_dependencies:
-          - prettier@3.9.6
+          - prettier@3.9.9
          - 'prettier-plugin-organize-imports@4.3.0'
  # Python hooks
  - repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: v0.16.7
+    rev: v0.16.9
    hooks:
      - id: ruff-check
      - id: ruff-format
@@ -30,7 +30,7 @@ RUN set -eux \
 # Purpose: Installs s6-overlay and rootfs
 # Comments:
 #  - Don't leave anything extra in here either
-FROM ghcr.io/astral-sh/uv:0.12.16-python3.14-trixie-slim AS s6-overlay-base
+FROM ghcr.io/astral-sh/uv:0.12.23-python3.14-trixie-slim AS s6-overlay-base

 WORKDIR /usr/src/s6

@@ -171,7 +171,9 @@ RUN set -eux \
        && cp /etc/ImageMagick-6/paperless-policy.xml /etc/ImageMagick-6/policy.xml \
      && echo "Cleaning up image layer" \
        && rm --force --verbose *.deb \
-    && rm --recursive --force --verbose /var/lib/apt/lists/*
+    && rm --recursive --force --verbose /var/lib/apt/lists/* \
+    && echo "Configuring interactive shells to source the s6 container environment" \
+      && echo '. /etc/profile.d/contenv.sh' >> /etc/bash.bashrc

 WORKDIR /usr/src/paperless/src/

@@ -24,7 +24,7 @@ services:
    network_mode: host
    restart: unless-stopped
  greenmail:
-    image: docker.io/greenmail/standalone:2.1.13
+    image: docker.io/greenmail/standalone:2.1.14
    hostname: greenmail
    container_name: greenmail
    environment:
@@ -0,0 +1,18 @@
+#!/bin/sh
+# Source s6 container environment for interactive shells.
+# Ensures variables resolved from *_FILE secret injection are visible
+# when using 'docker exec bash'. Does not affect s6 services (those
+# use with-contenv directly). Has no effect in non-container contexts
+# because the directory will not exist.
+# Note: sh/dash shells opened via 'docker exec sh' are not covered;
+# only bash-based sessions benefit from this file.
+_pngx_contenv="/run/s6/container_environment"
+if [ -d "${_pngx_contenv}" ]; then
+	for _pngx_f in "${_pngx_contenv}"/*; do
+		[ -f "${_pngx_f}" ] || continue
+		_pngx_name=$(basename "${_pngx_f}")
+		_pngx_val=$(cat "${_pngx_f}")
+		export "${_pngx_name}=${_pngx_val}"
+	done
+fi
+unset _pngx_contenv _pngx_f _pngx_name _pngx_val
@@ -136,13 +136,15 @@ for suggested generation and embedding models.
 ### AI-assisted suggestions

 With AI enabled, Paperless-ngx can suggest a title, tags, correspondent, document type,
-storage path and dates by sending the document to the LLM. This is **opt-in per request**
-and surfaces through the "Suggest" control on the document detail page, alongside the
-classic classifier-based suggestions — it does not disable them. Suggestions are requested
-automatically when you open a document that carries an inbox tag unless "Automatically request
-suggestions for inbox documents" under Settings > Documents is disabled. Suggestion output
-language can be steered with
-[`PAPERLESS_AI_LLM_OUTPUT_LANGUAGE`](configuration.md#PAPERLESS_AI_LLM_OUTPUT_LANGUAGE)
+storage path and dates by sending the document to the LLM using "Suggest" button on the document
+detail page. You can choose which type of suggestions are requested by default under Settings >
+Documents, either ML (classifier-based) suggestions, AI suggestions, or both. When both are requested
+the results are combined.
+
+Suggestions are requested automatically when you open a document that carries an inbox tag
+unless "Automatically request suggestions for inbox documents" under Settings > Documents is disabled.
+
+Suggestion output language can be steered with [`PAPERLESS_AI_LLM_OUTPUT_LANGUAGE`](configuration.md#PAPERLESS_AI_LLM_OUTPUT_LANGUAGE)
 (otherwise it follows the user's UI language).

 ### The LLM index (RAG) and similar documents
@@ -19,7 +19,6 @@ classifiers = [
 dependencies = [
  "azure-ai-documentintelligence>=1.0.2",
  "babel>=2.17",
-  "bleach~=6.4.0",
  "celery[redis]~=5.6.2",
  "channels~=4.2",
  "channels-redis~=4.2",
@@ -35,7 +34,7 @@ dependencies = [
  "django-cors-headers~=4.9.0",
  "django-extensions~=4.1",
  "django-filter~=25.1",
-  "django-guardian>=3.3.3,<3.5",
+  "django-guardian>=3.3.3,<3.6",
  "django-multiselectfield~=1.0.1",
  "django-rich~=2.2.0",
  "django-soft-delete~=1.0.18",
@@ -45,7 +44,7 @@ dependencies = [
  "drf-spectacular-sidecar>=2026.7.1,<2026.10",
  "drf-writable-nested~=0.7.1",
  "filelock~=3.32.0",
-  "flower>=2.0.1,<2.2",
+  "flower>=2.0.1,<2.3",
  "gotenberg-client[httpx]~=1.0",
  "httpx-oauth~=0.17",
  "ijson>=3.5.1",
@@ -77,6 +76,7 @@ dependencies = [
  "tantivy~=0.26.0",
  "tika-client[httpx]~=1.0",
  "torch>=2.13,<2.15",
+  "turbohtml~=1.10.0",
  "watchfiles>=1.2",
  "whitenoise~=6.11",
  "whoosh-compat[tantivy]==0.3",
@@ -84,13 +84,13 @@ dependencies = [
 ]
 [project.optional-dependencies]
 mariadb = [
-  "mysqlclient~=2.2.7",
+  "mysqlclient>=2.2.7,<2.4",
 ]
 postgres = [
  "psycopg[c,pool]==3.3.4",
  # Direct dependency for proper resolution of the pre-built wheels
  "psycopg-c==3.3.4",
-  "psycopg-pool==3.3.1",
+  "psycopg-pool==3.3.2",
 ]
 webserver = [
  "granian[uvloop]>=2.7,<2.9",
@@ -112,7 +112,7 @@ lint = [
 testing = [
  "daphne",
  "factory-boy~=3.3.1",
-  "faker>=40.36,<40.39",
+  "faker>=40.36,<40.40",
  "imagehash",
  "pytest~=9.1.1",
  "pytest-cov~=7.1.0",
@@ -136,7 +136,6 @@ typing = [
  "mypy",
  "mypy-baseline",
  "pyrefly",
-  "types-bleach",
  "types-channels",
  "types-colorama",
  "types-dateparser",
@@ -151,6 +150,7 @@ typing = [

 [tool.uv]
 required-version = ">=0.9.0"
+prerelease = "disallow"
 environments = [
  "sys_platform == 'darwin'",
  "sys_platform == 'linux'",
@@ -91,6 +91,11 @@
                "glob": "**/*",
                "input": "node_modules/pdfjs-dist/iccs/",
                "output": "/assets/iccs/"
+              },
+              {
+                "glob": "**/*",
+                "input": "node_modules/pdfjs-dist/cmaps/",
+                "output": "/assets/cmaps/"
              }
            ],
            "styles": [
@@ -15,16 +15,16 @@
  },
  "private": true,
  "dependencies": {
-    "@angular/cdk": "^22.1.6",
-    "@angular/common": "~22.1.6",
-    "@angular/compiler": "~22.1.6",
-    "@angular/core": "~22.1.6",
-    "@angular/forms": "~22.1.6",
-    "@angular/localize": "~22.1.6",
-    "@angular/platform-browser": "~22.1.6",
-    "@angular/router": "~22.1.6",
+    "@angular/cdk": "^22.2.0",
+    "@angular/common": "~22.2.0",
+    "@angular/compiler": "~22.2.0",
+    "@angular/core": "~22.2.0",
+    "@angular/forms": "~22.2.0",
+    "@angular/localize": "~22.2.0",
+    "@angular/platform-browser": "~22.2.0",
+    "@angular/router": "~22.2.0",
    "@ng-bootstrap/ng-bootstrap": "^21.0.0",
-    "@ng-select/ng-select": "~24.1.1",
+    "@ng-select/ng-select": "~24.1.2",
    "@ngneat/dirty-check-forms": "^3.0.3",
    "@popperjs/core": "^2.11.8",
    "bootstrap": "^5.3.8",
@@ -45,29 +45,29 @@
  },
  "devDependencies": {
    "@angular-builders/jest": "^22.0.1",
-    "@angular-devkit/core": "^22.1.8",
-    "@angular-devkit/schematics": "^22.1.8",
+    "@angular-devkit/core": "^22.2.0",
+    "@angular-devkit/schematics": "^22.2.0",
    "@angular-eslint/builder": "22.5.0",
    "@angular-eslint/eslint-plugin": "22.5.0",
    "@angular-eslint/eslint-plugin-template": "22.5.0",
    "@angular-eslint/schematics": "22.5.0",
    "@angular-eslint/template-parser": "22.5.0",
-    "@angular/build": "22.1.8",
-    "@angular/cli": "22.1.8",
-    "@angular/compiler-cli": "~22.1.6",
+    "@angular/build": "22.2.0",
+    "@angular/cli": "22.2.0",
+    "@angular/compiler-cli": "~22.2.0",
    "@playwright/test": "^1.62.1",
    "@types/jest": "^30.0.0",
-    "@types/node": "^26.5.0",
-    "@typescript-eslint/eslint-plugin": "^8.70.0",
-    "@typescript-eslint/parser": "^8.70.0",
-    "@typescript-eslint/utils": "^8.70.0",
-    "eslint": "^10.10.0",
-    "jest": "30.5.1",
-    "jest-environment-jsdom": "^30.5.1",
+    "@types/node": "^26.6.2",
+    "@typescript-eslint/eslint-plugin": "^8.70.1",
+    "@typescript-eslint/parser": "^8.70.1",
+    "@typescript-eslint/utils": "^8.70.1",
+    "eslint": "^10.11.0",
+    "jest": "30.5.2",
+    "jest-environment-jsdom": "^30.5.2",
    "jest-junit": "^17.0.0",
    "jest-preset-angular": "^17.0.0",
    "jest-websocket-mock": "^2.5.0",
-    "prettier": "^3.9.6",
+    "prettier": "^3.9.9",
    "prettier-plugin-organize-imports": "^4.3.0",
    "ts-node": "~10.9.2",
    "typescript": "^6.0.3"
@@ -253,6 +253,24 @@
              </div>
            </div>

+            @if (aiEnabled) {
+              <div class="row mb-3">
+                <div class="col-md-3 col-form-label pt-0">
+                  <span i18n>Suggestions default to</span>
+                </div>
+                <div class="col">
+                  <fieldset class="btn-group btn-group-sm">
+                    <input type="radio" class="btn-check" id="suggestionSourceBoth" [value]="SuggestionSource.Both" formControlName="documentEditingSuggestionSource">
+                    <label class="btn btn-outline-primary" for="suggestionSourceBoth"><ng-container i18n>Both</ng-container></label>
+                    <input type="radio" class="btn-check" id="suggestionSourceML" [value]="SuggestionSource.ML" formControlName="documentEditingSuggestionSource">
+                    <label class="btn btn-outline-primary" for="suggestionSourceML"><i-bs class="me-1" name="cpu"></i-bs><ng-container i18n>ML only</ng-container></label>
+                    <input type="radio" class="btn-check" id="suggestionSourceAI" [value]="SuggestionSource.AI" formControlName="documentEditingSuggestionSource">
+                    <label class="btn btn-outline-primary" for="suggestionSourceAI"><i-bs class="me-1" name="stars"></i-bs><ng-container i18n>AI only</ng-container></label>
+                  </fieldset>
+                </div>
+              </div>
+            }
+
            <div class="row">
              <div class="col">
                <pngx-input-check i18n-title title="Automatically request suggestions for inbox documents" i18n-hint hint="If un-checked, suggestions must be requested via the Suggest button." formControlName="documentEditingAutoSuggest"></pngx-input-check>
@@ -307,7 +307,7 @@ describe('SettingsComponent', () => {
    expect(toastErrorSpy).toHaveBeenCalled()
    expect(storeSpy).toHaveBeenCalled()
    expect(appearanceSettingsSpy).not.toHaveBeenCalled()
-    expect(setSpy).toHaveBeenCalledTimes(34)
+    expect(setSpy).toHaveBeenCalledTimes(35)
    expect(setSpy).toHaveBeenCalledWith(SETTINGS_KEYS.SIDEBAR_HIDDEN_ITEMS, [
      HideableSidebarItemID.Workflows,
    ])
@@ -44,6 +44,7 @@ import {
  HIDEABLE_SIDEBAR_ITEM_IDS,
  HideableSidebarItemID,
  SETTINGS_KEYS,
+  SuggestionSource,
 } from 'src/app/data/ui-settings'
 import { User } from 'src/app/data/user'
 import { IfPermissionsDirective } from 'src/app/directives/if-permissions.directive'
@@ -184,6 +185,7 @@ export class SettingsComponent
    documentEditingRemoveInboxTags: new FormControl(null),
    documentEditingOverlayThumbnail: new FormControl(null),
    documentEditingAutoSuggest: new FormControl(null),
+    documentEditingSuggestionSource: new FormControl(null),
    documentDetailsHiddenFields: new FormControl([]),
    searchDbOnly: new FormControl(null),
    searchLink: new FormControl(null),
@@ -217,6 +219,11 @@ export class SettingsComponent
  public readonly PdfZoomScale = PdfZoomScale

  public readonly PdfEditorEditMode = PdfEditorEditMode
+  public readonly SuggestionSource = SuggestionSource
+
+  get aiEnabled(): boolean {
+    return this.settings.get(SETTINGS_KEYS.AI_ENABLED)
+  }

  public readonly documentDetailFieldOptions = documentDetailFieldOptions
  public readonly sidebarItemOptions = HIDEABLE_SIDEBAR_ITEM_IDS.map((id) => ({
@@ -404,6 +411,9 @@ export class SettingsComponent
      documentEditingAutoSuggest: this.settings.get(
        SETTINGS_KEYS.DOCUMENT_EDITING_AUTO_SUGGEST
      ),
+      documentEditingSuggestionSource: this.settings.get(
+        SETTINGS_KEYS.DOCUMENT_EDITING_SUGGESTION_SOURCE
+      ),
      documentDetailsHiddenFields: this.settings.get(
        SETTINGS_KEYS.DOCUMENT_DETAILS_HIDDEN_FIELDS
      ),
@@ -625,6 +635,10 @@ export class SettingsComponent
      SETTINGS_KEYS.DOCUMENT_EDITING_AUTO_SUGGEST,
      this.settingsForm.value.documentEditingAutoSuggest
    )
+    this.settings.set(
+      SETTINGS_KEYS.DOCUMENT_EDITING_SUGGESTION_SOURCE,
+      this.settingsForm.value.documentEditingSuggestionSource
+    )
    this.settings.set(
      SETTINGS_KEYS.DOCUMENT_DETAILS_HIDDEN_FIELDS,
      this.settingsForm.value.documentDetailsHiddenFields
@@ -154,11 +154,29 @@
  & section {
    position: absolute;
    text-align: initial;
+    pointer-events: auto;
    box-sizing: border-box;
    transform-origin: 0 0;
  }

+  & :is(.linkAnnotation, .buttonWidgetAnnotation.pushButton) > a {
+    position: absolute;
+    inset: 0;
+    font-size: 1em;
+    transition: none;
+  }
+
+  & :is(.linkAnnotation, .buttonWidgetAnnotation.pushButton):not(.hasBorder)
+    > a:hover {
+    opacity: 0.2;
+    background-color: rgb(255 255 0);
+  }
+
  & .annotationTextContent {
    opacity: 0;
  }
 }
+
+:host ::ng-deep .textLayer.selecting ~ .annotationLayer section {
+  pointer-events: none;
+}
@@ -1,7 +1,11 @@
 import { SimpleChange } from '@angular/core'
 import { ComponentFixture, TestBed } from '@angular/core/testing'
 import * as pdfjs from 'pdfjs-dist/legacy/build/pdf.mjs'
-import { PDFSinglePageViewer, PDFViewer } from 'pdfjs-dist/web/pdf_viewer.mjs'
+import {
+  LinkTarget,
+  PDFSinglePageViewer,
+  PDFViewer,
+} from 'pdfjs-dist/web/pdf_viewer.mjs'
 import { PngxPdfViewerComponent } from './pdf-viewer.component'
 import { PdfRenderMode, PdfZoomLevel, PdfZoomScale } from './pdf-viewer.types'

@@ -58,6 +62,16 @@ describe('PngxPdfViewerComponent', () => {
    expect((component as any).pdfViewer).toBeInstanceOf(PDFViewer)
  })

+  it('opens external links in a new tab', () => {
+    const linkService = (component as any).linkService
+    expect(linkService.options).toEqual(
+      expect.objectContaining({
+        externalLinkTarget: LinkTarget.BLANK,
+        externalLinkRel: 'noopener noreferrer nofollow',
+      })
+    )
+  })
+
  it('resolves the worker source relative to the document base URI', async () => {
    setBaseHref('/paperless/')
    const getDocumentSpy = jest.spyOn(pdfjs, 'getDocument')
@@ -76,6 +90,7 @@ describe('PngxPdfViewerComponent', () => {
      withCredentials: true,
      wasmUrl: expect.stringContaining('/paperless/assets/wasm/'),
      iccUrl: expect.stringContaining('/paperless/assets/iccs/'),
+      cMapUrl: expect.stringContaining('/paperless/assets/cmaps/'),
    })
  })

@@ -21,6 +21,7 @@ import {
 } from 'pdfjs-dist/legacy/build/pdf.mjs'
 import {
  EventBus,
+  LinkTarget,
  PDFFindController,
  PDFLinkService,
  PDFSinglePageViewer,
@@ -75,7 +76,11 @@ export class PngxPdfViewerComponent
  private lastViewerPage?: number

  private readonly eventBus = new EventBus()
-  private readonly linkService = new PDFLinkService({ eventBus: this.eventBus })
+  private readonly linkService = new PDFLinkService({
+    eventBus: this.eventBus,
+    externalLinkTarget: LinkTarget.BLANK,
+    externalLinkRel: 'noopener noreferrer nofollow',
+  })
  private readonly findController = new PDFFindController({
    eventBus: this.eventBus,
    linkService: this.linkService,
@@ -185,6 +190,7 @@ export class PngxPdfViewerComponent
      withCredentials: true,
      wasmUrl: new URL('assets/wasm/', this.document.baseURI).toString(),
      iccUrl: new URL('assets/iccs/', this.document.baseURI).toString(),
+      cMapUrl: new URL('assets/cmaps/', this.document.baseURI).toString(),
    }
    this.loadingTask = getDocument(initOptions)
    try {
@@ -1,58 +1,84 @@
-<div class="btn-group">
-  <button type="button" class="btn btn-sm btn-outline-primary" (click)="clickSuggest()" [disabled]="disabled() || loading() || (suggestions() && !aiEnabled())" [aria-label]="noSuggestions ? 'No suggestions' : 'Suggest'" i18n-aria-label>
-    @if (loading()) {
-      <div class="spinner-border spinner-border-sm" role="status"></div>
-    } @else if (noSuggestions) {
-      <i-bs width="1.2em" height="1.2em" name="check-circle"></i-bs>
-    } @else {
-      <i-bs width="1.2em" height="1.2em" name="stars"></i-bs>
+<div class="d-flex align-items-center">
+  <div class="btn-group">
+    <button type="button" class="btn btn-sm btn-outline-primary" (click)="clickSuggest()" [disabled]="disabled() || loading() || (suggestions() && !aiEnabled())" [aria-label]="noSuggestions ? 'No suggestions' : 'Suggest'" i18n-aria-label>
+      @if (loading()) {
+        <div class="spinner-border spinner-border-sm" role="status"></div>
+      } @else if (noSuggestions) {
+        <i-bs width="1.2em" height="1.2em" name="check-circle"></i-bs>
+      } @else {
+        <i-bs width="1.2em" height="1.2em" name="lightbulb"></i-bs>
+      }
+      @if (noSuggestions) {
+        <span class="d-none d-lg-inline ps-1" i18n>No suggestions</span>
+      } @else {
+        <span class="d-none d-lg-inline ps-1" i18n>Suggest</span>
+      }
+      @if (totalSuggestions > 0) {
+        <span class="badge bg-primary ms-2">{{ totalSuggestions }}</span>
+      }
+    </button>
+
+    @if (aiEnabled()) {
+      <div class="btn-group" ngbDropdown #dropdown="ngbDropdown" [popperOptions]="popperOptions">
+        <button type="button" class="btn btn-sm btn-outline-primary" ngbDropdownToggle [disabled]="disabled() || loading() || !suggestions()" aria-expanded="false" aria-controls="suggestionsDropdown" aria-label="Suggestions dropdown">
+          <span class="visually-hidden" i18n>Show suggestions</span>
+        </button>
+
+        <div ngbDropdownMenu aria-labelledby="suggestionsDropdown" class="shadow suggestions-dropdown">
+          <div class="list-group list-group-flush small pb-0">
+            @if (novelSuggestions === 0 && reusableSuggestions === 0) {
+              <div class="list-group-item text-muted fst-italic">
+                <small class="text-muted small fst-italic" i18n>No novel suggestions</small>
+              </div>
+            }
+            @if (suggestions()?.suggested_tags?.length > 0) {
+              <small class="list-group-item text-uppercase text-muted small"><i-bs class="me-2" name="tags"></i-bs><ng-container i18n>Tags</ng-container></small>
+              @for (tag of suggestions().suggested_tags; track tag) {
+                <button type="button" class="list-group-item list-group-item-action bg-light" (click)="addTag.emit(tag)">{{ tag }}</button>
+              }
+            }
+            @if (suggestions()?.suggested_document_types?.length > 0) {
+              <div class="list-group-item text-uppercase text-muted small"><i-bs class="me-2" name="hash"></i-bs><ng-container i18n>Document Types</ng-container></div>
+              @for (type of suggestions().suggested_document_types; track type) {
+                <button type="button" class="list-group-item list-group-item-action bg-light" (click)="addDocumentType.emit(type)">{{ type }}</button>
+              }
+            }
+            @if (suggestions()?.suggested_correspondents?.length > 0) {
+              <div class="list-group-item text-uppercase text-muted small"><i-bs class="me-2" name="person"></i-bs><ng-container i18n>Correspondents</ng-container></div>
+              @for (correspondent of suggestions().suggested_correspondents; track correspondent) {
+                <button type="button" class="list-group-item list-group-item-action bg-light" (click)="addCorrespondent.emit(correspondent)">{{ correspondent }}</button>
+              }
+            }
+            @if (reusableSuggestions > 0) {
+              <div class="list-group-item text-muted fst-italic">
+                <small class="text-muted small fst-italic" i18n>{reusableSuggestions, plural, =1 {1 existing value suggested below} other {{{reusableSuggestions}} existing values suggested below}}</small>
+              </div>
+            }
+          </div>
+        </div>
+      </div>
    }
-    @if (noSuggestions) {
-      <span class="d-none d-lg-inline ps-1" i18n>No suggestions</span>
-    } @else {
-      <span class="d-none d-lg-inline ps-1" i18n>Suggest</span>
-    }
-    @if (totalSuggestions > 0) {
-      <span class="badge bg-primary ms-2">{{ totalSuggestions }}</span>
-    }
-  </button>
+  </div>

  @if (aiEnabled()) {
-    <div class="btn-group" ngbDropdown #dropdown="ngbDropdown" [popperOptions]="popperOptions">
-      <button type="button" class="btn btn-sm btn-outline-primary" ngbDropdownToggle [disabled]="disabled() || loading() || !suggestions()" aria-expanded="false" aria-controls="suggestionsDropdown" aria-label="Suggestions dropdown">
-        <span class="visually-hidden" i18n>Show suggestions</span>
+    <div ngbDropdown autoClose="outside" placement="bottom-end" [popperOptions]="popperOptions">
+      <button type="button" class="btn btn-sm btn-link position-relative" ngbDropdownToggle [disabled]="disabled() || loading()" i18n-title title="Suggestion options">
+        <i-bs name="three-dots"></i-bs>
+        @if (source() !== defaultSource()) {
+          <span class="position-absolute top-0 start-100 translate-middle p-1 bg-primary border border-light rounded-circle">
+            <span class="visually-hidden" i18n>Not using default</span>
+          </span>
+        }
      </button>
-
-      <div ngbDropdownMenu aria-labelledby="suggestionsDropdown" class="shadow suggestions-dropdown">
-        <div class="list-group list-group-flush small pb-0">
-          @if (novelSuggestions === 0 && reusableSuggestions === 0) {
-            <div class="list-group-item text-muted fst-italic">
-              <small class="text-muted small fst-italic" i18n>No novel suggestions</small>
-            </div>
-          }
-          @if (suggestions()?.suggested_tags?.length > 0) {
-            <small class="list-group-item text-uppercase text-muted small"><i-bs class="me-2" name="tags"></i-bs><ng-container i18n>Tags</ng-container></small>
-            @for (tag of suggestions().suggested_tags; track tag) {
-              <button type="button" class="list-group-item list-group-item-action bg-light" (click)="addTag.emit(tag)">{{ tag }}</button>
-            }
-          }
-          @if (suggestions()?.suggested_document_types?.length > 0) {
-            <div class="list-group-item text-uppercase text-muted small"><i-bs class="me-2" name="hash"></i-bs><ng-container i18n>Document Types</ng-container></div>
-            @for (type of suggestions().suggested_document_types; track type) {
-              <button type="button" class="list-group-item list-group-item-action bg-light" (click)="addDocumentType.emit(type)">{{ type }}</button>
-            }
-          }
-          @if (suggestions()?.suggested_correspondents?.length > 0) {
-            <div class="list-group-item text-uppercase text-muted small"><i-bs class="me-2" name="person"></i-bs><ng-container i18n>Correspondents</ng-container></div>
-            @for (correspondent of suggestions().suggested_correspondents; track correspondent) {
-              <button type="button" class="list-group-item list-group-item-action bg-light" (click)="addCorrespondent.emit(correspondent)">{{ correspondent }}</button>
-            }
-          }
-          @if (reusableSuggestions > 0) {
-            <div class="list-group-item text-muted fst-italic">
-              <small class="text-muted small fst-italic" i18n>{reusableSuggestions, plural, =1 {1 existing value suggested below} other {{{reusableSuggestions}} existing values suggested below}}</small>
-            </div>
-          }
+      <div ngbDropdownMenu class="shadow p-3">
+        <div class="small text-muted mb-2" i18n>Suggest using:</div>
+        <div class="form-check small">
+          <input class="form-check-input" type="checkbox" id="suggestionSourceML" [checked]="useML" [disabled]="useML && !useAI" (change)="setSources($event.target.checked, useAI)">
+          <label class="form-check-label d-inline-flex align-items-center gap-1" for="suggestionSourceML"><i-bs name="cpu"></i-bs><ng-container i18n>ML</ng-container></label>
+        </div>
+        <div class="form-check small">
+          <input class="form-check-input" type="checkbox" id="suggestionSourceAI" [checked]="useAI" [disabled]="useAI && !useML" (change)="setSources(useML, $event.target.checked)">
+          <label class="form-check-label d-inline-flex align-items-center gap-1" for="suggestionSourceAI"><i-bs name="stars"></i-bs><ng-container i18n>AI</ng-container></label>
        </div>
      </div>
    </div>
@@ -1,3 +1,7 @@
 .suggestions-dropdown {
  min-width: 250px;
 }
+
+.btn-link.dropdown-toggle::after {
+  display: none;
+}
@@ -1,6 +1,7 @@
 import { ComponentFixture, TestBed } from '@angular/core/testing'
 import { NgbDropdownModule } from '@ng-bootstrap/ng-bootstrap'
 import { NgxBootstrapIconsModule, allIcons } from 'ngx-bootstrap-icons'
+import { SuggestionSource } from 'src/app/data/ui-settings'
 import { SuggestionsDropdownComponent } from './suggestions-dropdown.component'

 describe('SuggestionsDropdownComponent', () => {
@@ -179,14 +180,71 @@ describe('SuggestionsDropdownComponent', () => {

  it('should toggle dropdown when clickSuggest is called and suggestions are not null', () => {
    fixture.componentRef.setInput('aiEnabled', true)
+    fixture.componentRef.setInput('fetchedSources', [SuggestionSource.ML])
    fixture.detectChanges()
    fixture.componentRef.setInput('suggestions', {
      suggested_correspondents: [],
      suggested_tags: [],
      suggested_document_types: [],
    })
+    fixture.detectChanges()
    component.clickSuggest()
-    expect(component.dropdown.open).toBeTruthy()
+    expect(component.dropdown.isOpen()).toBeTruthy()
    expect(fixture.nativeElement.textContent).toContain('No novel suggestions')
  })
+
+  it('should fetch unfetched sources and show existing suggestions', () => {
+    jest.spyOn(component.getSuggestions, 'emit')
+    fixture.componentRef.setInput('aiEnabled', true)
+    fixture.componentRef.setInput('source', SuggestionSource.Both)
+    fixture.componentRef.setInput('fetchedSources', [SuggestionSource.ML])
+    fixture.componentRef.setInput('suggestions', { tags: [1] })
+    fixture.detectChanges()
+    component.clickSuggest()
+    expect(component.getSuggestions.emit).toHaveBeenCalledWith(
+      SuggestionSource.Both
+    )
+    expect(component.dropdown.isOpen()).toBeTruthy()
+  })
+
+  it('should only show source options when AI is enabled', () => {
+    expect(
+      fixture.nativeElement.querySelector('#suggestionSourceML')
+    ).toBeNull()
+    fixture.componentRef.setInput('aiEnabled', true)
+    fixture.detectChanges()
+    fixture.nativeElement
+      .querySelector('button[title="Suggestion options"]')
+      .click()
+    fixture.detectChanges()
+    expect(
+      fixture.nativeElement.querySelector('#suggestionSourceML')
+    ).not.toBeNull()
+  })
+
+  it('should emit source changes and never allow no source', () => {
+    const emitSpy = jest.spyOn(component.sourceChange, 'emit')
+    component.setSources(true, true)
+    expect(emitSpy).toHaveBeenCalledWith(SuggestionSource.Both)
+    component.setSources(true, false)
+    expect(emitSpy).toHaveBeenCalledWith(SuggestionSource.ML)
+    component.setSources(false, true)
+    expect(emitSpy).toHaveBeenCalledWith(SuggestionSource.AI)
+
+    emitSpy.mockClear()
+    component.setSources(false, false)
+    expect(emitSpy).not.toHaveBeenCalled()
+  })
+
+  it('should indicate a non-default source', () => {
+    fixture.componentRef.setInput('aiEnabled', true)
+    fixture.componentRef.setInput('source', SuggestionSource.AI)
+    fixture.componentRef.setInput('defaultSource', SuggestionSource.AI)
+    fixture.detectChanges()
+    expect(fixture.nativeElement.textContent).not.toContain('Not using default')
+
+    fixture.componentRef.setInput('source', SuggestionSource.Both)
+    fixture.detectChanges()
+    expect(fixture.nativeElement.textContent).toContain('Not using default')
+  })
 })
@@ -8,6 +8,7 @@ import {
 import { NgbDropdown, NgbDropdownModule } from '@ng-bootstrap/ng-bootstrap'
 import { NgxBootstrapIconsModule } from 'ngx-bootstrap-icons'
 import { DocumentSuggestions } from 'src/app/data/document-suggestions'
+import { SuggestionSource } from 'src/app/data/ui-settings'
 import { pngxPopperOptions } from 'src/app/utils/popper-options'

@Component({
@@ -18,12 +19,16 @@ import { pngxPopperOptions } from 'src/app/utils/popper-options'
 })
 export class SuggestionsDropdownComponent {
  public popperOptions = pngxPopperOptions
+  public readonly SuggestionSource = SuggestionSource

  @ViewChild('dropdown') dropdown: NgbDropdown
  readonly suggestions = input<DocumentSuggestions>(null)
  readonly aiEnabled = input(false)
  readonly loading = input(false)
  readonly disabled = input(false)
+  readonly source = input<SuggestionSource>(SuggestionSource.ML)
+  readonly defaultSource = input<SuggestionSource>(SuggestionSource.ML)
+  readonly fetchedSources = input<SuggestionSource[]>([])

  readonly appliedTags = input<number[]>([])
  readonly appliedCorrespondent = input<number>(null)
@@ -31,8 +36,10 @@ export class SuggestionsDropdownComponent {
  readonly appliedStoragePath = input<number>(null)

  @Output()
-  getSuggestions: EventEmitter<SuggestionsDropdownComponent> =
-    new EventEmitter()
+  getSuggestions: EventEmitter<SuggestionSource> = new EventEmitter()
+
+  @Output()
+  sourceChange: EventEmitter<SuggestionSource> = new EventEmitter()

  @Output()
  addTag: EventEmitter<string> = new EventEmitter()
@@ -53,12 +60,42 @@ export class SuggestionsDropdownComponent {
    }

    if (!this.suggestions()) {
-      this.getSuggestions.emit(this)
+      this.getSuggestions.emit(this.source())
+    } else if (this.hasUnfetchedSources) {
+      // sources changed, fetch the rest and show what we have meanwhile
+      this.getSuggestions.emit(this.source())
+      this.dropdown?.open()
    } else {
      this.dropdown?.toggle()
    }
  }

+  get useML(): boolean {
+    return this.source() !== SuggestionSource.AI
+  }
+
+  get useAI(): boolean {
+    return this.source() !== SuggestionSource.ML
+  }
+
+  get hasUnfetchedSources(): boolean {
+    const fetched = this.fetchedSources()
+    return (
+      (this.useML && !fetched.includes(SuggestionSource.ML)) ||
+      (this.useAI && !fetched.includes(SuggestionSource.AI))
+    )
+  }
+
+  public setSources(ml: boolean, ai: boolean) {
+    if (ml && ai) {
+      this.sourceChange.emit(SuggestionSource.Both)
+    } else if (ml) {
+      this.sourceChange.emit(SuggestionSource.ML)
+    } else if (ai) {
+      this.sourceChange.emit(SuggestionSource.AI)
+    }
+  }
+
  get novelSuggestions(): number {
    return (
      (this.suggestions()?.suggested_correspondents?.length ?? 0) +
@@ -134,11 +134,15 @@
                [loading]="suggestionsLoading()"
                [suggestions]="suggestions()"
                [aiEnabled]="aiEnabled"
+                [source]="suggestionSource"
+                [defaultSource]="defaultSuggestionSource"
+                [fetchedSources]="fetchedSuggestionSources()"
                [appliedTags]="documentForm.value.tags"
                [appliedCorrespondent]="documentForm.value.correspondent"
                [appliedDocumentType]="documentForm.value.document_type"
                [appliedStoragePath]="documentForm.value.storage_path"
-                (getSuggestions)="getSuggestions()"
+                (getSuggestions)="getSuggestions($event)"
+                (sourceChange)="suggestionSourceOverride.set($event)"
                (addTag)="createTag($event)"
                (addDocumentType)="createDocumentType($event)"
                (addCorrespondent)="createCorrespondent($event)">
@@ -43,7 +43,7 @@ import {
 } from 'src/app/data/filter-rule-type'
 import { StoragePath } from 'src/app/data/storage-path'
 import { Tag } from 'src/app/data/tag'
-import { SETTINGS_KEYS } from 'src/app/data/ui-settings'
+import { SETTINGS_KEYS, SuggestionSource } from 'src/app/data/ui-settings'
 import { PermissionsGuard } from 'src/app/guards/permissions.guard'
 import { CustomDatePipe } from 'src/app/pipes/custom-date.pipe'
 import { DocumentTitlePipe } from 'src/app/pipes/document-title.pipe'
@@ -1528,6 +1528,113 @@ describe('DocumentDetailComponent', () => {
    expect(component.suggestionsLoading()).toBeFalsy()
  })

+  it('should get and merge ML and AI suggestions when source is both', () => {
+    settingsService.set(
+      SETTINGS_KEYS.DOCUMENT_EDITING_SUGGESTION_SOURCE,
+      SuggestionSource.Both
+    )
+    const getSetting = settingsService.get.bind(settingsService)
+    jest
+      .spyOn(settingsService, 'get')
+      .mockImplementation((key) =>
+        key === SETTINGS_KEYS.AI_ENABLED ? true : getSetting(key)
+      )
+    const suggestionsSpy = jest
+      .spyOn(documentService, 'getSuggestions')
+      .mockReturnValue(of({ tags: [42], dates: ['2024-01-01'] }))
+    const aiSuggestionsSpy = jest
+      .spyOn(documentService, 'getAiSuggestions')
+      .mockReturnValue(
+        of({ title: 'AI title', tags: [42, 43], suggested_tags: ['New'] })
+      )
+    initNormally()
+    expect(suggestionsSpy).toHaveBeenCalled()
+    expect(aiSuggestionsSpy).toHaveBeenCalled()
+    expect(component.suggestions().title).toEqual('AI title')
+    expect(component.suggestions().tags).toEqual([42, 43])
+    expect(component.suggestions().suggested_tags).toEqual(['New'])
+    expect(component.suggestions().dates).toEqual(['2024-01-01'])
+  })
+
+  it('should only fetch sources not yet fetched for the document', () => {
+    settingsService.set(SETTINGS_KEYS.DOCUMENT_EDITING_AUTO_SUGGEST, false)
+    settingsService.set(
+      SETTINGS_KEYS.DOCUMENT_EDITING_SUGGESTION_SOURCE,
+      SuggestionSource.ML
+    )
+    const getSetting = settingsService.get.bind(settingsService)
+    jest
+      .spyOn(settingsService, 'get')
+      .mockImplementation((key) =>
+        key === SETTINGS_KEYS.AI_ENABLED ? true : getSetting(key)
+      )
+    const suggestionsSpy = jest
+      .spyOn(documentService, 'getSuggestions')
+      .mockReturnValue(of({ tags: [42] }))
+    const aiSuggestionsSpy = jest
+      .spyOn(documentService, 'getAiSuggestions')
+      .mockReturnValue(of({ tags: [43] }))
+    initNormally()
+
+    component.getSuggestions()
+    expect(suggestionsSpy).toHaveBeenCalledTimes(1)
+    expect(aiSuggestionsSpy).not.toHaveBeenCalled()
+
+    component.getSuggestions(SuggestionSource.Both)
+    expect(suggestionsSpy).toHaveBeenCalledTimes(1)
+    expect(aiSuggestionsSpy).toHaveBeenCalledTimes(1)
+    expect(component.suggestions().tags).toEqual([42, 43])
+
+    component.getSuggestions(SuggestionSource.Both)
+    expect(suggestionsSpy).toHaveBeenCalledTimes(1)
+    expect(aiSuggestionsSpy).toHaveBeenCalledTimes(1)
+  })
+
+  it('should use the per-document source override and reset it on document change', () => {
+    settingsService.set(SETTINGS_KEYS.DOCUMENT_EDITING_AUTO_SUGGEST, false)
+    const getSetting = settingsService.get.bind(settingsService)
+    jest
+      .spyOn(settingsService, 'get')
+      .mockImplementation((key) =>
+        key === SETTINGS_KEYS.AI_ENABLED ? true : getSetting(key)
+      )
+    initNormally()
+    expect(component.suggestionSource).toEqual(SuggestionSource.AI)
+    component.suggestionSourceOverride.set(SuggestionSource.ML)
+    expect(component.suggestionSource).toEqual(SuggestionSource.ML)
+
+    jest
+      .spyOn(documentService, 'get')
+      .mockReturnValueOnce(of(Object.assign({}, doc)))
+    ;(component as any).loadDocument(doc.id, true)
+    expect(component.suggestionSourceOverride()).toBeNull()
+    expect(component.fetchedSuggestionSources()).toEqual([])
+  })
+
+  it('should keep suggestions from one source if the other fails', () => {
+    settingsService.set(
+      SETTINGS_KEYS.DOCUMENT_EDITING_SUGGESTION_SOURCE,
+      SuggestionSource.Both
+    )
+    const getSetting = settingsService.get.bind(settingsService)
+    jest
+      .spyOn(settingsService, 'get')
+      .mockImplementation((key) =>
+        key === SETTINGS_KEYS.AI_ENABLED ? true : getSetting(key)
+      )
+    const errorSpy = jest.spyOn(toastService, 'showError')
+    jest
+      .spyOn(documentService, 'getSuggestions')
+      .mockReturnValue(of({ tags: [42] }))
+    jest
+      .spyOn(documentService, 'getAiSuggestions')
+      .mockReturnValue(throwError(() => new Error('failed')))
+    initNormally()
+    expect(errorSpy).toHaveBeenCalled()
+    expect(component.suggestions().tags).toEqual([42])
+    expect(component.fetchedSuggestionSources()).toEqual([SuggestionSource.ML])
+  })
+
  it('should show error if needed for get suggestions', () => {
    const suggestionsSpy = jest.spyOn(documentService, 'getSuggestions')
    const errorSpy = jest.spyOn(toastService, 'showError')
@@ -28,7 +28,7 @@ import {
 import { dirtyCheck, DirtyComponent } from '@ngneat/dirty-check-forms'
 import { NgxBootstrapIconsModule } from 'ngx-bootstrap-icons'
 import { DeviceDetectorService } from 'ngx-device-detector'
-import { BehaviorSubject, Observable, of, Subject, timer } from 'rxjs'
+import { BehaviorSubject, merge, Observable, of, Subject, timer } from 'rxjs'
 import {
  catchError,
  debounceTime,
@@ -48,7 +48,10 @@ import { DataType } from 'src/app/data/datatype'
 import { Document, DocumentVersionInfo } from 'src/app/data/document'
 import { DocumentMetadata } from 'src/app/data/document-metadata'
 import { DocumentNote } from 'src/app/data/document-note'
-import { DocumentSuggestions } from 'src/app/data/document-suggestions'
+import {
+  DocumentSuggestions,
+  mergeSuggestions,
+} from 'src/app/data/document-suggestions'
 import { DocumentType } from 'src/app/data/document-type'
 import { FilterRule } from 'src/app/data/filter-rule'
 import {
@@ -63,7 +66,7 @@ import {
 import { ObjectWithId } from 'src/app/data/object-with-id'
 import { StoragePath } from 'src/app/data/storage-path'
 import { Tag } from 'src/app/data/tag'
-import { SETTINGS_KEYS } from 'src/app/data/ui-settings'
+import { SETTINGS_KEYS, SuggestionSource } from 'src/app/data/ui-settings'
 import { User } from 'src/app/data/user'
 import { IfPermissionsDirective } from 'src/app/directives/if-permissions.directive'
 import { CustomDatePipe } from 'src/app/pipes/custom-date.pipe'
@@ -240,6 +243,10 @@ export class DocumentDetailComponent
  private readonly autoSuggestSetting = this.settings.getSignal<boolean>(
    SETTINGS_KEYS.DOCUMENT_EDITING_AUTO_SUGGEST
  )
+  private readonly suggestionSourceSetting =
+    this.settings.getSignal<SuggestionSource>(
+      SETTINGS_KEYS.DOCUMENT_EDITING_SUGGESTION_SOURCE
+    )
  private readonly hiddenFieldsSetting = this.settings.getSignal<
    DocumentDetailFieldID[]
  >(SETTINGS_KEYS.DOCUMENT_DETAILS_HIDDEN_FIELDS)
@@ -261,6 +268,9 @@ export class DocumentDetailComponent
  readonly metadata = signal<DocumentMetadata>(undefined)
  readonly suggestions = signal<DocumentSuggestions>(undefined)
  readonly suggestionsLoading = signal(false)
+  // per-document, resets on navigation
+  readonly suggestionSourceOverride = signal<SuggestionSource>(null)
+  readonly fetchedSuggestionSources = signal<SuggestionSource[]>([])
  readonly users = signal<User[]>(undefined)

  readonly title = signal<string>(undefined)
@@ -365,6 +375,15 @@ export class DocumentDetailComponent
    return this.autoSuggestSetting()
  }

+  get defaultSuggestionSource(): SuggestionSource {
+    return this.aiEnabled ? this.suggestionSourceSetting() : SuggestionSource.ML
+  }
+
+  get suggestionSource(): SuggestionSource {
+    if (!this.aiEnabled) return SuggestionSource.ML
+    return this.suggestionSourceOverride() ?? this.defaultSuggestionSource
+  }
+
  get archiveContentRenderType(): ContentRenderType {
    const hasArchiveVersion =
      this.metadata()?.has_archive_version ??
@@ -590,6 +609,8 @@ export class DocumentDetailComponent
          }
          this.documentId.set(doc.id)
          this.suggestions.set(null)
+          this.suggestionSourceOverride.set(null)
+          this.fetchedSuggestionSources.set([])
          const openDocument = this.openDocumentService.getOpenDocument(
            this.documentId()
          )
@@ -1077,29 +1098,44 @@ export class DocumentDetailComponent
    return this.documentForm.get('custom_fields') as FormArray
  }

-  getSuggestions() {
+  getSuggestions(source: SuggestionSource = this.suggestionSource) {
+    const sources = (
+      source === SuggestionSource.Both
+        ? [SuggestionSource.ML, SuggestionSource.AI]
+        : [source]
+    ).filter((s) => !this.fetchedSuggestionSources().includes(s))
+    if (!sources.length) return
+
    this.suggestionsLoading.set(true)
-    const suggestionsObservable = this.aiEnabled
-      ? this.documentsService.getAiSuggestions(this.documentId())
-      : this.documentsService.getSuggestions(this.documentId())
-    suggestionsObservable
+    merge(
+      ...sources.map((s) =>
+        (s === SuggestionSource.AI
+          ? this.documentsService.getAiSuggestions(this.documentId())
+          : this.documentsService.getSuggestions(this.documentId())
+        ).pipe(
+          first(),
+          map((result) => ({ source: s, result })),
+          catchError((error) => {
+            this.toastService.showError(
+              $localize`Error retrieving suggestions.`,
+              error
+            )
+            return of(null)
+          })
+        )
+      )
+    )
      .pipe(
-        first(),
        takeUntil(this.unsubscribeNotifier),
        takeUntil(this.docChangeNotifier),
        finalize(() => this.suggestionsLoading.set(false))
      )
-      .subscribe({
-        next: (result) => {
-          this.suggestions.set(result)
-        },
-        error: (error) => {
-          this.suggestions.set(null)
-          this.toastService.showError(
-            $localize`Error retrieving suggestions.`,
-            error
-          )
-        },
+      .subscribe((response) => {
+        if (!response) return
+        this.fetchedSuggestionSources.update((f) => [...f, response.source])
+        this.suggestions.set(
+          mergeSuggestions(this.suggestions(), response.result)
+        )
      })
  }

@@ -15,3 +15,33 @@ export interface DocumentSuggestions {

  dates?: string[] // ISO-formatted date string e.g. 2022-11-03
 }
+
+const union = <T>(a: T[] = [], b: T[] = []): T[] => [...new Set([...a, ...b])]
+
+export function mergeSuggestions(
+  a: DocumentSuggestions,
+  b: DocumentSuggestions
+): DocumentSuggestions {
+  if (!a) return b
+  return {
+    title: a.title || b.title,
+    tags: union(a.tags, b.tags),
+    suggested_tags: union(a.suggested_tags, b.suggested_tags),
+    correspondents: union(a.correspondents, b.correspondents),
+    suggested_correspondents: union(
+      a.suggested_correspondents,
+      b.suggested_correspondents
+    ),
+    document_types: union(a.document_types, b.document_types),
+    suggested_document_types: union(
+      a.suggested_document_types,
+      b.suggested_document_types
+    ),
+    storage_paths: union(a.storage_paths, b.storage_paths),
+    suggested_storage_paths: union(
+      a.suggested_storage_paths,
+      b.suggested_storage_paths
+    ),
+    dates: union(a.dates, b.dates),
+  }
+}
@@ -20,6 +20,12 @@ export enum GlobalSearchType {
  TITLE_CONTENT = 'title-content',
 }

+export enum SuggestionSource {
+  ML = 'ml',
+  AI = 'ai',
+  Both = 'both',
+}
+
 export enum CollapsibleSection {
  ATTRIBUTES = 'attributes',
 }
@@ -98,6 +104,8 @@ export const SETTINGS_KEYS = {
    'general-settings:document-editing:overlay-thumbnail',
  DOCUMENT_EDITING_AUTO_SUGGEST:
    'general-settings:document-editing:auto-suggest',
+  DOCUMENT_EDITING_SUGGESTION_SOURCE:
+    'general-settings:document-editing:suggestion-source',
  DOCUMENT_DETAILS_HIDDEN_FIELDS:
    'general-settings:document-details:hidden-fields',
  SEARCH_DB_ONLY: 'general-settings:search:db-only',
@@ -326,6 +334,11 @@ export const SETTINGS: UiSetting[] = [
    type: 'boolean',
    default: true,
  },
+  {
+    key: SETTINGS_KEYS.DOCUMENT_EDITING_SUGGESTION_SOURCE,
+    type: 'string',
+    default: SuggestionSource.AI,
+  },
  {
    key: SETTINGS_KEYS.DOCUMENT_DETAILS_HIDDEN_FIELDS,
    type: 'array',
@@ -74,6 +74,7 @@ import {
  clipboardCheckFill,
  clipboardFill,
  clockHistory,
+  cpu,
  creditCard,
  dash,
  dashCircle,
@@ -118,6 +119,7 @@ import {
  infoCircle,
  journalBookmarkFill,
  journals,
+  lightbulb,
  link,
  list,
  listNested,
@@ -322,6 +324,7 @@ const icons = {
  clipboardCheckFill,
  clipboardFill,
  clockHistory,
+  cpu,
  cash,
  creditCard,
  dash,
@@ -367,6 +370,7 @@ const icons = {
  infoCircle,
  journalBookmarkFill,
  journals,
+  lightbulb,
  link,
  list,
  listNested,
@@ -292,6 +292,7 @@ a.btn-link:active,
 a.btn-link:focus-visible,
 .btn-link:hover,
 .btn-link:active,
+.btn-link.show,
 .btn-link:focus-visible {
  color: var(--pngx-primary-lighten-10) !important;
  .primary-light & {
@@ -25,10 +25,20 @@ export class PDFFindController {
  onIsPageVisible?: () => boolean
 }

+export const LinkTarget = {
+  NONE: 0,
+  SELF: 1,
+  BLANK: 2,
+  PARENT: 3,
+  TOP: 4,
+}
+
 export class PDFLinkService {
  private document?: unknown
  private viewer?: unknown

+  constructor(readonly options: Record<string, unknown> = {}) {}
+
  setDocument(document: unknown): void {
    this.document = document
  }
@@ -3,6 +3,7 @@ from __future__ import annotations
 import logging
 import tempfile
 import uuid
+from functools import partial
 from pathlib import Path
 from typing import TYPE_CHECKING
 from typing import Literal
@@ -17,6 +18,7 @@ from django.db.models import Max
 from django.db.models import Q
 from django.utils import timezone

+from documents import pdf_ops
 from documents.data_models import ConsumableDocument
 from documents.data_models import DocumentMetadataOverrides
 from documents.data_models import DocumentSource
@@ -116,6 +118,11 @@ def _resolve_root_and_source_doc(
    )


+def _scratch_path(name: str) -> Path:
+    """A path inside a fresh directory under SCRATCH_DIR."""
+    return Path(tempfile.mkdtemp(dir=settings.SCRATCH_DIR)) / name
+
+
 def set_correspondent(
    doc_ids: list[int],
    correspondent: Correspondent,
@@ -474,8 +481,6 @@ def rotate(
        pair = _resolve_root_and_source_doc(doc, source_mode=source_mode)
        docs_by_root_id.setdefault(pair.root_doc.id, pair)

-    import pikepdf
-
    for pair in docs_by_root_id.values():
        if pair.source_doc.mime_type != "application/pdf":
            logger.warning(
@@ -488,11 +493,7 @@ def rotate(
                Path(tempfile.mkdtemp(dir=settings.SCRATCH_DIR))
                / f"{pair.root_doc.id}_rotated.pdf"
            )
-            with pikepdf.open(pair.source_doc.source_path) as pdf:
-                for page in pdf.pages:
-                    page.rotate(degrees, relative=True)
-                pdf.remove_unreferenced_resources()
-                pdf.save(filepath)
+            pdf_ops.rotate_pdf(pair.source_doc.source_path, filepath, degrees)

            # Preserve metadata/permissions via overrides; mark as new version
            overrides = DocumentMetadataOverrides().from_document(pair.root_doc)
@@ -535,48 +536,45 @@ def merge(
    qs = Document.objects.select_related("root_document").filter(id__in=doc_ids)
    docs_by_id = {doc.id: doc for doc in qs}
    affected_docs: list[int] = []
-    import pikepdf
-
-    merged_pdf = pikepdf.new()
-    version: str = merged_pdf.pdf_version
    handoff_asn: int | None = None
-    # use doc_ids to preserve order
-    for doc_id in doc_ids:
-        doc = docs_by_id.get(doc_id)
-        if doc is None:
-            continue
-        pair = _resolve_root_and_source_doc(doc, source_mode=source_mode)
-        try:
-            doc_path = (
-                pair.source_doc.archive_path
-                if archive_fallback
-                and pair.source_doc.mime_type != "application/pdf"
-                and pair.source_doc.has_archive_version
-                else pair.source_doc.source_path
-            )
-            with pikepdf.open(str(doc_path)) as pdf:
-                version = max(version, pdf.pdf_version)
-                merged_pdf.pages.extend(pdf.pages)
-            affected_docs.append(doc.id)
-            if handoff_asn is None and doc.archive_serial_number is not None:
-                handoff_asn = doc.archive_serial_number
-        except Exception as e:
-            logger.exception(
-                f"Error merging document {doc.id}, it will not be included in the merge: {e}",
-            )
-    if len(affected_docs) == 0:
-        logger.warning("No documents were merged")
-        return "OK"
+    with pdf_ops.PdfMerger() as merger:
+        # use doc_ids to preserve order
+        for doc_id in doc_ids:
+            doc = docs_by_id.get(doc_id)
+            if doc is None:
+                continue
+            pair = _resolve_root_and_source_doc(doc, source_mode=source_mode)
+            try:
+                # archive_path is None when there is no archive version
+                archive_path = (
+                    pair.source_doc.archive_path
+                    if archive_fallback
+                    and pair.source_doc.mime_type != "application/pdf"
+                    else None
+                )
+                merger.add(
+                    archive_path
+                    if archive_path is not None
+                    else pair.source_doc.source_path,
+                )
+                affected_docs.append(doc.id)
+                if handoff_asn is None and doc.archive_serial_number is not None:
+                    handoff_asn = doc.archive_serial_number
+            except Exception as e:
+                logger.exception(
+                    f"Error merging document {doc.id}, it will not be included in the merge: {e}",
+                )
+        if len(affected_docs) == 0:
+            logger.warning("No documents were merged")
+            return "OK"

-    filepath = (
-        Path(
-            tempfile.mkdtemp(dir=settings.SCRATCH_DIR),
+        filepath = (
+            Path(
+                tempfile.mkdtemp(dir=settings.SCRATCH_DIR),
+            )
+            / f"{'_'.join([str(doc_id) for doc_id in affected_docs])[:100]}_merged.pdf"
        )
-        / f"{'_'.join([str(doc_id) for doc_id in affected_docs])[:100]}_merged.pdf"
-    )
-    merged_pdf.remove_unreferenced_resources()
-    merged_pdf.save(filepath, min_version=version)
-    merged_pdf.close()
+        merger.save(filepath)

    if metadata_document_id:
        metadata_document = qs.get(id=metadata_document_id)
@@ -752,64 +750,60 @@ def split(
    )
    doc = Document.objects.select_related("root_document").get(id=doc_ids[0])
    pair = _resolve_root_and_source_doc(doc, source_mode=source_mode)
-    import pikepdf

    consume_tasks = []

    try:
-        with pikepdf.open(pair.source_doc.source_path) as pdf:
-            for idx, split_doc in enumerate(pages):
-                dst: pikepdf.Pdf = pikepdf.new()
-                for page in split_doc:
-                    dst.pages.append(pdf.pages[page - 1])
-                filepath: Path = (
-                    Path(
-                        tempfile.mkdtemp(dir=settings.SCRATCH_DIR),
-                    )
-                    / f"{doc.id}_{split_doc[0]}-{split_doc[-1]}.pdf"
-                )
-                dst.remove_unreferenced_resources()
-                dst.save(filepath)
-                dst.close()
+        outputs = [
+            (
+                [pdf_ops.PageSpec(page) for page in split_doc],
+                partial(_scratch_path, f"{doc.id}_{split_doc[0]}-{split_doc[-1]}.pdf"),
+            )
+            for split_doc in pages
+        ]
+        filepaths = pdf_ops.build_pdfs(pair.source_doc.source_path, outputs)

-                overrides: DocumentMetadataOverrides = (
-                    DocumentMetadataOverrides().from_document(doc)
-                )
-                overrides.title = f"{doc.title} (split {idx + 1})"
-                if user is not None:
-                    overrides.owner_id = user.id
-                if not delete_originals:
-                    overrides.skip_asn_if_exists = True
-                logger.info(
-                    f"Adding split document with pages {split_doc} to the task queue.",
-                )
-                consume_tasks.append(
-                    consume_file.s(
-                        input_doc=ConsumableDocument(
-                            source=DocumentSource.ConsumeFolder,
-                            original_file=filepath,
-                        ),
-                        overrides=overrides,
-                    ).set(headers={"trigger_source": trigger_source}),
-                )
+        for idx, (split_doc, filepath) in enumerate(
+            zip(pages, filepaths, strict=True),
+        ):
+            overrides: DocumentMetadataOverrides = (
+                DocumentMetadataOverrides().from_document(doc)
+            )
+            overrides.title = f"{doc.title} (split {idx + 1})"
+            if user is not None:
+                overrides.owner_id = user.id
+            if not delete_originals:
+                overrides.skip_asn_if_exists = True
+            logger.info(
+                f"Adding split document with pages {split_doc} to the task queue.",
+            )
+            consume_tasks.append(
+                consume_file.s(
+                    input_doc=ConsumableDocument(
+                        source=DocumentSource.ConsumeFolder,
+                        original_file=filepath,
+                    ),
+                    overrides=overrides,
+                ).set(headers={"trigger_source": trigger_source}),
+            )

-            if delete_originals:
-                backup = release_archive_serial_numbers([doc.id])
-                logger.info(
-                    "Queueing removal of original document after consumption of the split documents",
-                )
-                try:
-                    chord(
-                        header=consume_tasks,
-                        body=delete.si([doc.id]),
-                    ).on_error(
-                        restore_archive_serial_numbers_task.s(backup),
-                    ).apply_async()
-                except Exception:
-                    restore_archive_serial_numbers(backup)
-                    raise
-            else:
-                group(consume_tasks).delay()
+        if delete_originals:
+            backup = release_archive_serial_numbers([doc.id])
+            logger.info(
+                "Queueing removal of original document after consumption of the split documents",
+            )
+            try:
+                chord(
+                    header=consume_tasks,
+                    body=delete.si([doc.id]),
+                ).on_error(
+                    restore_archive_serial_numbers_task.s(backup),
+                ).apply_async()
+            except Exception:
+                restore_archive_serial_numbers(backup)
+                raise
+        else:
+            group(consume_tasks).delay()

    except Exception as e:
        logger.exception(f"Error splitting document {doc.id}: {e}")
@@ -830,8 +824,7 @@ def delete_pages(
    )
    doc = Document.objects.select_related("root_document").get(id=doc_ids[0])
    pair = _resolve_root_and_source_doc(doc, source_mode=source_mode)
-    pages = sorted(pages)  # sort pages to avoid index issues
-    import pikepdf
+    pages = sorted(set(pages))

    try:
        # Produce edited PDF to a temp file and create a new version
@@ -839,13 +832,7 @@ def delete_pages(
            Path(tempfile.mkdtemp(dir=settings.SCRATCH_DIR))
            / f"{pair.root_doc.id}_pages_deleted.pdf"
        )
-        with pikepdf.open(pair.source_doc.source_path) as pdf:
-            offset = 1  # pages are 1-indexed
-            for page_num in pages:
-                pdf.pages.remove(pdf.pages[page_num - offset])
-                offset += 1  # remove() changes the index of the pages
-            pdf.remove_unreferenced_resources()
-            pdf.save(filepath)
+        pdf_ops.remove_pages(pair.source_doc.source_path, filepath, pages)

        overrides = DocumentMetadataOverrides().from_document(pair.root_doc)
        if user is not None:
@@ -894,47 +881,28 @@ def edit_pdf(
    )
    doc = Document.objects.select_related("root_document").get(id=doc_ids[0])
    pair = _resolve_root_and_source_doc(doc, source_mode=source_mode)
-    import pikepdf
-
-    pdf_docs: list[pikepdf.Pdf] = []
-
    try:
-        if not operations:
-            raise ValueError("Output document index is out of bounds")
-
-        max_idx = max(op.get("doc", 0) for op in operations)
-        if update_document and max_idx > 0:
-            logger.error(
-                "Update requested but multiple output documents specified",
+        output_count = pdf_ops.validate_page_operations(
+            operations,
+            single_output=update_document,
+        )
+        page_specs: list[list[pdf_ops.PageSpec]] = [[] for _ in range(output_count)]
+        for op in operations:
+            page_specs[op.get("doc", 0)].append(
+                pdf_ops.PageSpec(op["page"], op.get("rotate", 0)),
            )
-            raise ValueError("Multiple output documents specified")
-
-        if any(
-            op.get("doc", 0) < 0 or op.get("doc", 0) >= len(operations)
-            for op in operations
-        ):
-            raise ValueError("Output document index is out of bounds")
-
-        with pikepdf.open(pair.source_doc.source_path) as src:
-            # prepare output documents
-            pdf_docs = [pikepdf.new() for _ in range(max_idx + 1)]
-
-            for op in operations:
-                dst = pdf_docs[op.get("doc", 0)]
-                page = src.pages[op["page"] - 1]
-                dst.pages.append(page)
-                if op.get("rotate"):
-                    dst.pages[-1].rotate(op["rotate"], relative=True)

        if update_document:
            # Create a new version from the edited PDF rather than replacing in-place
-            pdf = pdf_docs[0]
-            pdf.remove_unreferenced_resources()
-            filepath: Path = (
-                Path(tempfile.mkdtemp(dir=settings.SCRATCH_DIR))
-                / f"{pair.root_doc.id}_edited.pdf"
+            (filepath,) = pdf_ops.build_pdfs(
+                pair.source_doc.source_path,
+                [
+                    (
+                        page_specs[0],
+                        partial(_scratch_path, f"{pair.root_doc.id}_edited.pdf"),
+                    ),
+                ],
            )
-            pdf.save(filepath)
            overrides = (
                DocumentMetadataOverrides().from_document(pair.root_doc)
                if include_metadata
@@ -955,6 +923,19 @@ def edit_pdf(
                headers={"trigger_source": trigger_source},
            )
        else:
+            version_filepaths = pdf_ops.build_pdfs(
+                pair.source_doc.source_path,
+                [
+                    (
+                        specs,
+                        partial(
+                            _scratch_path,
+                            f"{pair.root_doc.id}_edit_{idx}.pdf",
+                        ),
+                    )
+                    for idx, specs in enumerate(page_specs, start=1)
+                ],
+            )
            consume_tasks = []
            overrides = (
                DocumentMetadataOverrides().from_document(pair.root_doc)
@@ -966,15 +947,9 @@ def edit_pdf(
                overrides.actor_id = user.id
            if not delete_original:
                overrides.skip_asn_if_exists = True
-            if delete_original and len(pdf_docs) == 1:
+            if delete_original and output_count == 1:
                overrides.asn = pair.root_doc.archive_serial_number
-            for idx, pdf in enumerate(pdf_docs, start=1):
-                version_filepath: Path = (
-                    Path(tempfile.mkdtemp(dir=settings.SCRATCH_DIR))
-                    / f"{pair.root_doc.id}_edit_{idx}.pdf"
-                )
-                pdf.remove_unreferenced_resources()
-                pdf.save(version_filepath)
+            for version_filepath in version_filepaths:
                consume_tasks.append(
                    consume_file.s(
                        input_doc=ConsumableDocument(
@@ -1024,8 +999,6 @@ def remove_password(
    """
    Remove password protection from PDF documents.
    """
-    import pikepdf
-
    for doc_id in doc_ids:
        doc = Document.objects.select_related("root_document").get(id=doc_id)
        pair = _resolve_root_and_source_doc(doc, source_mode=source_mode)
@@ -1039,76 +1012,69 @@ def remove_password(
                doc.id,
                pair.source_doc.source_path,
            )
-            try:
-                with pikepdf.open(source_path) as pdf:
-                    if not pdf.is_encrypted:
-                        logger.info(
-                            "Skipping password removal for document %s because the "
-                            "source PDF is not encrypted",
-                            pair.root_doc.id,
-                        )
-                        continue
-            except pikepdf.PasswordError:
-                # Password-protected PDFs need the supplied password below.
-                pass
-
-            with pikepdf.open(source_path, password=password) as pdf:
-                filepath: Path = (
-                    Path(tempfile.mkdtemp(dir=settings.SCRATCH_DIR))
-                    / f"{pair.root_doc.id}_unprotected.pdf"
+            if not pdf_ops.needs_decrypt(source_path):
+                logger.info(
+                    "Skipping password removal for document %s because the "
+                    "source PDF is not encrypted",
+                    pair.root_doc.id,
                )
-                pdf.remove_unreferenced_resources()
-                pdf.save(filepath)
+                continue

-                if update_document:
-                    # Create a new version rather than modifying the root/original in place.
-                    overrides = (
-                        DocumentMetadataOverrides().from_document(pair.root_doc)
-                        if include_metadata
-                        else DocumentMetadataOverrides()
-                    )
-                    if user is not None:
-                        overrides.owner_id = user.id
-                        overrides.actor_id = user.id
-                    consume_file.apply_async(
-                        kwargs={
-                            "input_doc": ConsumableDocument(
-                                source=DocumentSource.ConsumeFolder,
-                                original_file=filepath,
-                                root_document_id=pair.root_doc.id,
-                            ),
-                            "overrides": overrides,
-                        },
-                        headers={"trigger_source": trigger_source},
-                    )
+            filepath = pdf_ops.decrypt_pdf(
+                source_path,
+                partial(_scratch_path, f"{pair.root_doc.id}_unprotected.pdf"),
+                password,
+            )
+
+            if update_document:
+                # Create a new version rather than modifying the root/original in place.
+                overrides = (
+                    DocumentMetadataOverrides().from_document(pair.root_doc)
+                    if include_metadata
+                    else DocumentMetadataOverrides()
+                )
+                if user is not None:
+                    overrides.owner_id = user.id
+                    overrides.actor_id = user.id
+                consume_file.apply_async(
+                    kwargs={
+                        "input_doc": ConsumableDocument(
+                            source=DocumentSource.ConsumeFolder,
+                            original_file=filepath,
+                            root_document_id=pair.root_doc.id,
+                        ),
+                        "overrides": overrides,
+                    },
+                    headers={"trigger_source": trigger_source},
+                )
+            else:
+                consume_tasks = []
+                overrides = (
+                    DocumentMetadataOverrides().from_document(pair.root_doc)
+                    if include_metadata
+                    else DocumentMetadataOverrides()
+                )
+                if user is not None:
+                    overrides.owner_id = user.id
+                    overrides.actor_id = user.id
+
+                consume_tasks.append(
+                    consume_file.s(
+                        input_doc=ConsumableDocument(
+                            source=DocumentSource.ConsumeFolder,
+                            original_file=filepath,
+                        ),
+                        overrides=overrides,
+                    ).set(headers={"trigger_source": trigger_source}),
+                )
+
+                if delete_original:
+                    chord(
+                        header=consume_tasks,
+                        body=delete.si([doc.id]),
+                    ).delay()
                else:
-                    consume_tasks = []
-                    overrides = (
-                        DocumentMetadataOverrides().from_document(pair.root_doc)
-                        if include_metadata
-                        else DocumentMetadataOverrides()
-                    )
-                    if user is not None:
-                        overrides.owner_id = user.id
-                        overrides.actor_id = user.id
-
-                    consume_tasks.append(
-                        consume_file.s(
-                            input_doc=ConsumableDocument(
-                                source=DocumentSource.ConsumeFolder,
-                                original_file=filepath,
-                            ),
-                            overrides=overrides,
-                        ).set(headers={"trigger_source": trigger_source}),
-                    )
-
-                    if delete_original:
-                        chord(
-                            header=consume_tasks,
-                            body=delete.si([doc.id]),
-                        ).delay()
-                    else:
-                        group(consume_tasks).delay()
+                    group(consume_tasks).delay()

        except Exception as e:
            logger.exception(
@@ -0,0 +1,184 @@
+"""
+Pure PDF page operations used by documents.bulk_edit.
+
+This module deliberately knows nothing about Django, Celery or the documents
+app: callers resolve documents, choose output paths and queue work. Every
+function that writes a PDF removes unreferenced resources before saving.
+
+pikepdf is always called as ``pikepdf.open(...)`` / ``pikepdf.new()`` (never
+``from pikepdf import open``) so tests can patch those module attributes.
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+from typing import NamedTuple
+
+import pikepdf
+
+if TYPE_CHECKING:
+    from collections.abc import Callable
+    from collections.abc import Iterable
+    from collections.abc import Mapping
+    from collections.abc import Sequence
+    from pathlib import Path
+    from types import TracebackType
+
+
+class PageSpec(NamedTuple):
+    """One page of an output PDF: a 1-indexed source page, optionally rotated."""
+
+    page: int
+    rotate: int = 0  # relative degrees, 0 leaves the page alone
+
+
+def _require_positive(pages: Iterable[int]) -> None:
+    for page in pages:
+        if page < 1:
+            raise ValueError(f"Page numbers start at 1, got {page}")
+
+
+def rotate_pdf(src: Path, dst: Path, degrees: int) -> None:
+    """
+    Rotate every page relatively on the opened document, not a rebuild, so Info,
+    XMP and outlines are kept. ``src`` is not modified.
+    """
+    with pikepdf.open(src) as pdf:
+        for page in pdf.pages:
+            page.rotate(degrees, relative=True)
+        pdf.remove_unreferenced_resources()
+        pdf.save(dst)
+
+
+def remove_pages(src: Path, dst: Path, pages: Iterable[int]) -> None:
+    """
+    Remove 1-indexed pages from the opened document, not a rebuild, so Info, XMP
+    and outlines are kept. ``src`` is not modified.
+
+    Duplicates are ignored. Pages are removed highest first so earlier removals
+    never shift the index of later ones.
+    """
+    unique = sorted(set(pages))
+    _require_positive(unique)
+    with pikepdf.open(src) as pdf:
+        for page_num in reversed(unique):
+            del pdf.pages[page_num - 1]
+        pdf.remove_unreferenced_resources()
+        pdf.save(dst)
+
+
+def build_pdfs(
+    src: Path,
+    outputs: Sequence[tuple[Sequence[PageSpec], Callable[[], Path]]],
+) -> list[Path]:
+    """
+    Build one new PDF per output from pages of ``src``, opening ``src`` once.
+
+    Each output is ``(page_specs, make_dst)``. ``make_dst`` is called after that
+    output's pages are copied and immediately before it is saved, so a bad page
+    number never leaves a destination behind. Document-level data (Info, XMP,
+    outlines) is not carried over. Returns the written paths in output order.
+    """
+    for specs, _ in outputs:
+        _require_positive(spec.page for spec in specs)
+
+    written: list[Path] = []
+    with pikepdf.open(src) as source:
+        for specs, make_dst in outputs:
+            dst = pikepdf.new()
+            for spec in specs:
+                dst.pages.append(source.pages[spec.page - 1])
+                if spec.rotate:
+                    dst.pages[-1].rotate(spec.rotate, relative=True)
+            dst.remove_unreferenced_resources()
+            path = make_dst()
+            dst.save(path)
+            dst.close()
+            written.append(path)
+    return written
+
+
+def validate_page_operations(
+    operations: Sequence[Mapping[str, int]],
+    *,
+    single_output: bool,
+) -> int:
+    """
+    Validate ``edit_pdf`` style operations and return the output document count.
+
+    Each operation has ``page`` and optionally ``rotate`` and ``doc`` (the output
+    document index, default 0). The bounds rule is kept as it was: a ``doc`` index
+    must be below the number of operations.
+    """
+    if not operations:
+        raise ValueError("Output document index is out of bounds")
+
+    max_idx = max(op.get("doc", 0) for op in operations)
+    if single_output and max_idx > 0:
+        raise ValueError("Multiple output documents specified")
+
+    if any(
+        op.get("doc", 0) < 0 or op.get("doc", 0) >= len(operations) for op in operations
+    ):
+        raise ValueError("Output document index is out of bounds")
+
+    return max_idx + 1
+
+
+def needs_decrypt(src: Path) -> bool:
+    """
+    True if ``src`` is encrypted. A PDF that needs a password to open at all
+    counts as encrypted.
+    """
+    try:
+        with pikepdf.open(src) as pdf:
+            return bool(pdf.is_encrypted)
+    except pikepdf.PasswordError:
+        return True
+
+
+def decrypt_pdf(src: Path, make_dst: Callable[[], Path], password: str) -> Path:
+    """
+    Write an unencrypted copy of ``src`` and return its path.
+
+    ``make_dst`` is only called once the password has been accepted, so a wrong
+    password never leaves a destination behind.
+    """
+    with pikepdf.open(src, password=password) as pdf:
+        pdf.remove_unreferenced_resources()
+        dst = make_dst()
+        pdf.save(dst)
+    return dst
+
+
+class PdfMerger:
+    """
+    Accumulates the pages of several PDFs into one new PDF.
+
+    ``add`` raises if a source cannot be read; deciding whether to skip it is the
+    caller's policy. Use as a context manager so the merged PDF is closed.
+    """
+
+    def __init__(self) -> None:
+        self._pdf = pikepdf.new()
+        self._version: str = self._pdf.pdf_version
+
+    def __enter__(self) -> PdfMerger:
+        return self
+
+    def __exit__(
+        self,
+        exc_type: type[BaseException] | None,
+        exc: BaseException | None,
+        tb: TracebackType | None,
+    ) -> None:
+        self._pdf.close()
+
+    def add(self, path: Path) -> None:
+        with pikepdf.open(str(path)) as pdf:
+            self._version = max(self._version, pdf.pdf_version)
+            self._pdf.pages.extend(pdf.pages)
+
+    def save(self, dst: Path) -> None:
+        self._pdf.remove_unreferenced_resources()
+        self._pdf.save(dst, min_version=self._version)
@@ -1079,6 +1079,16 @@ class DocumentSerializer(
    )

    def get_page_count(self, obj) -> int | None:
+        # Like content versions get their own page count from the newest version,
+        # use the prefetched versions cache to avoid an extra query
+        prefetched_cache = getattr(obj, "_prefetched_objects_cache", None)
+        prefetched_versions = (
+            prefetched_cache.get("versions")
+            if isinstance(prefetched_cache, dict)
+            else None
+        )
+        if obj.root_document_id is None and prefetched_versions:
+            return sort_versions_newest_first(prefetched_versions)[0].page_count
        return obj.page_count

    @extend_schema_field(DuplicateDocumentSummarySerializer(many=True))
@@ -2127,6 +2137,8 @@ class BulkEditSerializer(
            raise serializers.ValidationError("pages must be a list")
        if not all(isinstance(i, int) for i in parameters["pages"]):
            raise serializers.ValidationError("pages must be a list of integers")
+        if any(i < 1 for i in parameters["pages"]):
+            raise serializers.ValidationError("pages must be positive integers")

    def _validate_parameters_merge(self, parameters) -> None:
        if "delete_originals" in parameters:
@@ -12,7 +12,7 @@
 	<meta name="robots" content="noindex,nofollow">
 	<meta name="author" content="The Paperless-ngx Team">
 	<link rel="icon" type="image/x-icon" href="favicon.ico">
-	<link rel="manifest" href="{% static webmanifest %}">
+	<link rel="manifest" href="{% static webmanifest %}" crossorigin="use-credentials">
 	<link rel="stylesheet" href="{% static styles_css %}">
 	<link rel="apple-touch-icon" href="{% static apple_touch_icon %}">
 </head>
@@ -1843,6 +1843,36 @@ class TestBulkEditAPI(DirectoriesMixin, APITestCase):
        m.assert_called_once()
        self.assertEqual(m.call_args.kwargs["pages"], [[1], [2, 3, 4], [5]])

+    @mock.patch("documents.serialisers.bulk_edit.delete_pages")
+    def test_bulk_edit_delete_pages_rejects_pages_below_one(self, m) -> None:
+        """
+        GIVEN:
+            - A legacy delete_pages bulk edit
+        WHEN:
+            - API to bulk edit is called with a page number below 1
+        THEN:
+            - API returns HTTP 400
+            - delete_pages is not called
+        """
+        self.setup_mock(m, "delete_pages")
+
+        for pages in ([0], [-1], [1, 0]):
+            with self.subTest(pages=pages):
+                response = self.client.post(
+                    "/api/documents/bulk_edit/",
+                    json.dumps(
+                        {
+                            "documents": [self.doc2.id],
+                            "method": "delete_pages",
+                            "parameters": {"pages": pages},
+                        },
+                    ),
+                    content_type="application/json",
+                )
+                self.assertEqual(response.status_code, status.HTTP_400_BAD_REQUEST)
+                self.assertIn(b"pages must be positive integers", response.content)
+        m.assert_not_called()
+
    @mock.patch("documents.views.bulk_edit.rotate")
    def test_rotate_insufficient_permissions(self, m) -> None:
        self.doc1.owner = User.objects.get(username="temp_admin")
@@ -19,6 +19,7 @@ from documents.models import Document
 from documents.versioning import annotate_effective_content
 from documents.views import DocumentSelectionMixin
 from paperless_testing.dirs import DirectoriesMixin
+from paperless_testing.factories import DocumentFactory
 from paperless_testing.factories import UserFactory
 from paperless_testing.http import read_streaming_response
 from paperless_testing.permissions import grant_global
@@ -821,6 +822,26 @@ class TestDocumentVersioningApi(DirectoriesMixin, APITestCase):
        self.assertEqual(resp.status_code, status.HTTP_200_OK)
        self.assertEqual(resp.data["content"], "v1-content")

+    def test_page_count_resolves_to_latest_version(self) -> None:
+        root = DocumentFactory(page_count=2)
+        DocumentFactory(root_document=root, version_index=1, page_count=1)
+        unversioned = DocumentFactory(page_count=5)
+
+        resp = self.client.get("/api/documents/?fields=id,page_count")
+        self.assertEqual(resp.status_code, status.HTTP_200_OK)
+        self.assertEqual(
+            {doc["id"]: doc["page_count"] for doc in resp.data["results"]},
+            {root.id: 1, unversioned.id: 5},
+        )
+
+        resp = self.client.get(f"/api/documents/{root.id}/")
+        self.assertEqual(resp.status_code, status.HTTP_200_OK)
+        self.assertEqual(resp.data["page_count"], 1)
+
+        resp = self.client.get(f"/api/documents/{root.id}/?version={root.id}")
+        self.assertEqual(resp.status_code, status.HTTP_200_OK)
+        self.assertEqual(resp.data["page_count"], 2)
+
    def _make_root_with_out_of_order_versions(self) -> tuple[Document, ...]:
        """
        A root whose newest version has a *lower* id than an older one, which is
@@ -0,0 +1,386 @@
+"""
+Tests for documents.pdf_ops.
+
+These use real PDFs from the sample directories. No database, Celery or mocks.
+Pages are compared by a hash of their content stream, so page identity and order
+are easy to assert.
+"""
+
+import ast
+import hashlib
+from collections.abc import Callable
+from pathlib import Path
+
+import pikepdf
+import pytest
+
+from documents import pdf_ops
+from documents.pdf_ops import PageSpec
+
+SRC_ROOT = Path(__file__).parents[2]
+SAMPLES = Path(__file__).parent / "samples"
+THREE_PAGES = SAMPLES / "documents" / "originals" / "0000002.pdf"
+TWELVE_PAGES = SAMPLES / "barcodes" / "split-by-asn-2.pdf"
+ENCRYPTED = SAMPLES / "password-is-test.pdf"
+SIGNED = SRC_ROOT / "paperless" / "tests" / "samples" / "tesseract" / "signed.pdf"
+
+
+def _page_fingerprint(page: pikepdf.Page) -> str:
+    contents = page.obj.get("/Contents")
+    assert contents is not None, "sample page has no /Contents"
+    streams = list(contents) if isinstance(contents, pikepdf.Array) else [contents]
+    return hashlib.sha1(b"".join(s.read_bytes() for s in streams)).hexdigest()
+
+
+def fingerprints(path: Path) -> list[str]:
+    with pikepdf.open(path) as pdf:
+        return [_page_fingerprint(page) for page in pdf.pages]
+
+
+def rotations(path: Path) -> list[int]:
+    with pikepdf.open(path) as pdf:
+        return [int(page.obj.get("/Rotate", 0)) for page in pdf.pages]
+
+
+def docinfo_keys(path: Path) -> set[str]:
+    with pikepdf.open(path) as pdf:
+        return set(pdf.docinfo.keys())
+
+
+def constant(path: Path) -> Callable[[], Path]:
+    return lambda: path
+
+
+@pytest.fixture
+def source_fingerprints() -> list[str]:
+    fps = fingerprints(THREE_PAGES)
+    assert len(set(fps)) == 3, "sample must have three distinct pages"
+    return fps
+
+
+class TestRotatePdf:
+    def test_rotation_is_relative_and_applies_to_every_page(self, tmp_path: Path):
+        once = tmp_path / "once.pdf"
+        twice = tmp_path / "twice.pdf"
+
+        pdf_ops.rotate_pdf(THREE_PAGES, once, 90)
+        pdf_ops.rotate_pdf(once, twice, 90)
+
+        assert rotations(once) == [90, 90, 90]
+        assert rotations(twice) == [180, 180, 180]
+        assert fingerprints(twice) == fingerprints(THREE_PAGES)
+
+    def test_keeps_document_info(self, tmp_path: Path):
+        dst = tmp_path / "out.pdf"
+
+        pdf_ops.rotate_pdf(THREE_PAGES, dst, 90)
+
+        assert "/Creator" in docinfo_keys(dst)
+
+
+class TestRemovePages:
+    def test_removes_selected_pages(
+        self,
+        tmp_path: Path,
+        source_fingerprints: list[str],
+    ):
+        dst = tmp_path / "out.pdf"
+
+        pdf_ops.remove_pages(THREE_PAGES, dst, [2])
+
+        assert fingerprints(dst) == [source_fingerprints[0], source_fingerprints[2]]
+
+    def test_duplicate_page_numbers_remove_the_page_once(
+        self,
+        tmp_path: Path,
+        source_fingerprints: list[str],
+    ):
+        dst = tmp_path / "out.pdf"
+
+        pdf_ops.remove_pages(THREE_PAGES, dst, [2, 2])
+
+        assert fingerprints(dst) == [source_fingerprints[0], source_fingerprints[2]]
+
+    def test_unordered_pages(self, tmp_path: Path, source_fingerprints: list[str]):
+        dst = tmp_path / "out.pdf"
+
+        pdf_ops.remove_pages(THREE_PAGES, dst, [3, 1])
+
+        assert fingerprints(dst) == [source_fingerprints[1]]
+
+    def test_empty_list_keeps_every_page(
+        self,
+        tmp_path: Path,
+        source_fingerprints: list[str],
+    ):
+        dst = tmp_path / "out.pdf"
+
+        pdf_ops.remove_pages(THREE_PAGES, dst, [])
+
+        assert fingerprints(dst) == source_fingerprints
+
+    def test_removing_every_page_writes_an_empty_pdf(self, tmp_path: Path):
+        dst = tmp_path / "out.pdf"
+
+        pdf_ops.remove_pages(THREE_PAGES, dst, [1, 2, 3])
+
+        assert fingerprints(dst) == []
+
+    def test_keeps_document_info(self, tmp_path: Path):
+        dst = tmp_path / "out.pdf"
+
+        pdf_ops.remove_pages(THREE_PAGES, dst, [1])
+
+        assert "/Creator" in docinfo_keys(dst)
+
+    @pytest.mark.parametrize("bad_page", [0, -1])
+    def test_rejects_pages_below_one(self, tmp_path: Path, bad_page: int):
+        dst = tmp_path / "out.pdf"
+
+        with pytest.raises(ValueError, match="start at 1"):
+            pdf_ops.remove_pages(THREE_PAGES, dst, [1, bad_page])
+
+        assert not dst.exists()
+
+    def test_page_past_the_end_raises_and_writes_nothing(self, tmp_path: Path):
+        dst = tmp_path / "out.pdf"
+
+        with pytest.raises(IndexError):
+            pdf_ops.remove_pages(THREE_PAGES, dst, [99])
+
+        assert not dst.exists()
+
+
+class TestBuildPdfs:
+    def test_selects_and_orders_pages(
+        self,
+        tmp_path: Path,
+        source_fingerprints: list[str],
+    ):
+        dst = tmp_path / "out.pdf"
+
+        written = pdf_ops.build_pdfs(
+            THREE_PAGES,
+            [([PageSpec(3), PageSpec(1)], constant(dst))],
+        )
+
+        assert written == [dst]
+        assert fingerprints(dst) == [source_fingerprints[2], source_fingerprints[0]]
+
+    def test_rotates_only_the_requested_pages(self, tmp_path: Path):
+        dst = tmp_path / "out.pdf"
+
+        pdf_ops.build_pdfs(
+            THREE_PAGES,
+            [([PageSpec(1), PageSpec(2, 90), PageSpec(3, 180)], constant(dst))],
+        )
+
+        assert rotations(dst) == [0, 90, 180]
+
+    def test_writes_one_file_per_output_in_order(self, tmp_path: Path):
+        first = tmp_path / "first.pdf"
+        second = tmp_path / "second.pdf"
+        source = fingerprints(TWELVE_PAGES)
+
+        written = pdf_ops.build_pdfs(
+            TWELVE_PAGES,
+            [
+                ([PageSpec(p) for p in (1, 2, 3)], constant(first)),
+                ([PageSpec(p) for p in range(4, 13)], constant(second)),
+            ],
+        )
+
+        assert written == [first, second]
+        assert fingerprints(first) == source[:3]
+        assert fingerprints(second) == source[3:]
+
+    def test_empty_page_list_writes_a_zero_page_file(self, tmp_path: Path):
+        dst = tmp_path / "out.pdf"
+
+        pdf_ops.build_pdfs(THREE_PAGES, [([], constant(dst))])
+
+        assert fingerprints(dst) == []
+
+    def test_does_not_carry_over_document_info(self, tmp_path: Path):
+        dst = tmp_path / "out.pdf"
+        assert "/Creator" in docinfo_keys(THREE_PAGES)
+
+        pdf_ops.build_pdfs(THREE_PAGES, [([PageSpec(1)], constant(dst))])
+
+        assert "/Creator" not in docinfo_keys(dst)
+
+    def test_destination_is_not_requested_when_a_page_is_out_of_range(
+        self,
+        tmp_path: Path,
+    ):
+        requested: list[Path] = []
+
+        def make_dst() -> Path:
+            requested.append(tmp_path / "out.pdf")
+            return requested[-1]
+
+        with pytest.raises(IndexError):
+            pdf_ops.build_pdfs(THREE_PAGES, [([PageSpec(99)], make_dst)])
+
+        assert requested == []
+
+    @pytest.mark.parametrize("bad_page", [0, -1])
+    def test_rejects_pages_below_one_before_opening_anything(
+        self,
+        tmp_path: Path,
+        bad_page: int,
+    ):
+        requested: list[Path] = []
+
+        def make_dst() -> Path:
+            requested.append(tmp_path / "out.pdf")
+            return requested[-1]
+
+        with pytest.raises(ValueError, match="start at 1"):
+            pdf_ops.build_pdfs(
+                THREE_PAGES,
+                [([PageSpec(1)], make_dst), ([PageSpec(bad_page)], make_dst)],
+            )
+
+        assert requested == []
+
+
+class TestValidatePageOperations:
+    def test_returns_the_output_count(self):
+        operations = [{"page": 1}, {"page": 2}, {"page": 3}]
+
+        assert pdf_ops.validate_page_operations(operations, single_output=True) == 1
+
+    def test_gap_in_output_indices_counts_up_to_the_highest(self):
+        operations = [
+            {"page": 1, "doc": 0},
+            {"page": 2, "doc": 2},
+            {"page": 3, "doc": 0},
+        ]
+
+        count = pdf_ops.validate_page_operations(operations, single_output=False)
+
+        assert count == 3
+
+    def test_empty_operations_are_rejected(self):
+        with pytest.raises(ValueError, match="index is out of bounds"):
+            pdf_ops.validate_page_operations([], single_output=False)
+
+    def test_multiple_outputs_rejected_when_single_output_required(self):
+        operations = [{"page": 1, "doc": 0}, {"page": 2, "doc": 1}]
+
+        with pytest.raises(ValueError, match="Multiple output documents"):
+            pdf_ops.validate_page_operations(operations, single_output=True)
+
+    @pytest.mark.parametrize("doc", [-1, 2, 2**32])
+    def test_output_index_out_of_bounds(self, doc: int):
+        operations = [{"page": 1, "doc": 0}, {"page": 2, "doc": doc}]
+
+        with pytest.raises(ValueError, match="index is out of bounds"):
+            pdf_ops.validate_page_operations(operations, single_output=False)
+
+
+class TestDecrypt:
+    def test_needs_decrypt(self):
+        assert pdf_ops.needs_decrypt(ENCRYPTED) is True
+        assert pdf_ops.needs_decrypt(THREE_PAGES) is False
+
+    def test_pdf_that_opens_without_a_password_but_is_flagged_encrypted(self):
+        assert pdf_ops.needs_decrypt(SIGNED) is True
+
+    def test_decrypt_writes_an_unencrypted_copy(self, tmp_path: Path):
+        dst = tmp_path / "out.pdf"
+
+        result = pdf_ops.decrypt_pdf(ENCRYPTED, constant(dst), "test")
+
+        assert result == dst
+        assert pdf_ops.needs_decrypt(dst) is False
+
+    def test_wrong_password_raises_and_never_requests_a_destination(
+        self,
+        tmp_path: Path,
+    ):
+        requested: list[Path] = []
+
+        def make_dst() -> Path:
+            requested.append(tmp_path / "out.pdf")
+            return requested[-1]
+
+        with pytest.raises(pikepdf.PasswordError):
+            pdf_ops.decrypt_pdf(ENCRYPTED, make_dst, "wrong")
+
+        assert requested == []
+
+
+class TestPdfMerger:
+    def test_pages_are_appended_in_the_order_added(
+        self,
+        tmp_path: Path,
+        source_fingerprints: list[str],
+    ):
+        reordered = tmp_path / "reordered.pdf"
+        merged = tmp_path / "merged.pdf"
+        pdf_ops.build_pdfs(
+            THREE_PAGES,
+            [([PageSpec(3), PageSpec(1)], constant(reordered))],
+        )
+
+        with pdf_ops.PdfMerger() as merger:
+            merger.add(reordered)
+            merger.add(THREE_PAGES)
+            merger.save(merged)
+
+        assert fingerprints(merged) == [
+            source_fingerprints[2],
+            source_fingerprints[0],
+            *source_fingerprints,
+        ]
+
+    def test_output_version_is_at_least_the_highest_source_version(
+        self,
+        tmp_path: Path,
+    ):
+        merged = tmp_path / "merged.pdf"
+        with pikepdf.open(TWELVE_PAGES) as pdf:
+            source_versions = [pdf.pdf_version]
+        with pikepdf.open(THREE_PAGES) as pdf:
+            source_versions.append(pdf.pdf_version)
+
+        with pdf_ops.PdfMerger() as merger:
+            merger.add(TWELVE_PAGES)
+            merger.add(THREE_PAGES)
+            merger.save(merged)
+
+        with pikepdf.open(merged) as pdf:
+            assert pdf.pdf_version >= max(source_versions)
+
+    def test_unreadable_source_raises_so_the_caller_can_skip_it(
+        self,
+        tmp_path: Path,
+    ):
+        garbage = tmp_path / "garbage.pdf"
+        garbage.write_bytes(b"not a pdf")
+
+        with pdf_ops.PdfMerger() as merger:
+            with pytest.raises(pikepdf.PdfError):
+                merger.add(garbage)
+
+
+def test_pdf_ops_imports_only_the_standard_library_and_pikepdf():
+    tree = ast.parse(Path(pdf_ops.__file__).read_text())
+    imported: set[str] = set()
+    for node in ast.walk(tree):
+        if isinstance(node, ast.Import):
+            imported.update(alias.name.split(".")[0] for alias in node.names)
+        elif isinstance(node, ast.ImportFrom) and node.level == 0 and node.module:
+            imported.add(node.module.split(".")[0])
+
+    coupled = imported & {
+        "django",
+        "celery",
+        "documents",
+        "paperless",
+        "paperless_mail",
+        "paperless_ai",
+    }
+    assert not coupled
@@ -1192,6 +1192,7 @@ class DocumentViewSet(
                    "version_label",
                    "root_document_id",
                    "version_index",
+                    "page_count",
                ),
            ),
            "tags",
@@ -1271,13 +1272,16 @@ class DocumentViewSet(
        if (
            "version" not in request.query_params
            or not isinstance(response.data, dict)
-            or "content" not in response.data
+            or not ({"content", "page_count"} & response.data.keys())
        ):
            return response

        root_doc = self.get_object()
        content_doc = self._resolve_file_doc(root_doc, request)
-        response.data["content"] = content_doc.content or ""
+        if "content" in response.data:
+            response.data["content"] = content_doc.content or ""
+        if "page_count" in response.data:
+            response.data["page_count"] = content_doc.page_count
        return response

    def update(self, request, *args, **kwargs):
@@ -2,7 +2,7 @@ msgid ""
 msgstr ""
 "Project-Id-Version: paperless-ngx\n"
 "Report-Msgid-Bugs-To: \n"
-"POT-Creation-Date: 2026-09-23 19:00+0000\n"
+"POT-Creation-Date: 2026-10-02 23:09+0000\n"
 "PO-Revision-Date: 2022-02-17 04:17\n"
 "Last-Translator: \n"
 "Language-Team: English\n"
@@ -1632,7 +1632,7 @@ msgid "workflow runs"
 msgstr ""

 #: documents/serialisers.py:514 documents/serialisers.py:871
-#: documents/serialisers.py:2885 documents/views.py:342 documents/views.py:2725
+#: documents/serialisers.py:2895 documents/views.py:342 documents/views.py:2729
 #: paperless_mail/serialisers.py:156
 msgid "Insufficient permissions."
 msgstr ""
@@ -1641,39 +1641,39 @@ msgstr ""
 msgid "Invalid color."
 msgstr ""

-#: documents/serialisers.py:2352
+#: documents/serialisers.py:2362
 #, python-format
 msgid "File type %(type)s not supported"
 msgstr ""

-#: documents/serialisers.py:2396
+#: documents/serialisers.py:2406
 #, python-format
 msgid "Custom field id must be an integer: %(id)s"
 msgstr ""

-#: documents/serialisers.py:2403
+#: documents/serialisers.py:2413
 #, python-format
 msgid "Custom field with id %(id)s does not exist"
 msgstr ""

-#: documents/serialisers.py:2420 documents/serialisers.py:2430
+#: documents/serialisers.py:2430 documents/serialisers.py:2440
 msgid ""
 "Custom fields must be a list of integers or an object mapping ids to values."
 msgstr ""

-#: documents/serialisers.py:2425
+#: documents/serialisers.py:2435
 msgid "Some custom fields don't exist or were specified twice."
 msgstr ""

-#: documents/serialisers.py:2572
+#: documents/serialisers.py:2582
 msgid "Invalid variable detected."
 msgstr ""

-#: documents/serialisers.py:2941
+#: documents/serialisers.py:2951
 msgid "Duplicate document identifiers are not allowed."
 msgstr ""

-#: documents/serialisers.py:2971 documents/views.py:4780
+#: documents/serialisers.py:2981 documents/views.py:4784
 #, python-format
 msgid "Documents not found: %(ids)s"
 msgstr ""
@@ -1941,40 +1941,40 @@ msgstr ""
 msgid "Unable to parse URI {value}"
 msgstr ""

-#: documents/views.py:335 documents/views.py:2722
+#: documents/views.py:335 documents/views.py:2726
 msgid "Invalid more_like_id"
 msgstr ""

-#: documents/views.py:1669
+#: documents/views.py:1673
 msgid "Invalid AI configuration."
 msgstr ""

-#: documents/views.py:1680
+#: documents/views.py:1684
 msgid "AI backend request timed out."
 msgstr ""

-#: documents/views.py:1692
+#: documents/views.py:1696
 msgid "AI backend rejected the request. Check logs for details."
 msgstr ""

-#: documents/views.py:2547 documents/views.py:2863
+#: documents/views.py:2551 documents/views.py:2867
 msgid "Specify only one of text, title_search, query, or more_like_id."
 msgstr ""

-#: documents/views.py:4793
+#: documents/views.py:4797
 #, python-format
 msgid "Insufficient permissions to share document %(id)s."
 msgstr ""

-#: documents/views.py:4839
+#: documents/views.py:4843
 msgid "Bundle is already being processed."
 msgstr ""

-#: documents/views.py:4903
+#: documents/views.py:4907
 msgid "The share link bundle is still being prepared. Please try again later."
 msgstr ""

-#: documents/views.py:4917
+#: documents/views.py:4921
 msgid "The share link bundle is unavailable."
 msgstr ""

@@ -2262,155 +2262,155 @@ msgstr ""
 msgid "paperless application settings"
 msgstr ""

-#: paperless/settings/__init__.py:559
+#: paperless/settings/__init__.py:560
 msgid "English (US)"
 msgstr ""

-#: paperless/settings/__init__.py:560
+#: paperless/settings/__init__.py:561
 msgid "Arabic"
 msgstr ""

-#: paperless/settings/__init__.py:561
+#: paperless/settings/__init__.py:562
 msgid "Afrikaans"
 msgstr ""

-#: paperless/settings/__init__.py:562
+#: paperless/settings/__init__.py:563
 msgid "Belarusian"
 msgstr ""

-#: paperless/settings/__init__.py:563
+#: paperless/settings/__init__.py:564
 msgid "Bulgarian"
 msgstr ""

-#: paperless/settings/__init__.py:564
+#: paperless/settings/__init__.py:565
 msgid "Catalan"
 msgstr ""

-#: paperless/settings/__init__.py:565
+#: paperless/settings/__init__.py:566
 msgid "Czech"
 msgstr ""

-#: paperless/settings/__init__.py:566
+#: paperless/settings/__init__.py:567
 msgid "Danish"
 msgstr ""

-#: paperless/settings/__init__.py:567
+#: paperless/settings/__init__.py:568
 msgid "German"
 msgstr ""

-#: paperless/settings/__init__.py:568
+#: paperless/settings/__init__.py:569
 msgid "Greek"
 msgstr ""

-#: paperless/settings/__init__.py:569
+#: paperless/settings/__init__.py:570
 msgid "English (GB)"
 msgstr ""

-#: paperless/settings/__init__.py:570
+#: paperless/settings/__init__.py:571
 msgid "Spanish"
 msgstr ""

-#: paperless/settings/__init__.py:571
+#: paperless/settings/__init__.py:572
 msgid "Persian"
 msgstr ""

-#: paperless/settings/__init__.py:572
+#: paperless/settings/__init__.py:573
 msgid "Finnish"
 msgstr ""

-#: paperless/settings/__init__.py:573
+#: paperless/settings/__init__.py:574
 msgid "French"
 msgstr ""

-#: paperless/settings/__init__.py:574
+#: paperless/settings/__init__.py:575
 msgid "Hungarian"
 msgstr ""

-#: paperless/settings/__init__.py:575
+#: paperless/settings/__init__.py:576
 msgid "Indonesian"
 msgstr ""

-#: paperless/settings/__init__.py:576
+#: paperless/settings/__init__.py:577
 msgid "Italian"
 msgstr ""

-#: paperless/settings/__init__.py:577
+#: paperless/settings/__init__.py:578
 msgid "Japanese"
 msgstr ""

-#: paperless/settings/__init__.py:578
+#: paperless/settings/__init__.py:579
 msgid "Korean"
 msgstr ""

-#: paperless/settings/__init__.py:579
+#: paperless/settings/__init__.py:580
 msgid "Luxembourgish"
 msgstr ""

-#: paperless/settings/__init__.py:580
+#: paperless/settings/__init__.py:581
 msgid "Norwegian"
 msgstr ""

-#: paperless/settings/__init__.py:581
+#: paperless/settings/__init__.py:582
 msgid "Dutch"
 msgstr ""

-#: paperless/settings/__init__.py:582
+#: paperless/settings/__init__.py:583
 msgid "Polish"
 msgstr ""

-#: paperless/settings/__init__.py:583
+#: paperless/settings/__init__.py:584
 msgid "Portuguese (Brazil)"
 msgstr ""

-#: paperless/settings/__init__.py:584
+#: paperless/settings/__init__.py:585
 msgid "Portuguese"
 msgstr ""

-#: paperless/settings/__init__.py:585
+#: paperless/settings/__init__.py:586
 msgid "Romanian"
 msgstr ""

-#: paperless/settings/__init__.py:586
+#: paperless/settings/__init__.py:587
 msgid "Russian"
 msgstr ""

-#: paperless/settings/__init__.py:587
+#: paperless/settings/__init__.py:588
 msgid "Slovak"
 msgstr ""

-#: paperless/settings/__init__.py:588
+#: paperless/settings/__init__.py:589
 msgid "Slovenian"
 msgstr ""

-#: paperless/settings/__init__.py:589
+#: paperless/settings/__init__.py:590
 msgid "Serbian"
 msgstr ""

-#: paperless/settings/__init__.py:590
+#: paperless/settings/__init__.py:591
 msgid "Swedish"
 msgstr ""

-#: paperless/settings/__init__.py:591
+#: paperless/settings/__init__.py:592
 msgid "Turkish"
 msgstr ""

-#: paperless/settings/__init__.py:592
+#: paperless/settings/__init__.py:593
 msgid "Ukrainian"
 msgstr ""

-#: paperless/settings/__init__.py:593
+#: paperless/settings/__init__.py:594
 msgid "Vietnamese"
 msgstr ""

-#: paperless/settings/__init__.py:594
+#: paperless/settings/__init__.py:595
 msgid "Chinese Simplified"
 msgstr ""

-#: paperless/settings/__init__.py:595
+#: paperless/settings/__init__.py:596
 msgid "Chinese Traditional"
 msgstr ""

-#: paperless/urls.py:435
+#: paperless/urls.py:438
 msgid "Paperless-ngx administration"
 msgstr ""

@@ -22,8 +22,6 @@ from pathlib import Path
 from typing import TYPE_CHECKING
 from typing import Self

-from bleach import clean
-from bleach import linkify
 from django.conf import settings
 from django.utils import timezone
 from django.utils.timezone import is_naive
@@ -38,6 +36,9 @@ from humanize import naturalsize
 from imap_tools import MailAttachment
 from imap_tools import MailMessage
 from tika_client import TikaClient
+from turbohtml.clean import Linkify
+from turbohtml.clean import linkify
+from turbohtml.migration.bleach import clean

 from documents.parsers import ParseError
 from documents.parsers import make_thumbnail_from_pdf
@@ -58,12 +59,6 @@ _SUPPORTED_MIME_TYPES: dict[str, str] = {
    "message/rfc822": ".eml",
 }

-# Bleach's email-address linkifier uses a superlinear regular expression. Keep
-# email linkification for ordinary headers and short messages, but never run it
-# over an unbounded attacker-controlled field. URL linkification remains enabled
-# for longer text.
-_MAX_EMAIL_LINKIFY_LENGTH = 2048
-

 class MailDocumentParser:
    """Parse .eml email files for Paperless-ngx.
@@ -633,10 +628,7 @@ class MailDocumentParser:
                text = str(text)
            text = escape(text)
            text = clean(text)
-            text = linkify(
-                text,
-                parse_email="@" in text and len(text) <= _MAX_EMAIL_LINKIFY_LENGTH,
-            )
+            text = linkify(text, Linkify(parse_email=True))
            text = text.replace("\n", "<br>")
            return text

@@ -291,7 +291,7 @@ if _CHANNELS_BACKEND.startswith("channels_redis."):
 ###############################################################################

 EMAIL_HOST: Final[str] = os.getenv("PAPERLESS_EMAIL_HOST", "localhost")
-EMAIL_PORT: Final[int] = int(os.getenv("PAPERLESS_EMAIL_PORT", 25))
+EMAIL_PORT: Final[int] = get_int_from_env("PAPERLESS_EMAIL_PORT", 25)
 EMAIL_HOST_USER: Final[str] = os.getenv("PAPERLESS_EMAIL_HOST_USER", "")
 EMAIL_HOST_PASSWORD: Final[str] = os.getenv("PAPERLESS_EMAIL_HOST_PASSWORD", "")
 DEFAULT_FROM_EMAIL: Final[str] = os.getenv("PAPERLESS_EMAIL_FROM", EMAIL_HOST_USER)
@@ -378,8 +378,9 @@ ACCOUNT_SESSION_REMEMBER = get_bool_from_env(
    "True",
 )
 SESSION_EXPIRE_AT_BROWSER_CLOSE = not ACCOUNT_SESSION_REMEMBER
-SESSION_COOKIE_AGE = int(
-    os.getenv("PAPERLESS_SESSION_COOKIE_AGE", 60 * 60 * 24 * 7 * 3),
+SESSION_COOKIE_AGE = get_int_from_env(
+    "PAPERLESS_SESSION_COOKIE_AGE",
+    60 * 60 * 24 * 7 * 3,
 )
 # https://docs.djangoproject.com/en/5.1/ref/settings/#std-setting-SESSION_ENGINE
 SESSION_ENGINE = "django.contrib.sessions.backends.cached_db"
@@ -611,8 +612,8 @@ USE_TZ = True

 LOGGING_DIR.mkdir(parents=True, exist_ok=True)

-LOGROTATE_MAX_SIZE = os.getenv("PAPERLESS_LOGROTATE_MAX_SIZE", 1024 * 1024)
-LOGROTATE_MAX_BACKUPS = os.getenv("PAPERLESS_LOGROTATE_MAX_BACKUPS", 20)
+LOGROTATE_MAX_SIZE = get_int_from_env("PAPERLESS_LOGROTATE_MAX_SIZE", 1024 * 1024)
+LOGROTATE_MAX_BACKUPS = get_int_from_env("PAPERLESS_LOGROTATE_MAX_BACKUPS", 20)

 LOGGING = {
    "version": 1,
@@ -796,7 +797,7 @@ def default_threads_per_worker(task_workers) -> int:
        return 1


-THREADS_PER_WORKER = os.getenv(
+THREADS_PER_WORKER = get_int_from_env(
    "PAPERLESS_THREADS_PER_WORKER",
    default_threads_per_worker(CELERY_WORKER_CONCURRENCY),
 )
@@ -817,9 +818,9 @@ IGNORABLE_FILES: Final[list[str]] = [
    "Thumbs.db",
 ]

-CONSUMER_POLLING_INTERVAL = float(os.getenv("PAPERLESS_CONSUMER_POLLING_INTERVAL", 0))
+CONSUMER_POLLING_INTERVAL = get_float_from_env("PAPERLESS_CONSUMER_POLLING_INTERVAL", 0)

-CONSUMER_STABILITY_DELAY = float(os.getenv("PAPERLESS_CONSUMER_STABILITY_DELAY", 5))
+CONSUMER_STABILITY_DELAY = get_float_from_env("PAPERLESS_CONSUMER_STABILITY_DELAY", 5)

 CONSUMER_DELETE_DUPLICATES = get_bool_from_env("PAPERLESS_CONSUMER_DELETE_DUPLICATES")

@@ -735,33 +735,24 @@ class TestParser:

            assert expected_html == actual_html

-    def test_mail_to_html_bounds_email_linkification(
+    def test_mail_to_html_linkifies_email_in_long_text(
        self,
        mail_parser: MailDocumentParser,
    ) -> None:
        mail = mock.Mock(
-            subject="sender@example.com",
+            subject="",
            from_values=None,
            to_values=[],
            cc_values=[],
            bcc_values=[],
            attachments=[],
            date=timezone.now(),
-            text=("a." * 1500) + "@example.com",
+            text=("a." * 1500) + " sender@example.com",
        )

-        with mock.patch(
-            "paperless.parsers.mail.linkify",
-            side_effect=lambda text, **kwargs: text,
-        ) as mock_linkify:
-            mail_parser.mail_to_html(mail)
+        html_file = mail_parser.mail_to_html(mail)

-        parse_email_by_text = {
-            call.args[0]: call.kwargs["parse_email"]
-            for call in mock_linkify.call_args_list
-        }
-        assert parse_email_by_text["sender@example.com"] is True
-        assert parse_email_by_text[mail.text] is False
+        assert 'href="mailto:sender@example.com"' in html_file.read_text()

    def test_generate_pdf_from_mail(
        self,
@@ -0,0 +1,68 @@
+import time
+
+from allauth.mfa import app_settings as mfa_settings
+from allauth.mfa.totp.internal import auth as totp_auth
+from django.test import TestCase
+from django.urls import reverse
+
+from paperless_testing.factories import UserFactory
+
+
+class TestAdminAuth(TestCase):
+    def test_admin_login_redirects_to_allauth(self):
+        user = UserFactory(staff=True, password="testpassword")
+        admin_url = reverse("admin:index")
+        login_url = reverse("admin:login")
+        expected_url = f"{reverse('account_login')}?next={admin_url}"
+
+        response = self.client.get(login_url, {"next": admin_url})
+        self.assertRedirects(response, expected_url)
+
+        response = self.client.post(
+            login_url,
+            {"username": user.username, "password": "testpassword", "next": admin_url},
+        )
+        self.assertRedirects(response, expected_url)
+        self.assertNotIn("_auth_user_id", self.client.session)
+
+    def test_admin_access_requires_totp_for_enrolled_staff(self):
+        user = UserFactory(staff=True, password="testpassword")
+        secret = totp_auth.generate_totp_secret()
+        totp_auth.TOTP.activate(user, secret)
+        admin_url = reverse("admin:index")
+        mfa_url = reverse("mfa_authenticate")
+
+        response = self.client.post(
+            reverse("account_login"),
+            {"login": user.username, "password": "testpassword", "next": admin_url},
+        )
+        self.assertRedirects(response, mfa_url)
+        self.assertNotIn("_auth_user_id", self.client.session)
+        self.assertRedirects(
+            self.client.get(admin_url),
+            f"{reverse('admin:login')}?next={admin_url}",
+            fetch_redirect_response=False,
+        )
+
+        response = self.client.post(mfa_url, {"code": "invalid"})
+        self.assertEqual(response.status_code, 200)
+        self.assertNotIn("_auth_user_id", self.client.session)
+
+        code = totp_auth.format_hotp_value(
+            totp_auth.hotp_value(secret, int(time.time()) // mfa_settings.TOTP_PERIOD),
+        )
+        response = self.client.post(mfa_url, {"code": code})
+        self.assertRedirects(response, admin_url)
+        self.assertEqual(self.client.session["_auth_user_id"], str(user.pk))
+
+    def test_staff_without_totp_can_still_log_in(self):
+        user = UserFactory(staff=True, password="testpassword")
+        admin_url = reverse("admin:index")
+
+        response = self.client.post(
+            reverse("account_login"),
+            {"login": user.username, "password": "testpassword", "next": admin_url},
+        )
+
+        self.assertRedirects(response, admin_url)
+        self.assertEqual(self.client.session["_auth_user_id"], str(user.pk))
@@ -1,4 +1,5 @@
 from allauth.account import views as allauth_account_views
+from allauth.account.decorators import secure_admin_login
 from allauth.mfa.base import views as allauth_mfa_views
 from allauth.socialaccount import views as allauth_social_account_views
 from allauth.urls import build_provider_urlpatterns
@@ -68,6 +69,8 @@ from paperless_mail.views import MailRuleViewSet
 from paperless_mail.views import OauthCallbackView
 from paperless_mail.views import ProcessedMailViewSet

+admin.site.login = secure_admin_login(admin.site.login)
+
 api_router = DefaultRouter()
 api_router.register(r"correspondents", CorrespondentViewSet)
 api_router.register(r"document_types", DocumentTypeViewSet)
@@ -297,7 +300,7 @@ urlpatterns = [
    ),
    re_path(r"^share/(?P<slug>\w+)/?$", SharedLinkView.as_view()),
    re_path(r"^favicon.ico$", FaviconView.as_view(), name="favicon"),
-    re_path(r"admin/", admin.site.urls),
+    re_path(r"^admin/", admin.site.urls),
    re_path(
        r"^fetch/",
        include(
@@ -2,14 +2,14 @@ from __future__ import annotations

 import factory
 from django.utils import timezone
-from factory.django import DjangoModelFactory

 from paperless_mail.models import MailAccount
 from paperless_mail.models import MailRule
 from paperless_mail.models import ProcessedMail
+from paperless_testing.typed_factory import TypedModelFactory


-class MailAccountFactory(DjangoModelFactory[MailAccount]):
+class MailAccountFactory(TypedModelFactory[MailAccount]):
    class Meta:
        model = MailAccount

@@ -24,7 +24,7 @@ class MailAccountFactory(DjangoModelFactory[MailAccount]):
    is_token = False


-class MailRuleFactory(DjangoModelFactory[MailRule]):
+class MailRuleFactory(TypedModelFactory[MailRule]):
    class Meta:
        model = MailRule

@@ -44,7 +44,7 @@ class MailRuleFactory(DjangoModelFactory[MailRule]):
    stop_processing = False


-class ProcessedMailFactory(DjangoModelFactory[ProcessedMail]):
+class ProcessedMailFactory(TypedModelFactory[ProcessedMail]):
    class Meta:
        model = ProcessedMail

@@ -5,8 +5,7 @@ Factory-boy factories for documents app models.
 from __future__ import annotations

 import factory
-from django.contrib.auth import get_user_model
-from factory.django import DjangoModelFactory
+from django.contrib.auth.models import User

 from documents.models import Correspondent
 from documents.models import Document
@@ -15,11 +14,10 @@ from documents.models import MatchingModel
 from documents.models import PaperlessTask
 from documents.models import StoragePath
 from documents.models import Tag
-
-UserModelT = get_user_model()
+from paperless_testing.typed_factory import TypedModelFactory


-class CorrespondentFactory(DjangoModelFactory[Correspondent]):
+class CorrespondentFactory(TypedModelFactory[Correspondent]):
    class Meta:
        model = Correspondent

@@ -28,7 +26,7 @@ class CorrespondentFactory(DjangoModelFactory[Correspondent]):
    matching_algorithm = MatchingModel.MATCH_NONE


-class DocumentTypeFactory(DjangoModelFactory[DocumentType]):
+class DocumentTypeFactory(TypedModelFactory[DocumentType]):
    class Meta:
        model = DocumentType

@@ -37,7 +35,7 @@ class DocumentTypeFactory(DjangoModelFactory[DocumentType]):
    matching_algorithm = MatchingModel.MATCH_NONE


-class TagFactory(DjangoModelFactory[Tag]):
+class TagFactory(TypedModelFactory[Tag]):
    class Meta:
        model = Tag

@@ -47,7 +45,7 @@ class TagFactory(DjangoModelFactory[Tag]):
    is_inbox_tag = False


-class StoragePathFactory(DjangoModelFactory[StoragePath]):
+class StoragePathFactory(TypedModelFactory[StoragePath]):
    class Meta:
        model = StoragePath

@@ -59,7 +57,7 @@ class StoragePathFactory(DjangoModelFactory[StoragePath]):
    matching_algorithm = MatchingModel.MATCH_NONE


-class DocumentFactory(DjangoModelFactory[Document]):
+class DocumentFactory(TypedModelFactory[Document]):
    class Meta:
        model = Document

@@ -71,9 +69,9 @@ class DocumentFactory(DjangoModelFactory[Document]):
    storage_path = None


-class UserFactory(DjangoModelFactory[UserModelT]):
+class UserFactory(TypedModelFactory[User]):
    class Meta:
-        model = UserModelT
+        model = User

    username = factory.Sequence(lambda n: f"user{n}")
    is_staff = False
@@ -88,7 +86,7 @@ class UserFactory(DjangoModelFactory[UserModelT]):
        staff = factory.Trait(is_staff=True)


-class PaperlessTaskFactory(DjangoModelFactory[PaperlessTask]):
+class PaperlessTaskFactory(TypedModelFactory[PaperlessTask]):
    class Meta:
        model = PaperlessTask

@@ -0,0 +1,20 @@
+"""
+A DjangoModelFactory base whose calls are typed as the model they build.
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+from typing import Any
+from typing import TypeVar
+
+from factory.django import DjangoModelFactory
+
+T = TypeVar("T")
+
+
+class TypedModelFactory(DjangoModelFactory[T]):
+    if TYPE_CHECKING:
+        # factory-boy leaves Factory() unannotated, so mypy takes it to build a
+        # factory instance. At runtime it builds the model.
+        def __new__(cls, *args: Any, **kwargs: Any) -> T: ...  # type: ignore[misc]