Fix: annotate effective content in the filters rather than relying on callers

Auto translate strings
Enhancement: Apply AI suggestions workflow action (#13639 )
2026-08-27 05:03:20 +00:00 · 2026-08-26 09:54:18 -07:00 · 2026-08-26 16:50:29 +00:00 · 2026-08-26 16:41:21 +00:00 · 2026-08-26 14:29:05 +00:00 · 2026-08-26 14:27:31 +00:00
28 changed files with 2265 additions and 749 deletions
@@ -667,6 +667,35 @@ The action takes no options, its presence is what enables remote OCR for a match
 If the remote engine is not configured, or does not support the document's file type, the document is
 processed locally instead and a warning is written to the log.

+##### Apply AI Suggestions {#workflow-action-apply-ai-suggestions}
+
+"Apply AI Suggestions" actions ask the configured AI service for title and metadata suggestions,
+the same as the AI suggestions shown on the document detail page, except applied automatically and in bulk.
+It requires [AI features](configuration.md#ai) to be enabled. You can specify:
+
+- Which suggestions to apply: title, tags, correspondent, document type, storage path and / or created
+  date. Suggestions for fields you did not select are discarded.
+- Whether to create missing items. By default only tags, correspondents and document types that
+  already exist are assigned and any other suggestion is dropped. With this enabled, suggested items
+  that do not exist are created. Storage paths are never created.
+- Whether to overwrite existing values. By default a field is only filled in if it is currently empty.
+  Note that documents almost always already have a title and created date, so if you select those you
+  will usually want to enable this too. Tags are an exception: suggested tags are always added and
+  never replace the document's existing tags.
+
+The action works with every trigger **except Consumption Started**, because suggestions are made from
+the document's text, which does not exist until after the document has been processed.
+
+Because the query to the AI service is slow, the action is queued and runs in the background rather
+than as part of the workflow run itself. The document is updated once the suggestions come back.
+
+!!! warning
+
+    Every matching document results in a query to the AI service, which may incur costs and have privacy
+    implications. Queries can be slow, so a workflow matching a large number of documents can occupy the
+    task queue, and delay consumption of new documents, etc. Consider narrowing the trigger filters,
+    running in small batches and / or increasing workers.
+
 #### Workflow placeholders

 Titles and webhook payloads can be generated by workflows using [Jinja templates](https://jinja.palletsprojects.com/en/3.1.x/templates/).
@@ -1620,6 +1620,10 @@
          <context context-type="sourcefile">src/app/components/admin/settings/settings.component.ts</context>
          <context context-type="linenumber">99</context>
        </context-group>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
+          <context context-type="linenumber">173</context>
+        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/document-detail/document-detail.component.html</context>
          <context context-type="linenumber">174</context>
@@ -1651,6 +1655,10 @@
          <context context-type="sourcefile">src/app/components/admin/settings/settings.component.ts</context>
          <context context-type="linenumber">100</context>
        </context-group>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
+          <context context-type="linenumber">177</context>
+        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/document-detail/document-detail.component.html</context>
          <context context-type="linenumber">178</context>
@@ -1682,6 +1690,10 @@
          <context context-type="sourcefile">src/app/components/admin/settings/settings.component.ts</context>
          <context context-type="linenumber">101</context>
        </context-group>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
+          <context context-type="linenumber">181</context>
+        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/document-detail/document-detail.component.html</context>
          <context context-type="linenumber">182</context>
@@ -1713,6 +1725,10 @@
          <context context-type="sourcefile">src/app/components/app-frame/app-frame.component.html</context>
          <context context-type="linenumber">211</context>
        </context-group>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
+          <context context-type="linenumber">169</context>
+        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/input/tags/tags.component.ts</context>
          <context context-type="linenumber">86</context>
@@ -2087,7 +2103,7 @@
        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">95</context>
+          <context context-type="linenumber">96</context>
        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/share-link-bundle-manage-dialog/share-link-bundle-manage-dialog.component.html</context>
@@ -2402,7 +2418,7 @@
        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">80</context>
+          <context context-type="linenumber">81</context>
        </context-group>
      </trans-unit>
      <trans-unit id="228975554646076615" datatype="html">
@@ -2462,7 +2478,7 @@
        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">122</context>
+          <context context-type="linenumber">123</context>
        </context-group>
      </trans-unit>
      <trans-unit id="235571817610183244" datatype="html">
@@ -2473,7 +2489,7 @@
        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">84</context>
+          <context context-type="linenumber">85</context>
        </context-group>
      </trans-unit>
      <trans-unit id="526966086395145275" datatype="html">
@@ -2484,7 +2500,7 @@
        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">76</context>
+          <context context-type="linenumber">77</context>
        </context-group>
      </trans-unit>
      <trans-unit id="6520092939021340973" datatype="html">
@@ -4332,7 +4348,7 @@
        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">91</context>
+          <context context-type="linenumber">92</context>
        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/document-list/document-list.component.html</context>
@@ -5322,7 +5338,7 @@
        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">137</context>
+          <context context-type="linenumber">138</context>
        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/profile-edit-dialog/profile-edit-dialog.component.html</context>
@@ -5954,18 +5970,67 @@
          <context context-type="linenumber">461,462</context>
        </context-group>
      </trans-unit>
+      <trans-unit id="4231754882471903566" datatype="html">
+        <source>The document will be sent to the configured AI service for suggestions. Consider costs and privacy.</source>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.html</context>
+          <context context-type="linenumber">468,469</context>
+        </context-group>
+      </trans-unit>
+      <trans-unit id="3473589590476379802" datatype="html">
+        <source>Apply suggestions for</source>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.html</context>
+          <context context-type="linenumber">471,472</context>
+        </context-group>
+      </trans-unit>
+      <trans-unit id="6859826074028847234" datatype="html">
+        <source>Suggestions for fields that are not selected are discarded.</source>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.html</context>
+          <context context-type="linenumber">476,477</context>
+        </context-group>
+      </trans-unit>
+      <trans-unit id="1137167921211640884" datatype="html">
+        <source>Create missing items</source>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.html</context>
+          <context context-type="linenumber">486,487</context>
+        </context-group>
+      </trans-unit>
+      <trans-unit id="449779688394460071" datatype="html">
+        <source>Create suggested tags, correspondents and document types that do not exist yet.</source>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.html</context>
+          <context context-type="linenumber">488,489</context>
+        </context-group>
+      </trans-unit>
+      <trans-unit id="5192325026233872238" datatype="html">
+        <source>Overwrite existing values</source>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.html</context>
+          <context context-type="linenumber">496,497</context>
+        </context-group>
+      </trans-unit>
+      <trans-unit id="9010697250406492150" datatype="html">
+        <source>Apply suggestions even if the document already has a value. Tags are always added, never replaced.</source>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.html</context>
+          <context context-type="linenumber">498,499</context>
+        </context-group>
+      </trans-unit>
      <trans-unit id="4626030417479279989" datatype="html">
        <source>Consume Folder</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">72</context>
+          <context context-type="linenumber">73</context>
        </context-group>
      </trans-unit>
      <trans-unit id="3553216189604488439" datatype="html">
        <source>Modified</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">99</context>
+          <context context-type="linenumber">100</context>
        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/data/document.ts</context>
@@ -5976,63 +6041,63 @@
        <source>Custom Field</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">103</context>
+          <context context-type="linenumber">104</context>
        </context-group>
      </trans-unit>
      <trans-unit id="8696908693776094667" datatype="html">
        <source>Consumption Started</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">110</context>
+          <context context-type="linenumber">111</context>
        </context-group>
      </trans-unit>
      <trans-unit id="7858311467093621703" datatype="html">
        <source>Document Added</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">114</context>
+          <context context-type="linenumber">115</context>
        </context-group>
      </trans-unit>
      <trans-unit id="7955486237346046731" datatype="html">
        <source>Document Updated</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">118</context>
+          <context context-type="linenumber">119</context>
        </context-group>
      </trans-unit>
      <trans-unit id="5502398334173581061" datatype="html">
        <source>Assignment</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">129</context>
+          <context context-type="linenumber">130</context>
        </context-group>
      </trans-unit>
      <trans-unit id="6234812824772766804" datatype="html">
        <source>Removal</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">133</context>
+          <context context-type="linenumber">134</context>
        </context-group>
      </trans-unit>
      <trans-unit id="4206419737792796794" datatype="html">
        <source>Webhook</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">141</context>
+          <context context-type="linenumber">142</context>
        </context-group>
      </trans-unit>
      <trans-unit id="4824906895380506720" datatype="html">
        <source>Password removal</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">145</context>
+          <context context-type="linenumber">146</context>
        </context-group>
      </trans-unit>
      <trans-unit id="2048798344356757326" datatype="html">
        <source>Move to trash</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">149</context>
+          <context context-type="linenumber">150</context>
        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/document-detail/document-detail.component.ts</context>
@@ -6047,116 +6112,157 @@
        <source>Remote OCR</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">153</context>
+          <context context-type="linenumber">154</context>
        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/data/paperless-config.ts</context>
          <context context-type="linenumber">58</context>
        </context-group>
      </trans-unit>
+      <trans-unit id="4101630920511881483" datatype="html">
+        <source>Apply AI suggestions</source>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
+          <context context-type="linenumber">158</context>
+        </context-group>
+      </trans-unit>
+      <trans-unit id="5701618810648052610" datatype="html">
+        <source>Title</source>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
+          <context context-type="linenumber">165</context>
+        </context-group>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/document-detail/document-detail.component.html</context>
+          <context context-type="linenumber">167</context>
+        </context-group>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/document-list/document-list.component.html</context>
+          <context context-type="linenumber">218,219</context>
+        </context-group>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/document-list/filter-editor/filter-editor.component.ts</context>
+          <context context-type="linenumber">197</context>
+        </context-group>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/data/document.ts</context>
+          <context context-type="linenumber">30</context>
+        </context-group>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/data/document.ts</context>
+          <context context-type="linenumber">90</context>
+        </context-group>
+      </trans-unit>
+      <trans-unit id="5224451373257709572" datatype="html">
+        <source>Created date</source>
+        <context-group purpose="location">
+          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
+          <context context-type="linenumber">185</context>
+        </context-group>
+      </trans-unit>
      <trans-unit id="4522609911791833187" datatype="html">
        <source>Has any of these tags</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">222</context>
+          <context context-type="linenumber">254</context>
        </context-group>
      </trans-unit>
      <trans-unit id="4166903555074156852" datatype="html">
        <source>Has all of these tags</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">229</context>
+          <context context-type="linenumber">261</context>
        </context-group>
      </trans-unit>
      <trans-unit id="6624363795312783141" datatype="html">
        <source>Does not have these tags</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">236</context>
+          <context context-type="linenumber">268</context>
        </context-group>
      </trans-unit>
      <trans-unit id="7168528512669831184" datatype="html">
        <source>Has any of these correspondents</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">243</context>
+          <context context-type="linenumber">275</context>
        </context-group>
      </trans-unit>
      <trans-unit id="5281365940563983618" datatype="html">
        <source>Has correspondent</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">251</context>
+          <context context-type="linenumber">283</context>
        </context-group>
      </trans-unit>
      <trans-unit id="6884498632428600393" datatype="html">
        <source>Does not have correspondents</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">259</context>
+          <context context-type="linenumber">291</context>
        </context-group>
      </trans-unit>
      <trans-unit id="4806713133917046341" datatype="html">
        <source>Has document type</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">267</context>
+          <context context-type="linenumber">299</context>
        </context-group>
      </trans-unit>
      <trans-unit id="8801397520369995032" datatype="html">
        <source>Has any of these document types</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">275</context>
+          <context context-type="linenumber">307</context>
        </context-group>
      </trans-unit>
      <trans-unit id="1507843981661822403" datatype="html">
        <source>Does not have document types</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">283</context>
+          <context context-type="linenumber">315</context>
        </context-group>
      </trans-unit>
      <trans-unit id="4277260190522078330" datatype="html">
        <source>Has storage path</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">291</context>
+          <context context-type="linenumber">323</context>
        </context-group>
      </trans-unit>
      <trans-unit id="8858580062214623097" datatype="html">
        <source>Has any of these storage paths</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">299</context>
+          <context context-type="linenumber">331</context>
        </context-group>
      </trans-unit>
      <trans-unit id="6070943364927280151" datatype="html">
        <source>Does not have storage paths</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">307</context>
+          <context context-type="linenumber">339</context>
        </context-group>
      </trans-unit>
      <trans-unit id="6250799006816371860" datatype="html">
        <source>Matches custom field query</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">315</context>
+          <context context-type="linenumber">347</context>
        </context-group>
      </trans-unit>
      <trans-unit id="3138206142174978019" datatype="html">
        <source>Create new workflow</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">524</context>
+          <context context-type="linenumber">556</context>
        </context-group>
      </trans-unit>
      <trans-unit id="5996779210524133604" datatype="html">
        <source>Edit workflow</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/common/edit-dialog/workflow-edit-dialog/workflow-edit-dialog.component.ts</context>
-          <context context-type="linenumber">528</context>
+          <context context-type="linenumber">560</context>
        </context-group>
      </trans-unit>
      <trans-unit id="5457837313196342910" datatype="html">
@@ -8208,29 +8314,6 @@
          <context context-type="linenumber">164,165</context>
        </context-group>
      </trans-unit>
-      <trans-unit id="5701618810648052610" datatype="html">
-        <source>Title</source>
-        <context-group purpose="location">
-          <context context-type="sourcefile">src/app/components/document-detail/document-detail.component.html</context>
-          <context context-type="linenumber">167</context>
-        </context-group>
-        <context-group purpose="location">
-          <context context-type="sourcefile">src/app/components/document-list/document-list.component.html</context>
-          <context context-type="linenumber">218,219</context>
-        </context-group>
-        <context-group purpose="location">
-          <context context-type="sourcefile">src/app/components/document-list/filter-editor/filter-editor.component.ts</context>
-          <context context-type="linenumber">197</context>
-        </context-group>
-        <context-group purpose="location">
-          <context context-type="sourcefile">src/app/data/document.ts</context>
-          <context context-type="linenumber">30</context>
-        </context-group>
-        <context-group purpose="location">
-          <context context-type="sourcefile">src/app/data/document.ts</context>
-          <context context-type="linenumber">90</context>
-        </context-group>
-      </trans-unit>
      <trans-unit id="5114742157723900905" datatype="html">
        <source>Date created</source>
        <context-group purpose="location">
@@ -462,6 +462,45 @@
          </div>
        </div>
      }
+      @case (WorkflowActionType.ApplyAiSuggestions) {
+        <div class="row">
+          <div class="col">
+            <p class="text-muted small" i18n>The document will be sent to the configured AI service for suggestions. Consider costs and privacy.</p>
+            <pngx-input-select
+              i18n-title
+              title="Apply suggestions for"
+              [items]="aiSuggestionFieldOptions"
+              [multiple]="true"
+              formControlName="ai_suggestion_fields"
+              [error]="error?.actions?.[i]?.ai_suggestion_fields"
+              hint="Suggestions for fields that are not selected are discarded."
+              i18n-hint
+            ></pngx-input-select>
+          </div>
+        </div>
+        <div class="row">
+          <div class="col-md-6">
+            <pngx-input-switch
+              [horizontal]="true"
+              i18n-title
+              title="Create missing items"
+              formControlName="ai_create_missing"
+              hint="Create suggested tags, correspondents and document types that do not exist yet."
+              i18n-hint
+            ></pngx-input-switch>
+          </div>
+          <div class="col-md-6">
+            <pngx-input-switch
+              [horizontal]="true"
+              i18n-title
+              title="Overwrite existing values"
+              formControlName="ai_overwrite_existing"
+              hint="Apply suggestions even if the document already has a value. Tags are always added, never replaced."
+              i18n-hint
+            ></pngx-input-switch>
+          </div>
+        </div>
+      }
    }
  </div>
 </ng-template>
@@ -22,6 +22,7 @@ import {
 } from 'src/app/data/matching-model'
 import { Workflow } from 'src/app/data/workflow'
 import {
+  AISuggestionField,
  WorkflowAction,
  WorkflowActionType,
 } from 'src/app/data/workflow-action'
@@ -49,6 +50,7 @@ import { TagsComponent } from '../../input/tags/tags.component'
 import { TextComponent } from '../../input/text/text.component'
 import { EditDialogMode } from '../edit-dialog.component'
 import {
+  AI_SUGGESTION_FIELD_OPTIONS,
  DOCUMENT_SOURCE_OPTIONS,
  SCHEDULE_DATE_FIELD_OPTIONS,
  TriggerFilterType,
@@ -239,14 +241,15 @@ describe('WorkflowEditDialogComponent', () => {
      SCHEDULE_DATE_FIELD_OPTIONS
    )

-    // Email disabled
+    // Email, remote OCR and AI all disabled
    jest.spyOn(settingsService, 'get').mockReturnValue(false)
    component.ngOnInit()
    expect(component.actionTypeOptions).toEqual(
      WORKFLOW_ACTION_OPTIONS.filter(
        (a) =>
          a.id !== WorkflowActionType.Email &&
-          a.id !== WorkflowActionType.RemoteOcr
+          a.id !== WorkflowActionType.RemoteOcr &&
+          a.id !== WorkflowActionType.ApplyAiSuggestions
      )
    )
  })
@@ -344,6 +347,125 @@ describe('WorkflowEditDialogComponent', () => {
    )
  })

+  it('should offer apply AI suggestions unless every trigger is consumption', () => {
+    jest.spyOn(settingsService, 'get').mockReturnValue(true)
+
+    // Consumption runs before the document has been parsed, so there would be
+    // no content to make suggestions from
+    component.object = {
+      name: 'Workflow 1',
+      order: 0,
+      enabled: true,
+      triggers: [{ type: WorkflowTriggerType.Consumption }],
+      actions: [],
+    } as Workflow
+    component.ngOnInit()
+    expect(component.actionTypeOptions.map((a) => a.id)).not.toContain(
+      WorkflowActionType.ApplyAiSuggestions
+    )
+
+    // A second, usable trigger is enough
+    component.object = {
+      name: 'Workflow 2',
+      order: 0,
+      enabled: true,
+      triggers: [
+        { type: WorkflowTriggerType.Consumption },
+        { type: WorkflowTriggerType.DocumentAdded },
+      ],
+      actions: [],
+    } as Workflow
+    component.ngOnInit()
+    expect(component.actionTypeOptions.map((a) => a.id)).toContain(
+      WorkflowActionType.ApplyAiSuggestions
+    )
+  })
+
+  it('should keep apply AI suggestions listed when an action already uses it', () => {
+    jest.spyOn(settingsService, 'get').mockReturnValue(true)
+
+    // Otherwise changing the trigger would silently blank the selection
+    component.object = {
+      name: 'Workflow 1',
+      order: 0,
+      enabled: true,
+      triggers: [{ type: WorkflowTriggerType.Consumption }],
+      actions: [{ type: WorkflowActionType.ApplyAiSuggestions }],
+    } as Workflow
+    component.ngOnInit()
+
+    expect(component.actionTypeOptions.map((a) => a.id)).toContain(
+      WorkflowActionType.ApplyAiSuggestions
+    )
+  })
+
+  it('should not offer apply AI suggestions when AI is disabled', () => {
+    jest
+      .spyOn(settingsService, 'get')
+      .mockImplementation((key) => key !== SETTINGS_KEYS.AI_ENABLED)
+
+    component.object = {
+      name: 'Workflow 1',
+      order: 0,
+      enabled: true,
+      triggers: [{ type: WorkflowTriggerType.DocumentAdded }],
+      actions: [],
+    } as Workflow
+    component.ngOnInit()
+
+    expect(component.actionTypeOptions.map((a) => a.id)).not.toContain(
+      WorkflowActionType.ApplyAiSuggestions
+    )
+  })
+
+  it('should create form fields for apply AI suggestions options', () => {
+    component.object = {
+      name: 'Workflow 1',
+      order: 0,
+      enabled: true,
+      triggers: [{ type: WorkflowTriggerType.DocumentAdded }],
+      actions: [
+        {
+          type: WorkflowActionType.ApplyAiSuggestions,
+          ai_suggestion_fields: [
+            AISuggestionField.Title,
+            AISuggestionField.Tags,
+          ],
+          ai_create_missing: true,
+          ai_overwrite_existing: true,
+        },
+      ],
+    } as Workflow
+    component.ngOnInit()
+
+    const action = component.actionFields.at(0)
+    expect(action.get('ai_suggestion_fields').value).toEqual([
+      AISuggestionField.Title,
+      AISuggestionField.Tags,
+    ])
+    expect(action.get('ai_create_missing').value).toBeTruthy()
+    expect(action.get('ai_overwrite_existing').value).toBeTruthy()
+    expect(component.aiSuggestionFieldOptions).toEqual(
+      AI_SUGGESTION_FIELD_OPTIONS
+    )
+  })
+
+  it('should default apply AI suggestions options on a new action', () => {
+    component.object = {
+      name: 'Workflow 1',
+      order: 0,
+      enabled: true,
+      triggers: [{ type: WorkflowTriggerType.DocumentAdded }],
+      actions: [],
+    } as Workflow
+    component.addAction()
+
+    const action = component.actionFields.at(component.actionFields.length - 1)
+    expect(action.get('ai_suggestion_fields').value).toEqual([])
+    expect(action.get('ai_create_missing').value).toBeFalsy()
+    expect(action.get('ai_overwrite_existing').value).toBeFalsy()
+  })
+
  it('should support add and remove triggers and actions', () => {
    component.object = workflow
    component.addTrigger()
@@ -30,6 +30,7 @@ import { StoragePath } from 'src/app/data/storage-path'
 import { SETTINGS_KEYS } from 'src/app/data/ui-settings'
 import { Workflow } from 'src/app/data/workflow'
 import {
+  AISuggestionField,
  WorkflowAction,
  WorkflowActionType,
 } from 'src/app/data/workflow-action'
@@ -152,6 +153,37 @@ export const WORKFLOW_ACTION_OPTIONS = [
    id: WorkflowActionType.RemoteOcr,
    name: $localize`Remote OCR`,
  },
+  {
+    id: WorkflowActionType.ApplyAiSuggestions,
+    name: $localize`Apply AI suggestions`,
+  },
+]
+
+export const AI_SUGGESTION_FIELD_OPTIONS = [
+  {
+    id: AISuggestionField.Title,
+    name: $localize`Title`,
+  },
+  {
+    id: AISuggestionField.Tags,
+    name: $localize`Tags`,
+  },
+  {
+    id: AISuggestionField.Correspondent,
+    name: $localize`Correspondent`,
+  },
+  {
+    id: AISuggestionField.DocumentType,
+    name: $localize`Document type`,
+  },
+  {
+    id: AISuggestionField.StoragePath,
+    name: $localize`Storage path`,
+  },
+  {
+    id: AISuggestionField.Created,
+    name: $localize`Created date`,
+  },
 ]

 export enum TriggerFilterType {
@@ -576,6 +608,24 @@ export class WorkflowEditDialogComponent
      allowed = allowed.filter((a) => a.id !== WorkflowActionType.RemoteOcr)
    }

+    // Only available after consumption. Unlike remote OCR this is hidden only
+    // once every trigger is consumption, so it stays offered on a workflow
+    // that has no triggers yet.
+    const aiSuggestionsUsable =
+      this.settingsService.get(SETTINGS_KEYS.AI_ENABLED) &&
+      (!formWorkflow?.triggers?.length ||
+        formWorkflow.triggers.some(
+          (trigger) => trigger.type !== WorkflowTriggerType.Consumption
+        ) ||
+        formWorkflow.actions?.some(
+          (action) => action.type === WorkflowActionType.ApplyAiSuggestions
+        ))
+    if (!aiSuggestionsUsable) {
+      allowed = allowed.filter(
+        (a) => a.id !== WorkflowActionType.ApplyAiSuggestions
+      )
+    }
+
    if (
      this.allowedActionTypes?.length === allowed.length &&
      this.allowedActionTypes.every((a, i) => a.id === allowed[i].id)
@@ -1227,6 +1277,11 @@ export class WorkflowEditDialogComponent
        passwords: new FormControl(
          this.formatPasswords(action.passwords ?? [])
        ),
+        ai_suggestion_fields: new FormControl(
+          action.ai_suggestion_fields ?? []
+        ),
+        ai_create_missing: new FormControl(!!action.ai_create_missing),
+        ai_overwrite_existing: new FormControl(!!action.ai_overwrite_existing),
      }),
      { emitEvent }
    )
@@ -1316,6 +1371,10 @@ export class WorkflowEditDialogComponent
    return this.actionTypeOptions.find((t) => t.id === type)?.name ?? ''
  }

+  get aiSuggestionFieldOptions() {
+    return AI_SUGGESTION_FIELD_OPTIONS
+  }
+
  addAction() {
    if (!this.object) {
      this.object = Object.assign({}, this.objectForm.value)
@@ -1369,6 +1428,9 @@ export class WorkflowEditDialogComponent
        include_document: false,
      },
      passwords: [],
+      ai_suggestion_fields: [],
+      ai_create_missing: false,
+      ai_overwrite_existing: false,
    }
    this.object.actions.push(action)
    this.createActionField(action)
@@ -8,6 +8,17 @@ export enum WorkflowActionType {
  PasswordRemoval = 5,
  MoveToTrash = 6,
  RemoteOcr = 7,
+  ApplyAiSuggestions = 8,
+}
+
+// see src/documents/models.py AISuggestionField
+export enum AISuggestionField {
+  Title = 'title',
+  Tags = 'tags',
+  Correspondent = 'correspondent',
+  DocumentType = 'document_type',
+  StoragePath = 'storage_path',
+  Created = 'created',
 }

 export interface WorkflowActionEmail extends ObjectWithId {
@@ -102,4 +113,10 @@ export interface WorkflowAction extends ObjectWithId {
  webhook?: WorkflowActionWebhook

  passwords?: string[]
+
+  ai_suggestion_fields?: AISuggestionField[]
+
+  ai_create_missing?: boolean
+
+  ai_overwrite_existing?: boolean
 }
@@ -12,7 +12,6 @@ from typing import TYPE_CHECKING
 from typing import Any

 from django.contrib.contenttypes.models import ContentType
-from django.core.exceptions import FieldError
 from django.db.models import Case
 from django.db.models import CharField
 from django.db.models import Count
@@ -51,6 +50,7 @@ from documents.models import ShareLinkBundle
 from documents.models import StoragePath
 from documents.models import Tag
 from documents.permissions import permitted_object_ids
+from documents.versioning import ensure_effective_content

 if TYPE_CHECKING:
    from collections.abc import Callable
@@ -180,14 +180,9 @@ class TitleContentFilter(Filter):
            logger.warning(
                "Deprecated document filter parameter 'title_content' used; use `text` instead.",
            )
-            try:
-                return qs.filter(
-                    Q(title__icontains=value) | Q(effective_content__icontains=value),
-                )
-            except FieldError:
-                return qs.filter(
-                    Q(title__icontains=value) | Q(content__icontains=value),
-                )
+            return ensure_effective_content(qs).filter(
+                Q(title__icontains=value) | Q(effective_content__icontains=value),
+            )
        else:
            return qs

@@ -198,14 +193,9 @@ class EffectiveContentFilter(Filter):
        value = value.strip() if isinstance(value, str) else value
        if not value:
            return qs
-        try:
-            return qs.filter(
-                **{f"effective_content__{self.lookup_expr}": value},
-            )
-        except FieldError:
-            return qs.filter(
-                **{f"content__{self.lookup_expr}": value},
-            )
+        return ensure_effective_content(qs).filter(
+            **{f"effective_content__{self.lookup_expr}": value},
+        )


@extend_schema_field(serializers.BooleanField)
@@ -0,0 +1,84 @@
+# Generated by Django 5.2.16 on 2026-08-10 18:26
+
+from django.db import migrations
+from django.db import models
+
+
+class Migration(migrations.Migration):
+    dependencies = [
+        ("documents", "0024_alter_workflowaction_type"),
+    ]
+
+    operations = [
+        migrations.AddField(
+            model_name="workflowaction",
+            name="ai_create_missing",
+            field=models.BooleanField(
+                default=False,
+                help_text="Create suggested tags, correspondents, document types and storage paths that do not already exist instead of skipping them.",
+                verbose_name="create missing objects",
+            ),
+        ),
+        migrations.AddField(
+            model_name="workflowaction",
+            name="ai_overwrite_existing",
+            field=models.BooleanField(
+                default=False,
+                help_text="Apply suggestions even if the document already has a value for that field. Tags are always added to, never replaced.",
+                verbose_name="overwrite existing values",
+            ),
+        ),
+        migrations.AddField(
+            model_name="workflowaction",
+            name="ai_suggestion_fields",
+            field=models.JSONField(
+                blank=True,
+                help_text="Which of the AI-suggested fields to apply to the document.",
+                null=True,
+                verbose_name="AI suggestion fields",
+            ),
+        ),
+        migrations.AlterField(
+            model_name="workflowaction",
+            name="type",
+            field=models.PositiveSmallIntegerField(
+                choices=[
+                    (1, "Assignment"),
+                    (2, "Removal"),
+                    (3, "Email"),
+                    (4, "Webhook"),
+                    (5, "Password removal"),
+                    (6, "Move to trash"),
+                    (7, "Remote OCR"),
+                    (8, "Apply AI suggestions"),
+                ],
+                default=1,
+                verbose_name="Workflow Action Type",
+            ),
+        ),
+        migrations.AlterField(
+            model_name="paperlesstask",
+            name="task_type",
+            field=models.CharField(
+                choices=[
+                    ("consume_file", "Consume File"),
+                    ("train_classifier", "Train Classifier"),
+                    ("sanity_check", "Sanity Check"),
+                    ("index_optimize", "Index Optimize"),
+                    ("mail_fetch", "Mail Fetch"),
+                    ("llm_index", "LLM Index"),
+                    ("empty_trash", "Empty Trash"),
+                    ("check_workflows", "Check Workflows"),
+                    ("bulk_update", "Bulk Update"),
+                    ("reprocess_document", "Reprocess Document"),
+                    ("build_share_link", "Build Share Link"),
+                    ("bulk_delete", "Bulk Delete"),
+                    ("apply_ai_suggestions", "Apply AI Suggestions"),
+                ],
+                db_index=True,
+                help_text="The kind of work being performed",
+                max_length=50,
+                verbose_name="Task Type",
+            ),
+        ),
+    ]
@@ -766,6 +766,7 @@ class PaperlessTask(ModelWithOwner):
        REPROCESS_DOCUMENT = "reprocess_document", _("Reprocess Document")
        BUILD_SHARE_LINK = "build_share_link", _("Build Share Link")
        BULK_DELETE = "bulk_delete", _("Bulk Delete")
+        APPLY_AI_SUGGESTIONS = "apply_ai_suggestions", _("Apply AI Suggestions")

    COMPLETE_STATUSES = (
        Status.SUCCESS,
@@ -1674,6 +1675,18 @@ class WorkflowAction(models.Model):
            7,
            _("Remote OCR"),
        )
+        APPLY_AI_SUGGESTIONS = (
+            8,
+            _("Apply AI suggestions"),
+        )
+
+    class AISuggestionField(models.TextChoices):
+        TITLE = ("title", _("Title"))
+        TAGS = ("tags", _("Tags"))
+        CORRESPONDENT = ("correspondent", _("Correspondent"))
+        DOCUMENT_TYPE = ("document_type", _("Document type"))
+        STORAGE_PATH = ("storage_path", _("Storage path"))
+        CREATED = ("created", _("Created date"))

    type = models.PositiveSmallIntegerField(
        _("Workflow Action Type"),
@@ -1912,6 +1925,33 @@ class WorkflowAction(models.Model):
        ),
    )

+    ai_suggestion_fields = models.JSONField(
+        _("AI suggestion fields"),
+        null=True,
+        blank=True,
+        help_text=_(
+            "Which of the AI-suggested fields to apply to the document.",
+        ),
+    )
+
+    ai_create_missing = models.BooleanField(
+        _("create missing objects"),
+        default=False,
+        help_text=_(
+            "Create suggested tags, correspondents, document types and storage "
+            "paths that do not already exist instead of skipping them.",
+        ),
+    )
+
+    ai_overwrite_existing = models.BooleanField(
+        _("overwrite existing values"),
+        default=False,
+        help_text=_(
+            "Apply suggestions even if the document already has a value for that "
+            "field. Tags are always added to, never replaced.",
+        ),
+    )
+
    class Meta:
        verbose_name = _("workflow action")
        verbose_name_plural = _("workflow actions")
@@ -3235,6 +3235,9 @@ class WorkflowActionSerializer(serializers.ModelSerializer[WorkflowAction]):
            "email",
            "webhook",
            "passwords",
+            "ai_suggestion_fields",
+            "ai_create_missing",
+            "ai_overwrite_existing",
        ]

    def validate(self, attrs):
@@ -3292,6 +3295,23 @@ class WorkflowActionSerializer(serializers.ModelSerializer[WorkflowAction]):
                    "Passwords are required for password removal actions",
                )

+        if (
+            "type" in attrs
+            and attrs["type"] == WorkflowAction.WorkflowActionType.APPLY_AI_SUGGESTIONS
+        ):
+            fields = attrs.get("ai_suggestion_fields")
+            valid_fields = set(WorkflowAction.AISuggestionField.values)
+            if (
+                fields is None
+                or not isinstance(fields, list)
+                or len(fields) == 0
+                or any(field not in valid_fields for field in fields)
+            ):
+                raise serializers.ValidationError(
+                    "At least one valid field is required for apply AI "
+                    f"suggestions actions, options are: {sorted(valid_fields)}",
+                )
+
        return attrs


@@ -3320,24 +3340,43 @@ class WorkflowSerializer(serializers.ModelSerializer[Workflow]):
                action.get("type") == WorkflowAction.WorkflowActionType.REMOTE_OCR
                for action in attrs["actions"]
            )
+            has_ai_suggestions_action = any(
+                action.get("type")
+                == WorkflowAction.WorkflowActionType.APPLY_AI_SUGGESTIONS
+                for action in attrs["actions"]
+            )
        else:
            has_remote_ocr_action = self.instance is not None and (
                self.instance.actions.filter(
                    type=WorkflowAction.WorkflowActionType.REMOTE_OCR,
                ).exists()
            )
+            has_ai_suggestions_action = self.instance is not None and (
+                self.instance.actions.filter(
+                    type=WorkflowAction.WorkflowActionType.APPLY_AI_SUGGESTIONS,
+                ).exists()
+            )

        if "triggers" in attrs:
            has_consumption_trigger = any(
                trigger.get("type") == WorkflowTrigger.WorkflowTriggerType.CONSUMPTION
                for trigger in attrs["triggers"]
            )
+            has_non_consumption_trigger = any(
+                trigger.get("type") != WorkflowTrigger.WorkflowTriggerType.CONSUMPTION
+                for trigger in attrs["triggers"]
+            )
        else:
            has_consumption_trigger = self.instance is not None and (
                self.instance.triggers.filter(
                    type=WorkflowTrigger.WorkflowTriggerType.CONSUMPTION,
                ).exists()
            )
+            has_non_consumption_trigger = self.instance is not None and (
+                self.instance.triggers.exclude(
+                    type=WorkflowTrigger.WorkflowTriggerType.CONSUMPTION,
+                ).exists()
+            )

        # Remote OCR can only work with consumption triggers
        if has_remote_ocr_action and not has_consumption_trigger:
@@ -3345,6 +3384,14 @@ class WorkflowSerializer(serializers.ModelSerializer[Workflow]):
                "Remote OCR actions require a consumption started trigger",
            )

+        # Suggestions are made from the document content, which does not exist
+        # until after consumption has finished
+        if has_ai_suggestions_action and not has_non_consumption_trigger:
+            raise serializers.ValidationError(
+                "Apply AI suggestions actions require a trigger other than "
+                "consumption started",
+            )
+
        return attrs

    def update_triggers_and_actions(
@@ -984,6 +984,28 @@ def run_workflows(
                            "triggers, ignoring",
                            extra={"group": logging_group},
                        )
+                elif (
+                    action.type
+                    == WorkflowAction.WorkflowActionType.APPLY_AI_SUGGESTIONS
+                ):
+                    if use_overrides:
+                        # The document has not been parsed yet, so there is no
+                        # content for the LLM to make suggestions from
+                        logger.debug(
+                            "Apply AI suggestions action does not apply to "
+                            "consumption triggers, ignoring",
+                            extra={"group": logging_group},
+                        )
+                    else:
+                        # Queued rather than run sync
+                        from documents.tasks import apply_ai_suggestions
+
+                        # kwargs so the PaperlessTask record can note the
+                        # document, see _extract_input_data
+                        apply_ai_suggestions.delay(
+                            action_id=action.pk,
+                            document_id=document.pk,
+                        )

            if not use_overrides:
                # limit title to 128 characters
@@ -1039,6 +1061,7 @@ TRACKED_TASKS: dict[str, PaperlessTask.TaskType] = {
    "documents.tasks.update_document_content_maybe_archive_file": PaperlessTask.TaskType.REPROCESS_DOCUMENT,
    "documents.tasks.build_share_link_bundle": PaperlessTask.TaskType.BUILD_SHARE_LINK,
    "documents.bulk_edit.delete": PaperlessTask.TaskType.BULK_DELETE,
+    "documents.tasks.apply_ai_suggestions": PaperlessTask.TaskType.APPLY_AI_SUGGESTIONS,
 }

 _CELERY_STATE_TO_STATUS: dict[str, PaperlessTask.Status] = {
@@ -1092,6 +1115,12 @@ def _extract_input_data(
            return {"account_ids": account_ids}
        return {}

+    if task_type == PaperlessTask.TaskType.APPLY_AI_SUGGESTIONS:
+        document_id = task_kwargs.get("document_id")
+        if document_id is not None:
+            return {"document_id": document_id}
+        return {}
+
    return {}


@@ -71,6 +71,7 @@ from paperless.config import RemoteOCRConfig
 from paperless.logging import consume_task_id
 from paperless.parsers import ParserContext
 from paperless.parsers.registry import get_parser_registry
+from paperless_ai.exceptions import LLMTimeoutError
 from paperless_ai.indexing import llm_index_add_or_update_document
 from paperless_ai.indexing import llm_index_remove_document
 from paperless_ai.indexing import update_llm_index
@@ -714,6 +715,45 @@ def llmindex_index(
    )


+@shared_task(
+    bind=True,
+    autoretry_for=(LLMTimeoutError,),
+    max_retries=3,
+    retry_backoff=60,
+    retry_backoff_max=600,
+    retry_jitter=True,
+)
+def apply_ai_suggestions(self, action_id: int, document_id: int) -> None:
+    """
+    Deferred "apply AI suggestions" workflow action.
+    """
+    from documents.models import WorkflowAction
+    from documents.workflows.ai import apply_ai_suggestions_to_document
+
+    try:
+        action = WorkflowAction.objects.get(pk=action_id)
+        document = Document.objects.select_related("owner").get(pk=document_id)
+    except (WorkflowAction.DoesNotExist, Document.DoesNotExist):
+        logger.warning(
+            "Workflow action %s or document %s no longer exists, "
+            "not applying AI suggestions",
+            action_id,
+            document_id,
+        )
+        return
+
+    if not apply_ai_suggestions_to_document(action, document):
+        return
+
+    # No document_updated signal to avoid loop
+    clear_document_caches(document.pk)
+    index_document.delay(document.pk)
+
+    ai_config = AIConfig()
+    if ai_config.llm_index_enabled:
+        update_document_in_llm_index.apply_async(kwargs={"document": document})
+
+
@shared_task
 def update_document_in_llm_index(document) -> None:
    llm_index_add_or_update_document(document)
@@ -2,15 +2,14 @@ from __future__ import annotations

 import datetime
 from typing import TYPE_CHECKING
-from unittest import TestCase
 from unittest import mock

 from auditlog.models import LogEntry  # type: ignore[import-untyped]
 from django.contrib.auth.models import Permission
 from django.contrib.auth.models import User
 from django.contrib.contenttypes.models import ContentType
-from django.core.exceptions import FieldError
 from django.core.files.uploadedfile import SimpleUploadedFile
+from django.test import TestCase as DjangoTestCase
 from django.utils import timezone
 from rest_framework import status
 from rest_framework.test import APITestCase
@@ -21,6 +20,8 @@ from documents.filters import TitleContentFilter
 from documents.models import Document
 from documents.tests.utils import DirectoriesMixin
 from documents.tests.utils import read_streaming_response
+from documents.versioning import annotate_effective_content
+from documents.views import DocumentSelectionMixin

 if TYPE_CHECKING:
    from pathlib import Path
@@ -889,32 +890,103 @@ class TestDocumentVersioningApi(DirectoriesMixin, APITestCase):
        )


-class TestVersionAwareFilters(TestCase):
-    def test_title_content_filter_falls_back_to_content(self) -> None:
-        queryset = mock.Mock()
-        fallback_queryset = mock.Mock()
-        queryset.filter.side_effect = [FieldError("missing field"), fallback_queryset]
+class TestVersionAwareFilters(DjangoTestCase):
+    """
+    The filters annotate effective_content themselves rather than relying on
+    the caller's queryset carrying it, so they stay version-aware on a plain
+    Document queryset (e.g. the bulk-edit "select all matching" path).
+    """

-        result = TitleContentFilter().filter(queryset, " latest ")
+    def setUp(self) -> None:
+        super().setUp()
+        self.root = Document.objects.create(
+            title="root",
+            checksum="root",
+            mime_type="application/pdf",
+            content="superseded-content",
+        )
+        Document.objects.create(
+            title="version",
+            checksum="version",
+            mime_type="application/pdf",
+            root_document=self.root,
+            version_index=1,
+            content="latest-content",
+        )
+        self.unversioned = Document.objects.create(
+            title="unversioned",
+            checksum="unversioned",
+            mime_type="application/pdf",
+            content="latest-content",
+        )

-        self.assertIs(result, fallback_queryset)
-        self.assertEqual(queryset.filter.call_count, 2)
-
-    def test_effective_content_filter_falls_back_to_content_lookup(self) -> None:
-        queryset = mock.Mock()
-        fallback_queryset = mock.Mock()
-        queryset.filter.side_effect = [FieldError("missing field"), fallback_queryset]
-
-        result = EffectiveContentFilter(lookup_expr="icontains").filter(
-            queryset,
+    def test_title_content_filter_matches_latest_version_content(self) -> None:
+        result = TitleContentFilter().filter(
+            Document.objects.filter(root_document__isnull=True),
            " latest ",
        )

-        self.assertIs(result, fallback_queryset)
-        first_kwargs = queryset.filter.call_args_list[0].kwargs
-        second_kwargs = queryset.filter.call_args_list[1].kwargs
-        self.assertEqual(first_kwargs, {"effective_content__icontains": "latest"})
-        self.assertEqual(second_kwargs, {"content__icontains": "latest"})
+        self.assertCountEqual(
+            [doc.id for doc in result],
+            [self.root.id, self.unversioned.id],
+        )
+
+    def test_effective_content_filter_matches_latest_version_content(self) -> None:
+        result = EffectiveContentFilter(lookup_expr="icontains").filter(
+            Document.objects.filter(root_document__isnull=True),
+            " latest ",
+        )
+
+        self.assertCountEqual(
+            [doc.id for doc in result],
+            [self.root.id, self.unversioned.id],
+        )
+
+    def test_effective_content_filter_ignores_superseded_content(self) -> None:
+        result = EffectiveContentFilter(lookup_expr="icontains").filter(
+            Document.objects.filter(root_document__isnull=True),
+            "superseded",
+        )
+
+        self.assertEqual(list(result), [])
+
+    def test_filters_reuse_an_existing_annotation(self) -> None:
+        """
+        Annotating twice under the same alias is an error, so an already
+        annotated queryset (the search path) has to be left alone.
+        """
+        annotated = annotate_effective_content(
+            Document.objects.filter(root_document__isnull=True),
+        )
+
+        result = EffectiveContentFilter(lookup_expr="icontains").filter(
+            annotated,
+            "latest",
+        )
+
+        self.assertCountEqual(
+            [doc.id for doc in result],
+            [self.root.id, self.unversioned.id],
+        )
+
+    def test_bulk_selection_does_not_match_superseded_content(self) -> None:
+        """
+        Bulk edit's "select all matching" builds its own queryset, so before
+        the filters annotated for themselves it matched the root document's
+        superseded content -- selecting documents the list view, filtered by
+        the same term, does not show.
+        """
+        user = User.objects.create_superuser(username="bulk_selection")
+
+        selected = DocumentSelectionMixin()._resolve_document_ids(
+            user=user,
+            validated_data={
+                "all": True,
+                "filters": {"content__icontains": "superseded"},
+            },
+        )
+
+        self.assertEqual(selected, [])

    def test_effective_content_filter_returns_input_for_empty_values(self) -> None:
        queryset = mock.Mock()
@@ -923,3 +995,36 @@ class TestVersionAwareFilters(TestCase):

        self.assertIs(result, queryset)
        queryset.filter.assert_not_called()
+
+
+class TestBulkSelectionExcludesVersions(DjangoTestCase):
+    def test_select_all_matching_does_not_select_version_documents(self) -> None:
+        """
+        "Select all matching" reconstructs the document list, which never
+        contains version documents as rows of their own.
+        """
+        user = User.objects.create_superuser(username="bulk_versions")
+        root = Document.objects.create(
+            title="shared-title root",
+            checksum="bulk-root",
+            mime_type="application/pdf",
+            content="root",
+        )
+        Document.objects.create(
+            title="shared-title version",
+            checksum="bulk-version",
+            mime_type="application/pdf",
+            root_document=root,
+            version_index=1,
+            content="version",
+        )
+
+        selected = DocumentSelectionMixin()._resolve_document_ids(
+            user=user,
+            validated_data={
+                "all": True,
+                "filters": {"title__icontains": "shared-title"},
+            },
+        )
+
+        self.assertEqual(selected, [root.id])
@@ -1917,6 +1917,29 @@ class TestDocumentSearchApi(DirectoriesMixin, APITestCase):
        self.assertEqual(len(response.data["documents"]), 1)
        self.assertEqual(response.data["documents"][0]["id"], title_match.id)

+    def test_global_search_returns_latest_version_content(self) -> None:
+        root = Document.objects.create(
+            title="bank statement",
+            content="superseded content",
+            checksum="GSV1",
+            pk=23,
+        )
+        Document.objects.create(
+            title="bank statement v2",
+            content="latest content",
+            checksum="GSV2",
+            pk=24,
+            root_document=root,
+            version_index=1,
+        )
+
+        self.client.force_authenticate(self.user)
+
+        response = self.client.get("/api/search/?query=bank&db_only=true")
+        self.assertEqual(response.status_code, status.HTTP_200_OK)
+        returned = {doc["id"]: doc["content"] for doc in response.data["documents"]}
+        self.assertEqual(returned.get(root.id), "latest content")
+
    def test_global_search_filters_owned_mail_objects(self) -> None:
        user1 = User.objects.create_user("mail-search-user")
        user2 = User.objects.create_user("other-mail-search-user")
@@ -641,6 +641,145 @@ class TestApiWorkflows(DirectoriesMixin, APITestCase):
        self.assertEqual(response.status_code, status.HTTP_400_BAD_REQUEST)
        self.assertEqual(self.workflow.triggers.get(), self.trigger)

+    def _post_ai_suggestions_workflow(self, *, trigger_types, action: dict):
+        def trigger(trigger_type):
+            # consumption triggers require a filter of their own
+            if trigger_type == WorkflowTrigger.WorkflowTriggerType.CONSUMPTION:
+                return {"type": trigger_type, "filter_filename": "*.pdf"}
+            return {"type": trigger_type}
+
+        return self.client.post(
+            self.ENDPOINT,
+            json.dumps(
+                {
+                    "name": "Apply AI suggestions",
+                    "order": 1,
+                    "triggers": [trigger(t) for t in trigger_types],
+                    "actions": [
+                        {
+                            "type": WorkflowAction.WorkflowActionType.APPLY_AI_SUGGESTIONS,
+                            **action,
+                        },
+                    ],
+                },
+            ),
+            content_type="application/json",
+        )
+
+    def test_api_create_apply_ai_suggestions_action(self) -> None:
+        """
+        GIVEN:
+            - API request to create a workflow with an apply AI suggestions
+              action and a valid set of fields
+        WHEN:
+            - API is called
+        THEN:
+            - The workflow is created with the chosen options
+        """
+        response = self._post_ai_suggestions_workflow(
+            trigger_types=[WorkflowTrigger.WorkflowTriggerType.DOCUMENT_ADDED],
+            action={
+                "ai_suggestion_fields": ["title", "tags", "correspondent"],
+                "ai_create_missing": True,
+                "ai_overwrite_existing": True,
+            },
+        )
+
+        self.assertEqual(response.status_code, status.HTTP_201_CREATED)
+        action = Workflow.objects.get(name="Apply AI suggestions").actions.first()
+        self.assertEqual(
+            action.ai_suggestion_fields,
+            ["title", "tags", "correspondent"],
+        )
+        self.assertTrue(action.ai_create_missing)
+        self.assertTrue(action.ai_overwrite_existing)
+
+    def test_api_create_apply_ai_suggestions_action_requires_fields(self) -> None:
+        """
+        GIVEN:
+            - API request to create an apply AI suggestions action with no
+              fields selected, which could never do anything
+        WHEN:
+            - API is called
+        THEN:
+            - Correct HTTP 400 response
+            - No objects are created
+        """
+        existing_count = Workflow.objects.count()
+
+        response = self._post_ai_suggestions_workflow(
+            trigger_types=[WorkflowTrigger.WorkflowTriggerType.DOCUMENT_ADDED],
+            action={"ai_suggestion_fields": []},
+        )
+
+        self.assertEqual(response.status_code, status.HTTP_400_BAD_REQUEST)
+        self.assertEqual(Workflow.objects.count(), existing_count)
+
+    def test_api_create_apply_ai_suggestions_action_rejects_unknown_field(
+        self,
+    ) -> None:
+        """
+        GIVEN:
+            - API request to create an apply AI suggestions action naming a
+              field that does not exist
+        WHEN:
+            - API is called
+        THEN:
+            - Correct HTTP 400 response
+        """
+        response = self._post_ai_suggestions_workflow(
+            trigger_types=[WorkflowTrigger.WorkflowTriggerType.DOCUMENT_ADDED],
+            action={"ai_suggestion_fields": ["title", "not_a_field"]},
+        )
+
+        self.assertEqual(response.status_code, status.HTTP_400_BAD_REQUEST)
+
+    def test_api_create_apply_ai_suggestions_action_rejects_consumption_only(
+        self,
+    ) -> None:
+        """
+        GIVEN:
+            - API request to create an apply AI suggestions action whose only
+              trigger is consumption started, so there is no document content
+              to make suggestions from yet
+        WHEN:
+            - API is called
+        THEN:
+            - Correct HTTP 400 response
+            - No objects are created
+        """
+        existing_count = Workflow.objects.count()
+
+        response = self._post_ai_suggestions_workflow(
+            trigger_types=[WorkflowTrigger.WorkflowTriggerType.CONSUMPTION],
+            action={"ai_suggestion_fields": ["title"]},
+        )
+
+        self.assertEqual(response.status_code, status.HTTP_400_BAD_REQUEST)
+        self.assertEqual(Workflow.objects.count(), existing_count)
+
+    def test_api_create_apply_ai_suggestions_action_allows_extra_consumption_trigger(
+        self,
+    ) -> None:
+        """
+        GIVEN:
+            - API request to create an apply AI suggestions action with a
+              consumption trigger alongside a usable one
+        WHEN:
+            - API is called
+        THEN:
+            - The workflow is created, the action applies to the other trigger
+        """
+        response = self._post_ai_suggestions_workflow(
+            trigger_types=[
+                WorkflowTrigger.WorkflowTriggerType.CONSUMPTION,
+                WorkflowTrigger.WorkflowTriggerType.DOCUMENT_ADDED,
+            ],
+            action={"ai_suggestion_fields": ["title"]},
+        )
+
+        self.assertEqual(response.status_code, status.HTTP_201_CREATED)
+
    def test_api_create_workflow_trigger_action_empty_fields(self) -> None:
        """
        GIVEN:
@@ -385,6 +385,25 @@ class TestTaskFailureHandler:
        task_failure_handler(task_id=None, exception=ValueError("x"), traceback=None)


+@pytest.mark.django_db
+class TestApplyAiSuggestionsTracking:
+    def test_records_the_document_it_is_for(self) -> None:
+        """
+        The action queues one task per document, so the tracked record notes
+        which document it is for -- otherwise a bulk run is an indistinguishable
+        wall of identical entries in the tasks list.
+        """
+        task_id = send_publish(
+            "documents.tasks.apply_ai_suggestions",
+            (),
+            {"action_id": 1, "document_id": 42},
+        )
+
+        task = PaperlessTask.objects.get(task_id=task_id)
+        assert task.task_type == PaperlessTask.TaskType.APPLY_AI_SUGGESTIONS
+        assert task.input_data == {"document_id": 42}
+
+
@pytest.mark.django_db
 class TestTaskRevokedHandler:
    def test_marks_task_revoked(self, mocker: pytest_mock.MockerFixture) -> None:
@@ -14,6 +14,7 @@ from documents.models import Correspondent
 from documents.models import Document
 from documents.models import DocumentType
 from documents.models import Tag
+from documents.models import WorkflowAction
 from documents.sanity_checker import SanityCheckFailedException
 from documents.sanity_checker import SanityCheckMessages
 from documents.tests.test_classifier import dummy_preprocess
@@ -447,3 +448,110 @@ class TestAIIndex(DirectoriesMixin, TestCase):
                rebuild=False,
                document_ids=doc_ids,
            )
+
+
+class TestApplyAISuggestionsTask(DirectoriesMixin, TestCase):
+    def setUp(self) -> None:
+        super().setUp()
+        self.doc = Document.objects.create(
+            title="doc",
+            content="content",
+            checksum="apply-ai-suggestions",
+        )
+        self.action = WorkflowAction.objects.create(
+            type=WorkflowAction.WorkflowActionType.APPLY_AI_SUGGESTIONS,
+            ai_suggestion_fields=[WorkflowAction.AISuggestionField.TITLE],
+        )
+
+    def test_reindexes_without_sending_document_updated(self) -> None:
+        """
+        GIVEN:
+            - An apply AI suggestions action that changes the document
+        WHEN:
+            - The task runs
+        THEN:
+            - The search index and caches are refreshed directly, deliberately
+              not via the document_updated signal: that re-runs updated
+              workflows, which for this action means queueing another LLM
+              query for a document it just changed, forever
+        """
+        with (
+            mock.patch(
+                "documents.workflows.ai.apply_ai_suggestions_to_document",
+                return_value=["title"],
+            ),
+            mock.patch("documents.tasks.index_document") as index_document,
+            mock.patch("documents.tasks.clear_document_caches") as clear_caches,
+            mock.patch("documents.tasks.document_updated") as document_updated,
+        ):
+            tasks.apply_ai_suggestions(self.action.pk, self.doc.pk)
+
+        index_document.delay.assert_called_once_with(self.doc.pk)
+        clear_caches.assert_called_once_with(self.doc.pk)
+        document_updated.send.assert_not_called()
+
+    def test_no_changes_skips_reindex(self) -> None:
+        """
+        GIVEN:
+            - An apply AI suggestions action that changes nothing
+        WHEN:
+            - The task runs
+        THEN:
+            - No reindexing work is queued
+        """
+        with (
+            mock.patch(
+                "documents.workflows.ai.apply_ai_suggestions_to_document",
+                return_value=[],
+            ),
+            mock.patch("documents.tasks.index_document") as index_document,
+        ):
+            tasks.apply_ai_suggestions(self.action.pk, self.doc.pk)
+
+        index_document.delay.assert_not_called()
+
+    @override_settings(AI_ENABLED=True, LLM_EMBEDDING_BACKEND="huggingface")
+    def test_updates_llm_index_when_enabled(self) -> None:
+        """
+        GIVEN:
+            - An apply AI suggestions action that changes the document
+            - The LLM index is enabled
+        WHEN:
+            - The task runs
+        THEN:
+            - The document is updated in the LLM index too
+        """
+        with (
+            mock.patch(
+                "documents.workflows.ai.apply_ai_suggestions_to_document",
+                return_value=["title"],
+            ),
+            mock.patch("documents.tasks.index_document"),
+            mock.patch(
+                "documents.tasks.update_document_in_llm_index",
+            ) as update_in_llm_index,
+        ):
+            tasks.apply_ai_suggestions(self.action.pk, self.doc.pk)
+
+        update_in_llm_index.apply_async.assert_called_once()
+
+    def test_deleted_document_is_a_noop(self) -> None:
+        """
+        GIVEN:
+            - A document that was deleted between the workflow running and the
+              queued task starting
+        WHEN:
+            - The task runs
+        THEN:
+            - It logs and exits rather than raising
+        """
+        with (
+            mock.patch(
+                "documents.workflows.ai.apply_ai_suggestions_to_document",
+            ) as apply_suggestions,
+            self.assertLogs("paperless.tasks", level="WARNING") as cm,
+        ):
+            tasks.apply_ai_suggestions(self.action.pk, self.doc.pk + 1000)
+
+        apply_suggestions.assert_not_called()
+        self.assertIn("no longer exists", "".join(cm.output))
@@ -31,7 +31,10 @@ from documents.file_handling import create_source_path_directory
 from documents.file_handling import generate_filename
 from documents.file_handling import generate_unique_filename
 from documents.signals.handlers import run_workflows
+from documents.workflows.ai import apply_ai_suggestions_to_document
 from documents.workflows.webhooks import send_webhook
+from paperless_ai.base_model import ClassificationSuggestions
+from paperless_ai.exceptions import LLMTimeoutError

 if TYPE_CHECKING:
    from django.db.models import QuerySet
@@ -5488,3 +5491,481 @@ class TestRemoteOCRWorkflowAction(DirectoriesMixin, SampleDirMixin, APITestCase)
            )

        self.assertIn("only applies to consumption triggers", "".join(cm.output))
+
+
+SUGGESTIONS: ClassificationSuggestions = {
+    "title": "Suggested Title",
+    "tags": {
+        "existing_ids": [],
+        "new_names": ["Existing Tag", "Suggested Tag"],
+    },
+    "correspondents": {
+        "existing_ids": [],
+        "new_names": ["Existing Correspondent", "Suggested Correspondent"],
+    },
+    "document_types": {
+        "existing_ids": [],
+        "new_names": ["Suggested Document Type"],
+    },
+    "storage_paths": {
+        "existing_ids": [],
+        "new_names": ["Suggested Storage Path"],
+    },
+    "dates": ["2024-03-05"],
+}
+
+ALL_SUGGESTION_FIELDS = [
+    WorkflowAction.AISuggestionField.TITLE,
+    WorkflowAction.AISuggestionField.TAGS,
+    WorkflowAction.AISuggestionField.CORRESPONDENT,
+    WorkflowAction.AISuggestionField.DOCUMENT_TYPE,
+    WorkflowAction.AISuggestionField.STORAGE_PATH,
+    WorkflowAction.AISuggestionField.CREATED,
+]
+
+
+@override_settings(AI_ENABLED=True)
+class TestApplyAISuggestionsWorkflowAction(
+    DirectoriesMixin,
+    SampleDirMixin,
+    APITestCase,
+):
+    def setUp(self) -> None:
+        super().setUp()
+        self.user = User.objects.create(username="ai-user")
+        self.doc = Document.objects.create(
+            title="original.pdf",
+            content="the document content",
+            checksum="ai-suggestions-checksum",
+            mime_type="application/pdf",
+            created=datetime.date(2020, 1, 1),
+            owner=self.user,
+        )
+
+    def make_action(self, **kwargs) -> WorkflowAction:
+        return WorkflowAction.objects.create(
+            type=WorkflowAction.WorkflowActionType.APPLY_AI_SUGGESTIONS,
+            ai_suggestion_fields=kwargs.pop(
+                "ai_suggestion_fields",
+                ALL_SUGGESTION_FIELDS,
+            ),
+            **kwargs,
+        )
+
+    def make_workflow(self, action: WorkflowAction, trigger_type) -> Workflow:
+        trigger = WorkflowTrigger.objects.create(type=trigger_type)
+        w = Workflow.objects.create(name="Apply AI suggestions", order=0)
+        w.triggers.add(trigger)
+        w.actions.add(action)
+        w.save()
+        return w
+
+    def apply(
+        self,
+        action: WorkflowAction,
+        suggestions: ClassificationSuggestions = SUGGESTIONS,
+    ) -> list[str]:
+        with mock.patch(
+            "documents.workflows.ai.get_ai_document_classification",
+            return_value=suggestions,
+        ):
+            changed = apply_ai_suggestions_to_document(action, self.doc)
+        self.doc.refresh_from_db()
+        return changed
+
+    def test_document_added_trigger_queues_task(self) -> None:
+        """
+        GIVEN:
+            - A document added workflow with an apply AI suggestions action
+        WHEN:
+            - A matching document is added
+        THEN:
+            - The work is queued rather than run inline, so a slow LLM query
+              cannot stall the rest of the workflow run
+        """
+        action = self.make_action()
+        self.make_workflow(action, WorkflowTrigger.WorkflowTriggerType.DOCUMENT_ADDED)
+
+        with mock.patch("documents.tasks.apply_ai_suggestions.delay") as delay:
+            run_workflows(
+                WorkflowTrigger.WorkflowTriggerType.DOCUMENT_ADDED,
+                self.doc,
+            )
+
+        delay.assert_called_once_with(action_id=action.pk, document_id=self.doc.pk)
+
+    def test_consumption_trigger_is_ignored(self) -> None:
+        """
+        GIVEN:
+            - A workflow with an apply AI suggestions action and a consumption
+              trigger alongside a valid one
+        WHEN:
+            - The consumption trigger fires
+        THEN:
+            - The action is skipped, since the document has not been parsed
+              yet and so has no content to make suggestions from
+        """
+        action = self.make_action()
+        w = self.make_workflow(
+            action,
+            WorkflowTrigger.WorkflowTriggerType.DOCUMENT_ADDED,
+        )
+        w.triggers.add(
+            WorkflowTrigger.objects.create(
+                type=WorkflowTrigger.WorkflowTriggerType.CONSUMPTION,
+            ),
+        )
+
+        test_file = shutil.copy(
+            self.SAMPLE_DIR / "simple.pdf",
+            self.dirs.scratch_dir / "simple.pdf",
+        )
+
+        with (
+            mock.patch("documents.tasks.apply_ai_suggestions.delay") as delay,
+            self.assertLogs("paperless.handlers", level="DEBUG") as cm,
+        ):
+            run_workflows(
+                WorkflowTrigger.WorkflowTriggerType.CONSUMPTION,
+                ConsumableDocument(
+                    source=DocumentSource.ConsumeFolder,
+                    original_file=test_file,
+                ),
+                overrides=DocumentMetadataOverrides(),
+            )
+
+        delay.assert_not_called()
+        self.assertIn("does not apply to consumption triggers", "".join(cm.output))
+
+    def test_no_selected_fields_does_nothing(self) -> None:
+        """
+        GIVEN:
+            - An action with no suggestion fields selected
+        WHEN:
+            - The action is applied
+        THEN:
+            - Nothing is changed and it is logged
+        """
+        action = self.make_action(ai_suggestion_fields=[])
+
+        with self.assertLogs("paperless.workflows.ai", level="WARNING") as cm:
+            changed = self.apply(action)
+
+        self.assertEqual(changed, [])
+        self.assertIn("no AI suggestion fields selected", "".join(cm.output))
+
+    @override_settings(AI_ENABLED=False)
+    def test_ai_disabled_does_nothing(self) -> None:
+        """
+        GIVEN:
+            - An action on an install where AI has since been disabled
+        WHEN:
+            - The action is applied
+        THEN:
+            - Nothing is changed and it is logged
+        """
+        action = self.make_action()
+
+        with self.assertLogs("paperless.workflows.ai", level="ERROR") as cm:
+            changed = self.apply(action)
+
+        self.assertEqual(changed, [])
+        self.assertIn("AI is not enabled", "".join(cm.output))
+
+    def test_invalid_configuration_leaves_document_untouched(self) -> None:
+        """
+        GIVEN:
+            - An AI backend that is misconfigured
+        WHEN:
+            - The action is applied
+        THEN:
+            - The failure is logged and the document is left alone. It is not
+              re-raised, because retrying will not fix a bad configuration
+        """
+        action = self.make_action()
+
+        with (
+            mock.patch(
+                "documents.workflows.ai.get_ai_document_classification",
+                side_effect=ValueError("nope"),
+            ),
+            self.assertLogs("paperless.workflows.ai", level="ERROR") as cm,
+        ):
+            changed = apply_ai_suggestions_to_document(action, self.doc)
+
+        self.assertEqual(changed, [])
+        self.doc.refresh_from_db()
+        self.assertEqual(self.doc.title, "original.pdf")
+        self.assertIn("Invalid AI configuration", "".join(cm.output))
+
+    def test_transient_llm_failure_is_raised_for_retry(self) -> None:
+        """
+        GIVEN:
+            - An LLM backend that times out, or rate limits the request
+        WHEN:
+            - The action is applied
+        THEN:
+            - The error propagates so the queued task can back off and retry,
+              rather than silently dropping this document's suggestions
+        """
+        action = self.make_action()
+
+        with (
+            mock.patch(
+                "documents.workflows.ai.get_ai_document_classification",
+                side_effect=LLMTimeoutError(),
+            ),
+            self.assertRaises(LLMTimeoutError),
+        ):
+            apply_ai_suggestions_to_document(action, self.doc)
+
+        self.doc.refresh_from_db()
+        self.assertEqual(self.doc.title, "original.pdf")
+
+    def test_only_matching_objects_are_applied(self) -> None:
+        """
+        GIVEN:
+            - An action without create missing, and only some of the suggested
+              objects existing
+        WHEN:
+            - The action is applied
+        THEN:
+            - Only the existing objects are assigned, unmatched suggestions are
+              dropped rather than creating anything
+        """
+        tag = Tag.objects.create(name="Existing Tag", owner=self.user)
+        correspondent = Correspondent.objects.create(
+            name="Existing Correspondent",
+            owner=self.user,
+        )
+        action = self.make_action(ai_overwrite_existing=True)
+
+        changed = self.apply(action)
+
+        self.assertEqual(self.doc.correspondent, correspondent)
+        self.assertEqual(list(self.doc.tags.all()), [tag])
+        # Nothing matched for these and create missing is off
+        self.assertIsNone(self.doc.document_type)
+        self.assertIsNone(self.doc.storage_path)
+        self.assertNotIn("document_type", changed)
+        self.assertEqual(Tag.objects.count(), 1)
+        self.assertEqual(Correspondent.objects.count(), 1)
+
+    def test_existing_id_suggestions_are_applied(self) -> None:
+        """
+        GIVEN:
+            - AI suggestions that select existing taxonomy candidates by ID
+        WHEN:
+            - The suggestions are applied
+        THEN:
+            - Each selected object is assigned to the document
+        """
+        tag = Tag.objects.create(name="Existing Tag", owner=self.user)
+        correspondent = Correspondent.objects.create(
+            name="Existing Correspondent",
+            owner=self.user,
+        )
+        document_type = DocumentType.objects.create(
+            name="Existing Document Type",
+            owner=self.user,
+        )
+        storage_path = StoragePath.objects.create(
+            name="Existing Storage Path",
+            path="{{ title }}",
+            owner=self.user,
+        )
+        action = self.make_action(ai_overwrite_existing=True)
+        suggestions: ClassificationSuggestions = {
+            **SUGGESTIONS,
+            "tags": {"existing_ids": [tag.pk], "new_names": []},
+            "correspondents": {
+                "existing_ids": [correspondent.pk],
+                "new_names": [],
+            },
+            "document_types": {
+                "existing_ids": [document_type.pk],
+                "new_names": [],
+            },
+            "storage_paths": {
+                "existing_ids": [storage_path.pk],
+                "new_names": [],
+            },
+        }
+
+        changed = self.apply(action, suggestions)
+
+        self.assertEqual(list(self.doc.tags.all()), [tag])
+        self.assertEqual(self.doc.correspondent, correspondent)
+        self.assertEqual(self.doc.document_type, document_type)
+        self.assertEqual(self.doc.storage_path, storage_path)
+        self.assertTrue(
+            {"tags", "correspondent", "document_type", "storage_path"} <= set(changed),
+        )
+
+    def test_create_missing_creates_objects_owned_by_document_owner(self) -> None:
+        """
+        GIVEN:
+            - An action with create missing enabled
+        WHEN:
+            - The action is applied and suggestions match nothing
+        THEN:
+            - Tags, correspondents and document types are created, owned by the
+              document owner so they stay private to them
+            - Storage paths are never created, since a path template cannot be
+              inferred from a name
+        """
+        action = self.make_action(
+            ai_create_missing=True,
+            ai_overwrite_existing=True,
+        )
+
+        changed = self.apply(action)
+
+        self.assertEqual(
+            sorted(t.name for t in self.doc.tags.all()),
+            ["Existing Tag", "Suggested Tag"],
+        )
+        self.assertEqual(self.doc.correspondent.name, "Existing Correspondent")
+        self.assertEqual(self.doc.correspondent.owner, self.user)
+        self.assertEqual(self.doc.document_type.name, "Suggested Document Type")
+        self.assertEqual(self.doc.document_type.owner, self.user)
+
+        self.assertIsNone(self.doc.storage_path)
+        self.assertFalse(StoragePath.objects.exists())
+        self.assertNotIn("storage_path", changed)
+
+    def test_overwrite_disabled_keeps_existing_values(self) -> None:
+        """
+        GIVEN:
+            - An action without overwrite existing
+            - A document that already has a title, created date and
+              correspondent
+        WHEN:
+            - The action is applied
+        THEN:
+            - The existing values are kept, only the empty document type is
+              filled in
+        """
+        existing = Correspondent.objects.create(name="Mine", owner=self.user)
+        self.doc.correspondent = existing
+        self.doc.save()
+        action = self.make_action(ai_create_missing=True)
+
+        changed = self.apply(action)
+
+        self.assertEqual(self.doc.title, "original.pdf")
+        self.assertEqual(self.doc.created, datetime.date(2020, 1, 1))
+        self.assertEqual(self.doc.correspondent, existing)
+        self.assertEqual(self.doc.document_type.name, "Suggested Document Type")
+        self.assertNotIn("title", changed)
+        self.assertNotIn("correspondent", changed)
+
+    def test_overwrite_enabled_replaces_existing_values(self) -> None:
+        """
+        GIVEN:
+            - An action with overwrite existing
+            - A document that already has a title and created date
+        WHEN:
+            - The action is applied
+        THEN:
+            - The suggested values replace them
+        """
+        action = self.make_action(
+            ai_create_missing=True,
+            ai_overwrite_existing=True,
+        )
+
+        changed = self.apply(action)
+
+        self.assertEqual(self.doc.title, "Suggested Title")
+        self.assertEqual(self.doc.created, datetime.date(2024, 3, 5))
+        self.assertIn("title", changed)
+        self.assertIn("created", changed)
+
+    def test_tags_are_added_not_replaced(self) -> None:
+        """
+        GIVEN:
+            - A document that already has a tag unrelated to the suggestions
+        WHEN:
+            - The action is applied with overwrite existing enabled
+        THEN:
+            - The existing tag is kept, since suggested tags are always
+              additive regardless of the overwrite setting
+        """
+        kept = Tag.objects.create(name="Do Not Remove", owner=self.user)
+        self.doc.tags.add(kept)
+        Tag.objects.create(name="Existing Tag", owner=self.user)
+        action = self.make_action(ai_overwrite_existing=True)
+
+        self.apply(action)
+
+        self.assertEqual(
+            sorted(t.name for t in self.doc.tags.all()),
+            ["Do Not Remove", "Existing Tag"],
+        )
+
+    def test_unselected_fields_are_untouched(self) -> None:
+        """
+        GIVEN:
+            - An action that only selects the title
+        WHEN:
+            - The action is applied
+        THEN:
+            - Only the title changes, even though the LLM suggested everything
+        """
+        action = self.make_action(
+            ai_suggestion_fields=[WorkflowAction.AISuggestionField.TITLE],
+            ai_create_missing=True,
+            ai_overwrite_existing=True,
+        )
+
+        changed = self.apply(action)
+
+        self.assertEqual(changed, ["title"])
+        self.assertEqual(self.doc.title, "Suggested Title")
+        self.assertEqual(self.doc.tags.count(), 0)
+        self.assertIsNone(self.doc.correspondent)
+        self.assertEqual(self.doc.created, datetime.date(2020, 1, 1))
+
+    def test_another_users_private_objects_are_not_matched(self) -> None:
+        """
+        GIVEN:
+            - A suggested tag name that exists, but is owned by someone else
+        WHEN:
+            - The action is applied
+        THEN:
+            - It is not assigned, because the document owner cannot see it
+        """
+        other = User.objects.create(username="someone-else")
+        Tag.objects.create(name="Existing Tag", owner=other)
+        action = self.make_action(
+            ai_suggestion_fields=[WorkflowAction.AISuggestionField.TAGS],
+        )
+
+        self.apply(action)
+
+        self.assertEqual(self.doc.tags.count(), 0)
+
+    def test_unparsable_dates_are_skipped(self) -> None:
+        """
+        GIVEN:
+            - Suggested dates that are not all valid
+        WHEN:
+            - The action is applied
+        THEN:
+            - The first usable date is applied and the rest ignored
+        """
+        action = self.make_action(
+            ai_suggestion_fields=[WorkflowAction.AISuggestionField.CREATED],
+            ai_overwrite_existing=True,
+        )
+
+        with mock.patch(
+            "documents.workflows.ai.get_ai_document_classification",
+            return_value={**SUGGESTIONS, "dates": ["not a date", "2019-07-04"]},
+        ):
+            changed = apply_ai_suggestions_to_document(action, self.doc)
+
+        self.doc.refresh_from_db()
+        self.assertEqual(changed, ["created"])
+        self.assertEqual(self.doc.created, datetime.date(2019, 7, 4))
@@ -43,6 +43,21 @@ def annotate_effective_content(documents: QuerySet[Document]) -> QuerySet[Docume
    )


+def ensure_effective_content(documents: QuerySet[Document]) -> QuerySet[Document]:
+    """
+    Annotates effective_content unless the queryset already carries it.
+
+    Lets a filter depend on effective_content without having to assume its
+    caller annotated one -- annotating twice under the same alias is an error,
+    and silently matching on the root document's own content instead is worse,
+    because the same filter then selects different documents depending on which
+    queryset it was handed.
+    """
+    if "effective_content" in documents.query.annotations:
+        return documents
+    return annotate_effective_content(documents)
+
+
 def sort_versions_newest_first(documents: list[Document]) -> list[Document]:
    """
    Same sorting as versions_newest_first()
@@ -230,6 +230,7 @@ from documents.tasks import train_classifier
 from documents.tasks import update_document_parent_tags
 from documents.utils import get_boolean
 from documents.versioning import VersionResolutionError
+from documents.versioning import annotate_effective_content
 from documents.versioning import get_latest_version_for_root
 from documents.versioning import get_request_version_param
 from documents.versioning import get_root_document
@@ -247,6 +248,7 @@ from paperless.serialisers import GroupSerializer
 from paperless.serialisers import UserSerializer
 from paperless.views import StandardPagination
 from paperless_ai.ai_classifier import get_ai_document_classification
+from paperless_ai.ai_classifier import get_llm_output_language
 from paperless_ai.chat import stream_chat_with_documents
 from paperless_ai.exceptions import LLMTimeoutError
 from paperless_ai.matching import extract_unmatched_names
@@ -665,20 +667,6 @@ class TagViewSet(PermissionsAwareDocumentCountMixin, ModelViewSet[Tag]):
            update_document_parent_tags(tag, new_parent)


-def _get_llm_output_language(ai_config: AIConfig, request) -> str | None:
-    output_language = ai_config.llm_output_language
-    if (
-        not output_language
-        and hasattr(request.user, "ui_settings")
-        and isinstance(
-            request.user.ui_settings.settings,
-            dict,
-        )
-    ):
-        output_language = request.user.ui_settings.settings.get("language")
-    return output_language
-
-
@extend_schema_view(**generate_object_with_permissions_schema(DocumentTypeSerializer))
 class DocumentTypeViewSet(
    PermissionsAwareDocumentCountMixin,
@@ -1542,7 +1530,10 @@ class DocumentViewSet(
        if not ai_config.ai_enabled:
            return HttpResponseBadRequest("AI is required for this feature")

-        output_language = _get_llm_output_language(ai_config=ai_config, request=request)
+        output_language = get_llm_output_language(
+            ai_config=ai_config,
+            user=request.user,
+        )
        llm_cache_backend = ":".join(
            part
            for part in (
@@ -2331,7 +2322,10 @@ class ChatStreamingView(GenericAPIView[Any]):
                id__in=permitted_document_ids(request.user),
            )

-        output_language = _get_llm_output_language(ai_config=ai_config, request=request)
+        output_language = get_llm_output_language(
+            ai_config=ai_config,
+            user=request.user,
+        )

        response = StreamingHttpResponse(
            stream_chat_with_documents(
@@ -2795,8 +2789,11 @@ class DocumentSelectionMixin:
            for key, value in filters.items()
            if key not in _TANTIVY_SEARCH_PARAM_NAMES
        }
+        # Operations are addressed to roots, a caller that wants
+        # to act on a specific version passes its id explicitly instead
        permitted_documents = Document.objects.filter(
            id__in=permitted_document_ids(user),
+            root_document__isnull=True,
        )
        # orm-filtered docs
        filtered_documents = DocumentFilterSet(
@@ -3617,8 +3614,13 @@ class GlobalSearchView(PassUserMixin):
        OBJECT_LIMIT = 3
        docs = []
        if request.user.has_perm("documents.view_document"):
-            all_docs = Document.objects.filter(
-                id__in=permitted_document_ids(request.user),
+            # Never more than OBJECT_LIMIT rows come back here, so annotating
+            # is cheap -- and without it these results show the root
+            # document's superseded content.
+            all_docs = annotate_effective_content(
+                Document.objects.filter(
+                    id__in=permitted_document_ids(request.user),
+                ),
            )
            if db_only:
                docs = all_docs.filter(title__icontains=query)[:OBJECT_LIMIT]
@@ -0,0 +1,259 @@
+import logging
+from datetime import date
+from datetime import datetime
+from typing import TypeVar
+
+from django.contrib.auth.models import User
+
+from documents.models import Correspondent
+from documents.models import Document
+from documents.models import DocumentType
+from documents.models import MatchingModel
+from documents.models import StoragePath
+from documents.models import Tag
+from documents.models import WorkflowAction
+from paperless.config import AIConfig
+from paperless_ai.ai_classifier import get_ai_document_classification
+from paperless_ai.ai_classifier import get_llm_output_language
+from paperless_ai.matching import extract_unmatched_names
+from paperless_ai.matching import match_correspondents_by_name
+from paperless_ai.matching import match_document_types_by_name
+from paperless_ai.matching import match_storage_paths_by_name
+from paperless_ai.matching import match_tags_by_name
+from paperless_ai.matching import resolve_correspondent_ids
+from paperless_ai.matching import resolve_document_type_ids
+from paperless_ai.matching import resolve_storage_path_ids
+from paperless_ai.matching import resolve_tag_ids
+
+logger = logging.getLogger("paperless.workflows.ai")
+
+AISuggestionField = WorkflowAction.AISuggestionField
+ObjT = TypeVar("ObjT", bound=MatchingModel)
+
+# Tags use m2m relation instead
+DIRECT_FIELDS: dict[str, str] = {
+    AISuggestionField.TITLE: "title",
+    AISuggestionField.CORRESPONDENT: "correspondent",
+    AISuggestionField.DOCUMENT_TYPE: "document_type",
+    AISuggestionField.STORAGE_PATH: "storage_path",
+    AISuggestionField.CREATED: "created",
+}
+
+
+def resolve_date(dates: list[str]) -> date | None:
+    """
+    First usable date out of the suggestions, which are expected as
+    YYYY-MM-DD. Document.created is a DateField, so only one can be applied.
+    """
+    for value in dates:
+        try:
+            return datetime.strptime(value, "%Y-%m-%d").date()
+        except (TypeError, ValueError):
+            logger.debug("Ignoring unparsable suggested date %s", value)
+    return None
+
+
+def resolve_object(
+    model: type[ObjT],
+    names: list[str],
+    matched: list[ObjT],
+    *,
+    create_missing: bool,
+    owner: User | None,
+) -> ObjT | None:
+    """
+    Single object from a suggestion list. The best match if there was one, else
+    optionally a newly-created object. StoragePaths are excluded.
+    """
+    if matched:
+        return matched[0]
+
+    if not create_missing or model is StoragePath:
+        return None
+
+    unmatched = extract_unmatched_names(names, matched)
+    if not unmatched:
+        return None
+
+    # (name, owner) is what MatchingModel is unique on
+    obj, created = model.objects.get_or_create(
+        name=unmatched[0][:128],
+        owner=owner,
+    )
+    if created:
+        logger.info("Created %s '%s' from AI suggestion", model.__name__, obj.name)
+    return obj
+
+
+def resolve_tags(
+    names: list[str],
+    matched: list[Tag],
+    *,
+    create_missing: bool,
+    owner: User | None,
+) -> list[Tag]:
+    """
+    Matched tags, plus newly created ones if create_missing is set.
+    """
+    tags = list(matched)
+    if not create_missing:
+        return tags
+
+    for name in extract_unmatched_names(names, matched):
+        tag, created = Tag.objects.get_or_create(
+            name=name[:128],
+            owner=owner,
+        )
+        if created:
+            logger.info("Created tag '%s' from AI suggestion", tag.name)
+        tags.append(tag)
+    return tags
+
+
+def apply_ai_suggestions_to_document(
+    action: WorkflowAction,
+    document: Document,
+    logging_group=None,
+) -> list[str]:
+    """
+    Get suggestions about `document` and write the chosen fields.
+
+    Returns the names of the fields that were actually changed.
+    """
+    selected = set(action.ai_suggestion_fields or [])
+    if not selected:
+        logger.warning(
+            "Workflow action %s has no AI suggestion fields selected, skipping",
+            action.pk,
+            extra={"group": logging_group},
+        )
+        return []
+
+    ai_config = AIConfig()
+    if not ai_config.ai_enabled:
+        logger.error(
+            "AI is not enabled, cannot apply AI suggestions for document %s",
+            document.pk,
+            extra={"group": logging_group},
+        )
+        return []
+
+    # Workflows run without a user, so we use the document owner
+    owner = document.owner
+
+    try:
+        suggestions = get_ai_document_classification(
+            document,
+            owner,
+            get_llm_output_language(ai_config, owner),
+        )
+    except ValueError:
+        # A bad AI config will not fix itself, so swallow it rather than
+        # letting the caller retry. Timeouts, rate limits, network errors etc
+        # propagate so the queued task can back off and try again.
+        logger.exception(
+            "Invalid AI configuration, cannot get suggestions for document %s",
+            document.pk,
+            extra={"group": logging_group},
+        )
+        return []
+
+    overwrite = action.ai_overwrite_existing
+    create_missing = action.ai_create_missing
+    updated_fields: list[str] = []
+
+    def should_set(field: str) -> bool:
+        # The field is selected and (overwrite or it's empty)
+        return field in selected and (
+            overwrite or getattr(document, DIRECT_FIELDS[field]) in (None, "")
+        )
+
+    if should_set(AISuggestionField.TITLE):
+        title = suggestions["title"].strip()
+        if title:
+            # title is capped at 128 characters
+            document.title = title[:128]
+            updated_fields.append("title")
+
+    if should_set(AISuggestionField.CORRESPONDENT):
+        choice = suggestions["correspondents"]
+        names = choice["new_names"]
+        correspondent = resolve_object(
+            Correspondent,
+            names,
+            resolve_correspondent_ids(choice["existing_ids"], owner)
+            + match_correspondents_by_name(names, owner),
+            create_missing=create_missing,
+            owner=owner,
+        )
+        if correspondent:
+            document.correspondent = correspondent
+            updated_fields.append("correspondent")
+
+    if should_set(AISuggestionField.DOCUMENT_TYPE):
+        choice = suggestions["document_types"]
+        names = choice["new_names"]
+        document_type = resolve_object(
+            DocumentType,
+            names,
+            resolve_document_type_ids(choice["existing_ids"], owner)
+            + match_document_types_by_name(names, owner),
+            create_missing=create_missing,
+            owner=owner,
+        )
+        if document_type:
+            document.document_type = document_type
+            updated_fields.append("document_type")
+
+    if should_set(AISuggestionField.STORAGE_PATH):
+        choice = suggestions["storage_paths"]
+        names = choice["new_names"]
+        storage_path = resolve_object(
+            StoragePath,
+            names,
+            resolve_storage_path_ids(choice["existing_ids"], owner)
+            + match_storage_paths_by_name(names, owner),
+            create_missing=create_missing,
+            owner=owner,
+        )
+        if storage_path:
+            document.storage_path = storage_path
+            updated_fields.append("storage_path")
+
+    if should_set(AISuggestionField.CREATED):
+        created = resolve_date(suggestions["dates"])
+        if created:
+            document.created = created
+            updated_fields.append("created")
+
+    if AISuggestionField.TAGS in selected:
+        choice = suggestions["tags"]
+        names = choice["new_names"]
+        tags = resolve_tags(
+            names,
+            resolve_tag_ids(choice["existing_ids"], owner)
+            + match_tags_by_name(names, owner),
+            create_missing=create_missing,
+            owner=owner,
+        )
+        if tags:
+            # Suggested tags are always added, so overwrite_existing
+            # does not really apply here
+            document.add_nested_tags(tags)
+            updated_fields.append("tags")
+
+    if updated_fields:
+        # save fields and update modified (excluding m2m tags from update_fields)
+        direct_updated_fields = [
+            field for field in updated_fields if field in DIRECT_FIELDS.values()
+        ]
+        document.save(update_fields=[*direct_updated_fields, "modified"])
+
+    logger.info(
+        "Applied AI suggestions %s to document %s",
+        updated_fields or "(none)",
+        document.pk,
+        extra={"group": logging_group},
+    )
+
+    return updated_fields
@@ -5,12 +5,13 @@ from django.conf import settings
 from django.contrib.auth.models import User

 from documents.models import Document
-from documents.permissions import permitted_object_ids
+from documents.permissions import get_objects_for_user_owner_aware
 from paperless.config import AIConfig
 from paperless_ai.base_model import ClassificationSuggestions
 from paperless_ai.base_model import TaxonomyChoiceDict
 from paperless_ai.client import AIClient
 from paperless_ai.db import db_connection_released
+from paperless_ai.indexing import _node_document_ids
 from paperless_ai.indexing import retrieve_similar_nodes
 from paperless_ai.indexing import truncate_content
 from paperless_ai.prompts.context import ClassificationPromptContext
@@ -18,9 +19,7 @@ from paperless_ai.prompts.context import LocalizationPromptContext
 from paperless_ai.prompts.context import RagContextPromptContext
 from paperless_ai.prompts.render import render_prompt
 from paperless_ai.taxonomy import AssignedMetadata
-from paperless_ai.taxonomy import SimilarDocument
 from paperless_ai.taxonomy import TaxonomyCandidates
-from paperless_ai.taxonomy import _node_document_weights
 from paperless_ai.taxonomy import build_taxonomy_candidates
 from paperless_ai.taxonomy import empty_taxonomy_candidates
 from paperless_ai.taxonomy import format_taxonomy_for_prompt
@@ -40,35 +39,6 @@ logger = logging.getLogger("paperless_ai.rag_classifier")
 TAXONOMY_CANDIDATE_TOP_K = 15


-def _fulltext_similar_documents(
-    document: Document,
-    user: User | None,
-    top_k: int,
-) -> list[SimilarDocument]:
-    """Rank-based fallback when no embedding backend is configured. Uses
-    Tantivy's "More Like This" (term-overlap similarity) instead of vector
-    similarity - cruder, but far better than no candidates at all.
-    more_like_this_ids returns only a ranked ID list, no scores, so weight is
-    synthesized from rank (descending from top_k) rather than claiming a
-    similarity magnitude that doesn't exist. A superuser is normalized to
-    ``None`` before calling, since the backend's permission filter has no
-    superuser short-circuit of its own.
-    """
-    from documents.search import get_backend
-
-    search_user = None if user is not None and user.is_superuser else user
-    backend = get_backend()
-    similar_ids = backend.more_like_this_ids(
-        document.pk,
-        user=search_user,
-        limit=top_k,
-    )
-    return [
-        SimilarDocument(document_id=doc_id, weight=float(top_k - rank))
-        for rank, doc_id in enumerate(similar_ids)
-    ]
-
-
 def get_language_name(language_code: str) -> str:
    normalized_language_code = language_code.lower()
    for code, name in settings.LANGUAGES:
@@ -77,6 +47,22 @@ def get_language_name(language_code: str) -> str:
    return language_code


+def get_llm_output_language(ai_config: AIConfig, user: User | None) -> str | None:
+    """
+    Language to localize LLM output into: the configured language, falling back
+    to the user's own UI language when unset.
+    """
+    output_language = ai_config.llm_output_language
+    if (
+        not output_language
+        and user is not None
+        and hasattr(user, "ui_settings")
+        and isinstance(user.ui_settings.settings, dict)
+    ):
+        output_language = user.ui_settings.settings.get("language")
+    return output_language
+
+
 def build_prompt_without_rag(
    document: Document,
    config: AIConfig,
@@ -161,52 +147,44 @@ def get_taxonomy_context(
    user: User | None = None,
    max_docs: int = 5,
 ) -> tuple[TaxonomyCandidates, AssignedMetadata, str]:
-    """One retrieval feeds both taxonomy candidates and RAG text context. Uses
-    vector similarity when an embedding backend is configured, otherwise
-    falls back to Tantivy full-text "More Like This" similarity - see
-    _fulltext_similar_documents. On any retrieval failure, degrades to empty
-    candidates/context rather than propagating the exception - neither a
-    vector-store outage nor a search-index issue should block classification,
-    only its context-assisted enrichment.
+    """One retrieval feeds both taxonomy candidates and RAG text context.
+    On any retrieval failure, degrades to empty candidates/context rather than
+    propagating the exception - a vector-store outage should not block
+    classification, only its RAG-assisted enrichment.
    """
    assigned = get_assigned_metadata(document, user)
-    ai_config = AIConfig()
    try:
-        if ai_config.llm_embedding_backend:
-            # None means "no restriction" to retrieve_similar_nodes. A superuser
-            # (like no user at all) can see every document, so skip materializing
-            # every visible pk into a Python list and passing it through as an IN
-            # filter: for a large library that is a wasted quadratic scan in the
-            # vector store at best, and past ~32,763 documents a hard
-            # sqlite3.OperationalError (SQLite's bound-parameter limit) at worst.
-            # permitted_object_ids() has its own superuser shortcut that would
-            # return every Document's id anyway, so this changes nothing about
-            # which documents are considered -- only how we get there.
-            visible_document_ids = (
-                None
-                if user is None or user.is_superuser
-                else list(permitted_object_ids(user, Document, "view_document"))
-            )
-            nodes = retrieve_similar_nodes(
-                document,
-                top_k=TAXONOMY_CANDIDATE_TOP_K,
-                document_ids=visible_document_ids,
-            )
-            similar_documents = _node_document_weights(nodes)
-        else:
-            # See _fulltext_similar_documents: it applies its own permission
-            # filter via `user`, so no visible-document-id list is needed here.
-            similar_documents = _fulltext_similar_documents(
-                document,
-                user,
-                top_k=TAXONOMY_CANDIDATE_TOP_K,
+        # None means "no restriction" to retrieve_similar_nodes. A superuser
+        # (like no user at all) can see every document, so skip materializing
+        # every visible pk into a Python list and passing it through as an IN
+        # filter: for a large library that is a wasted quadratic scan in the
+        # vector store at best, and past ~32,763 documents a hard
+        # sqlite3.OperationalError (SQLite's bound-parameter limit) at worst.
+        # get_objects_for_user_owner_aware() would return every Document for a
+        # superuser anyway (guardian's own with_superuser shortcut), so this
+        # changes nothing about which documents are considered -- only how we
+        # get there.
+        visible_document_ids = (
+            None
+            if user is None or user.is_superuser
+            else list(
+                get_objects_for_user_owner_aware(
+                    user,
+                    "view_document",
+                    Document,
+                ).values_list("pk", flat=True),
            )
+        )
+        nodes = retrieve_similar_nodes(
+            document,
+            top_k=TAXONOMY_CANDIDATE_TOP_K,
+            document_ids=visible_document_ids,
+        )

-        candidates = build_taxonomy_candidates(similar_documents, user)
+        candidates = build_taxonomy_candidates(nodes, user)

-        similar_doc_ids = [s["document_id"] for s in similar_documents]
        similar_docs = list(
-            Document.objects.filter(pk__in=similar_doc_ids)[:max_docs],
+            Document.objects.filter(pk__in=_node_document_ids(nodes))[:max_docs],
        )
        context_blocks = []
        for similar in similar_docs:
@@ -215,8 +193,8 @@ def get_taxonomy_context(
            context_blocks.append(f"TITLE: {title}\n{text}")
    except Exception:
        logger.exception(
-            "Failed to retrieve similar-document context for document %s; "
-            "continuing without taxonomy candidates or similar-document context.",
+            "Failed to retrieve RAG neighbours for document %s; continuing "
+            "without taxonomy candidates or similar-document context.",
            document.pk,
        )
        return empty_taxonomy_candidates(), assigned, ""
@@ -299,14 +277,23 @@ def get_ai_document_classification(
 ) -> ClassificationSuggestions:
    ai_config = AIConfig()

-    candidates, assigned, context = get_taxonomy_context(document, user)
-    prompt = build_prompt_with_rag(
-        document,
-        ai_config,
-        candidates=candidates,
-        assigned=assigned,
-        context=context,
-    )
+    if ai_config.llm_embedding_backend:
+        candidates, assigned, context = get_taxonomy_context(document, user)
+        prompt = build_prompt_with_rag(
+            document,
+            ai_config,
+            candidates=candidates,
+            assigned=assigned,
+            context=context,
+        )
+    else:
+        candidates = empty_taxonomy_candidates()
+        prompt = build_prompt_without_rag(
+            document,
+            ai_config,
+            candidates=candidates,
+            assigned=get_assigned_metadata(document, user),
+        )

    client = AIClient()
    # Hand the pooled DB connection back while the (slow) LLM query runs so it
@@ -11,7 +11,7 @@ from documents.models import Correspondent
 from documents.models import DocumentType
 from documents.models import StoragePath
 from documents.models import Tag
-from documents.permissions import get_objects_for_user_owner_aware
+from documents.permissions import permitted_object_ids
 from documents.permissions import restrict_queryset_to_visible

 MATCH_THRESHOLD = 0.8
@@ -63,30 +63,40 @@ def resolve_storage_path_ids(ids: list[int], user: User | None) -> list[StorageP

 def _match_by_name(
    names: list[str],
-    user: User,
+    user: User | None,
    model: type[ModelT],
    perm: str,
 ) -> list[ModelT]:
-    queryset = get_objects_for_user_owner_aware(user, [perm], model)
+    # A workflow may have no user. In that case permitted_object_ids limits
+    # matching to unowned objects, avoiding another user's private taxonomy.
+    queryset = model.objects.filter(
+        pk__in=permitted_object_ids(user, model, perm),
+    )
    return _match_names_to_queryset(names, queryset)


-def match_tags_by_name(names: list[str], user: User) -> list[Tag]:
+def match_tags_by_name(names: list[str], user: User | None) -> list[Tag]:
    return _match_by_name(names, user, Tag, "view_tag")


 def match_correspondents_by_name(
    names: list[str],
-    user: User,
+    user: User | None,
 ) -> list[Correspondent]:
    return _match_by_name(names, user, Correspondent, "view_correspondent")


-def match_document_types_by_name(names: list[str], user: User) -> list[DocumentType]:
+def match_document_types_by_name(
+    names: list[str],
+    user: User | None,
+) -> list[DocumentType]:
    return _match_by_name(names, user, DocumentType, "view_documenttype")


-def match_storage_paths_by_name(names: list[str], user: User) -> list[StoragePath]:
+def match_storage_paths_by_name(
+    names: list[str],
+    user: User | None,
+) -> list[StoragePath]:
    return _match_by_name(names, user, StoragePath, "view_storagepath")


@@ -33,11 +33,6 @@ class TaxonomyCandidate(TypedDict):
    weight: float


-class SimilarDocument(TypedDict):
-    document_id: int
-    weight: float
-
-
 class TaxonomyCandidates(TypedDict):
    tags: list[TaxonomyCandidate]
    document_types: list[TaxonomyCandidate]
@@ -110,10 +105,10 @@ def get_assigned_metadata(document: Document, user: User | None) -> AssignedMeta
    )


-def _node_document_weights(nodes: list["NodeWithScore"]) -> list[SimilarDocument]:
-    """Sum each node's similarity score into its document_id (a document can
-    appear via multiple chunks/nodes) and return one SimilarDocument per
-    distinct document_id."""
+def _node_document_weights(nodes: list["NodeWithScore"]) -> dict[int, float]:
+    """document_id -> that node's similarity score, summed if a document_id
+    appears more than once across the retrieved nodes (e.g. multiple chunks
+    of the same source document)."""
    weights: dict[int, float] = defaultdict(float)
    for node in nodes:
        document_id = node.metadata.get("document_id")
@@ -126,10 +121,7 @@ def _node_document_weights(nodes: list["NodeWithScore"]) -> list[SimilarDocument
            weights[int(document_id)] += float(node.score or 0.0)
        except (TypeError, ValueError):  # pragma: no cover
            continue
-    return [
-        SimilarDocument(document_id=document_id, weight=weight)
-        for document_id, weight in weights.items()
-    ]
+    return weights


 def _visible_ranked_candidates(
@@ -165,25 +157,20 @@ def _visible_ranked_candidates(


 def build_taxonomy_candidates(
-    similar_documents: list[SimilarDocument],
+    nodes: list["NodeWithScore"],
    user: User | None,
 ) -> TaxonomyCandidates:
-    """Resolve each similar document's id to a live Document, read its
-    *current* tags/type/correspondent/storage_path via the ORM (never any
-    possibly-stale names an adapter's source might have cached), weight each
-    distinct taxonomy object by aggregate similarity weight, permission-filter
+    """Resolve each neighbour node's document_id to a live Document, read its
+    *current* tags/type/correspondent/storage_path via the ORM (never the
+    possibly-stale names cached in vector-index node metadata), weight each
+    distinct taxonomy object by aggregate neighbour similarity, permission-filter
    against what ``user`` can see, and return each category ranked by weight
-    and capped. ``similar_documents`` may come from either the vector-RAG
-    adapter or the full-text fallback adapter - both produce this same shape.
+    and capped.
    """
-    if not similar_documents:
-        return empty_taxonomy_candidates()

-    # Both adapters guarantee at most one SimilarDocument per document_id, so
-    # this never silently drops a duplicate's weight.
-    document_weights: dict[int, float] = {
-        s["document_id"]: s["weight"] for s in similar_documents
-    }
+    document_weights = _node_document_weights(nodes)
+    if not document_weights:
+        return empty_taxonomy_candidates()

    # Only .tags.all() needs prefetching (a reverse M2M, one extra query for
    # the whole batch). document_type/correspondent/storage_path are read
@@ -1,4 +1,3 @@
-from collections.abc import Generator
 from types import SimpleNamespace
 from unittest.mock import MagicMock
 from unittest.mock import patch
@@ -8,13 +7,10 @@ import pytest_mock
 from django.test import override_settings

 from documents.models import Document
-from documents.search import TantivyBackend
 from documents.tests.factories import DocumentFactory
 from documents.tests.factories import TagFactory
 from documents.tests.factories import UserFactory
 from paperless.config import AIConfig
-from paperless_ai.ai_classifier import TAXONOMY_CANDIDATE_TOP_K
-from paperless_ai.ai_classifier import _fulltext_similar_documents
 from paperless_ai.ai_classifier import _restrict_to_shown_candidates
 from paperless_ai.ai_classifier import build_localization_prompt
 from paperless_ai.ai_classifier import build_prompt_with_rag
@@ -24,7 +20,6 @@ from paperless_ai.ai_classifier import get_language_name
 from paperless_ai.ai_classifier import get_taxonomy_context
 from paperless_ai.base_model import ClassificationSuggestions
 from paperless_ai.base_model import TaxonomyChoiceDict
-from paperless_ai.taxonomy import SimilarDocument
 from paperless_ai.taxonomy import TaxonomyCandidate
 from paperless_ai.taxonomy import TaxonomyCandidates
 from paperless_ai.taxonomy import empty_taxonomy_candidates
@@ -209,10 +204,12 @@ def test_use_rag_if_configured(

@pytest.mark.django_db
@patch("paperless_ai.client.AIClient.run_llm_query")
-@patch("paperless_ai.ai_classifier.build_prompt_with_rag")
+@patch("paperless_ai.ai_classifier.build_prompt_without_rag")
+@patch("paperless_ai.ai_classifier.AIConfig")
@override_settings(LLM_BACKEND="ollama", LLM_MODEL="some_model")
-def test_use_rag_prompt_even_without_embedding_backend(
-    mock_build_prompt_with_rag,
+def test_use_without_rag_if_not_configured(
+    mock_ai_config,
+    mock_build_prompt_without_rag,
    mock_run_llm_query,
    mock_document,
 ):
@@ -222,13 +219,13 @@ def test_use_rag_prompt_even_without_embedding_backend(
    WHEN:
        - get_ai_document_classification() is called
    THEN:
-        - The RAG-context prompt builder is still used (fed by the full-text
-          fallback's context/candidates instead of the vector store's)
+        - The non-RAG prompt builder is used
    """
-    mock_build_prompt_with_rag.return_value = "Prompt with RAG"
+    mock_ai_config.return_value.llm_embedding_backend = None
+    mock_build_prompt_without_rag.return_value = "Prompt without RAG"
    mock_run_llm_query.return_value = NESTED_SUGGESTIONS
    get_ai_document_classification(mock_document)
-    mock_build_prompt_with_rag.assert_called_once()
+    mock_build_prompt_without_rag.assert_called_once()


@pytest.mark.django_db
@@ -306,7 +303,6 @@ def test_build_localization_prompt_preserves_unicode_characters():


@pytest.mark.django_db
-@override_settings(LLM_EMBEDDING_BACKEND="huggingface")
 def test_get_taxonomy_context_assembles_rag_text_and_candidates():
    """
    GIVEN:
@@ -348,7 +344,6 @@ def test_get_taxonomy_context_assembles_rag_text_and_candidates():


@pytest.mark.django_db
-@override_settings(LLM_EMBEDDING_BACKEND="huggingface")
 def test_get_taxonomy_context_no_similar_docs():
    """
    GIVEN:
@@ -372,67 +367,6 @@ def test_get_taxonomy_context_no_similar_docs():
    }


-@pytest.mark.django_db
-def test_get_taxonomy_context_uses_fulltext_fallback_when_no_embedding_backend(
-    mocker: pytest_mock.MockerFixture,
-) -> None:
-    """
-    GIVEN:
-        - No LLM embedding backend is configured (the default test settings)
-    WHEN:
-        - get_taxonomy_context() is called
-    THEN:
-        - _fulltext_similar_documents() is called with the document, the user
-          and TAXONOMY_CANDIDATE_TOP_K
-        - retrieve_similar_nodes() (the vector path) is never called
-    """
-    document = DocumentFactory.create(content="Some content")
-    mock_fulltext = mocker.patch(
-        "paperless_ai.ai_classifier._fulltext_similar_documents",
-        return_value=[],
-    )
-    mock_retrieve = mocker.patch("paperless_ai.ai_classifier.retrieve_similar_nodes")
-
-    get_taxonomy_context(document, user=None)
-
-    mock_fulltext.assert_called_once_with(
-        document,
-        None,
-        top_k=TAXONOMY_CANDIDATE_TOP_K,
-    )
-    mock_retrieve.assert_not_called()
-
-
-@pytest.mark.django_db
-@override_settings(LLM_EMBEDDING_BACKEND="huggingface")
-def test_get_taxonomy_context_uses_vector_path_when_embedding_backend_configured(
-    mocker: pytest_mock.MockerFixture,
-) -> None:
-    """
-    GIVEN:
-        - An LLM embedding backend is configured
-    WHEN:
-        - get_taxonomy_context() is called
-    THEN:
-        - retrieve_similar_nodes() (the vector path) is called
-        - _fulltext_similar_documents() (the no-embedding-backend fallback)
-          is never called
-    """
-    document = DocumentFactory.create(content="Some content")
-    mock_retrieve = mocker.patch(
-        "paperless_ai.ai_classifier.retrieve_similar_nodes",
-        return_value=[],
-    )
-    mock_fulltext = mocker.patch(
-        "paperless_ai.ai_classifier._fulltext_similar_documents",
-    )
-
-    get_taxonomy_context(document, user=None)
-
-    mock_retrieve.assert_called_once()
-    mock_fulltext.assert_not_called()
-
-
 class TestGetTaxonomyContextVisibility:
    """get_taxonomy_context must not materialize every visible document id
    for a user who can already see the whole library: a superuser (like no
@@ -445,7 +379,6 @@ class TestGetTaxonomyContextVisibility:
    """

    @pytest.mark.django_db
-    @override_settings(LLM_EMBEDDING_BACKEND="huggingface")
    def test_skips_permission_lookup_for_superuser(
        self,
        mocker: pytest_mock.MockerFixture,
@@ -464,18 +397,17 @@ class TestGetTaxonomyContextVisibility:
            "paperless_ai.ai_classifier.retrieve_similar_nodes",
            return_value=[],
        )
-        mock_permitted = mocker.patch(
-            "paperless_ai.ai_classifier.permitted_object_ids",
+        mock_get_objects = mocker.patch(
+            "paperless_ai.ai_classifier.get_objects_for_user_owner_aware",
        )
        user = UserFactory.create(is_superuser=True)

        get_taxonomy_context(document, user)

-        mock_permitted.assert_not_called()
+        mock_get_objects.assert_not_called()
        assert mock_retrieve.call_args.kwargs["document_ids"] is None

    @pytest.mark.django_db
-    @override_settings(LLM_EMBEDDING_BACKEND="huggingface")
    def test_skips_permission_lookup_when_no_user(
        self,
        mocker: pytest_mock.MockerFixture,
@@ -494,17 +426,16 @@ class TestGetTaxonomyContextVisibility:
            "paperless_ai.ai_classifier.retrieve_similar_nodes",
            return_value=[],
        )
-        mock_permitted = mocker.patch(
-            "paperless_ai.ai_classifier.permitted_object_ids",
+        mock_get_objects = mocker.patch(
+            "paperless_ai.ai_classifier.get_objects_for_user_owner_aware",
        )

        get_taxonomy_context(document, None)

-        mock_permitted.assert_not_called()
+        mock_get_objects.assert_not_called()
        assert mock_retrieve.call_args.kwargs["document_ids"] is None

    @pytest.mark.django_db
-    @override_settings(LLM_EMBEDDING_BACKEND="huggingface")
    def test_restricts_to_visible_documents_for_non_superuser(
        self,
        mocker: pytest_mock.MockerFixture,
@@ -515,7 +446,7 @@ class TestGetTaxonomyContextVisibility:
        WHEN:
            - get_taxonomy_context() is called
        THEN:
-            - The user's permitted document ids are looked up and passed to
+            - The user's visible document ids are looked up and passed to
              retrieve_similar_nodes() as a restriction
        """
        document = DocumentFactory.create(content="Some content")
@@ -523,185 +454,21 @@ class TestGetTaxonomyContextVisibility:
            "paperless_ai.ai_classifier.retrieve_similar_nodes",
            return_value=[],
        )
-        mock_permitted = mocker.patch(
-            "paperless_ai.ai_classifier.permitted_object_ids",
-            return_value=[1, 2, 3],
+        mock_queryset = mocker.MagicMock()
+        mock_queryset.values_list.return_value = [1, 2, 3]
+        mock_get_objects = mocker.patch(
+            "paperless_ai.ai_classifier.get_objects_for_user_owner_aware",
+            return_value=mock_queryset,
        )
        user = UserFactory.create(is_superuser=False)

        get_taxonomy_context(document, user)

-        mock_permitted.assert_called_once_with(user, Document, "view_document")
+        mock_get_objects.assert_called_once_with(user, "view_document", Document)
        assert mock_retrieve.call_args.kwargs["document_ids"] == [1, 2, 3]


@pytest.mark.django_db
-class TestFulltextSimilarDocuments:
-    """_fulltext_similar_documents is the no-embedding-backend fallback: it
-    asks the Tantivy full-text index for "More Like This" neighbours instead
-    of the vector store, and synthesizes a rank-based weight since Tantivy's
-    more_like_this_ids returns only an ordered id list, no scores.
-    """
-
-    @pytest.fixture
-    def fulltext_backend(
-        self,
-        mocker: pytest_mock.MockerFixture,
-    ) -> Generator[TantivyBackend, None, None]:
-        """An in-memory Tantivy backend, wired up as the module-level
-        singleton _fulltext_similar_documents resolves via get_backend()."""
-        backend = TantivyBackend(path=None)
-        backend.open()
-        mocker.patch("documents.search.get_backend", return_value=backend)
-        try:
-            yield backend
-        finally:
-            backend.close()
-
-    def test_ranks_by_rank_based_weight_descending(
-        self,
-        fulltext_backend: TantivyBackend,
-    ) -> None:
-        """
-        GIVEN:
-            - A source document and two similar documents indexed in Tantivy
-        WHEN:
-            - _fulltext_similar_documents() is called
-        THEN:
-            - Each result's weight reflects its rank (first result weighted
-              higher than the second), not a raw similarity score
-        """
-        source = DocumentFactory.create(content="quarterly financial report details")
-        first = DocumentFactory.create(content="quarterly financial report details")
-        second = DocumentFactory.create(content="financial report")
-        for doc in (source, first, second):
-            fulltext_backend.add_or_update(doc)
-
-        result = _fulltext_similar_documents(source, user=None, top_k=5)
-
-        assert len(result) == 2
-        weight_by_id = {s["document_id"]: s["weight"] for s in result}
-        assert weight_by_id[first.pk] > weight_by_id[second.pk]
-
-    def test_excludes_source_document(
-        self,
-        fulltext_backend: TantivyBackend,
-    ) -> None:
-        """
-        GIVEN:
-            - A source document indexed in Tantivy with no other documents
-        WHEN:
-            - _fulltext_similar_documents() is called
-        THEN:
-            - An empty list is returned - the source document is never its
-              own similar document
-        """
-        source = DocumentFactory.create(content="unique unrelated content")
-        fulltext_backend.add_or_update(source)
-
-        result = _fulltext_similar_documents(source, user=None, top_k=5)
-
-        assert result == []
-
-    def test_empty_index_returns_empty_list(
-        self,
-        fulltext_backend: TantivyBackend,
-    ) -> None:
-        """
-        GIVEN:
-            - A document that has never been indexed (fresh/empty Tantivy index)
-        WHEN:
-            - _fulltext_similar_documents() is called
-        THEN:
-            - An empty list is returned rather than raising
-        """
-        source = DocumentFactory.create(content="never indexed")
-
-        result = _fulltext_similar_documents(source, user=None, top_k=5)
-
-        assert result == []
-
-    def test_respects_top_k_limit(
-        self,
-        fulltext_backend: TantivyBackend,
-    ) -> None:
-        """
-        GIVEN:
-            - A source document and four similar documents indexed
-        WHEN:
-            - _fulltext_similar_documents() is called with top_k=2
-        THEN:
-            - At most 2 results are returned
-        """
-        source = DocumentFactory.create(content="shared overlapping keyword text")
-        for _ in range(4):
-            fulltext_backend.add_or_update(
-                DocumentFactory.create(content="shared overlapping keyword text"),
-            )
-
-        result = _fulltext_similar_documents(source, user=None, top_k=2)
-
-        assert len(result) <= 2
-
-    def test_result_shape_is_similar_document(
-        self,
-        fulltext_backend: TantivyBackend,
-    ) -> None:
-        """
-        GIVEN:
-            - A source document and one similar document indexed
-        WHEN:
-            - _fulltext_similar_documents() is called
-        THEN:
-            - Each result is a SimilarDocument (document_id + weight only)
-        """
-        source = DocumentFactory.create(content="shared content phrase")
-        other = DocumentFactory.create(content="shared content phrase")
-        fulltext_backend.add_or_update(source)
-        fulltext_backend.add_or_update(other)
-
-        result = _fulltext_similar_documents(source, user=None, top_k=5)
-
-        # rank 0 (the only/best result) with top_k=5 -> weight = top_k - rank = 5.0,
-        # per the "first result gets top_k, the last gets 1" formula.
-        assert result == [SimilarDocument(document_id=other.pk, weight=5.0)]
-
-    def test_superuser_sees_other_users_documents(
-        self,
-        fulltext_backend: TantivyBackend,
-    ) -> None:
-        """
-        GIVEN:
-            - A source document owned by one user and a similar document
-              owned by a different user, with no sharing between them
-        WHEN:
-            - _fulltext_similar_documents() is called with a superuser
-        THEN:
-            - The other user's document is still returned as a similar
-              document - a superuser must not be narrowed by the backend's
-              owner-based permission filter
-        """
-        owner = UserFactory.create()
-        other_owner = UserFactory.create()
-        superuser = UserFactory.create(is_superuser=True)
-        source = DocumentFactory.create(
-            content="shared content phrase",
-            owner=owner,
-        )
-        other = DocumentFactory.create(
-            content="shared content phrase",
-            owner=other_owner,
-        )
-        fulltext_backend.add_or_update(source)
-        fulltext_backend.add_or_update(other)
-
-        result = _fulltext_similar_documents(source, user=superuser, top_k=5)
-
-        assert [s["document_id"] for s in result] == [other.pk]
-
-
-@pytest.mark.django_db
-@override_settings(LLM_EMBEDDING_BACKEND="huggingface")
@patch("paperless_ai.ai_classifier.retrieve_similar_nodes")
 def test_get_taxonomy_context_retrieval_failure_degrades_to_no_hints(mock_retrieve):
    """
@@ -728,7 +495,6 @@ def test_get_taxonomy_context_retrieval_failure_degrades_to_no_hints(mock_retrie


@pytest.mark.django_db
-@override_settings(LLM_EMBEDDING_BACKEND="huggingface")
@patch("paperless_ai.ai_classifier.build_taxonomy_candidates")
@patch("paperless_ai.ai_classifier.retrieve_similar_nodes")
 def test_get_taxonomy_context_candidate_building_failure_degrades_to_no_hints(
@@ -1,5 +1,4 @@
 from collections.abc import Callable
-from unittest.mock import patch

 import pytest
 import pytest_mock
@@ -45,33 +44,25 @@ class TestAIMatching(TestCase):
        self.storage_path1 = StoragePath.objects.create(name="Test Storage Path 1")
        self.storage_path2 = StoragePath.objects.create(name="Test Storage Path 2")

-    @patch("paperless_ai.matching.get_objects_for_user_owner_aware")
-    def test_match_tags_by_name(self, mock_get_objects) -> None:
-        mock_get_objects.return_value = Tag.objects.all()
+    def test_match_tags_by_name(self) -> None:
        names = ["Test Tag 1", "Nonexistent Tag"]
        result = match_tags_by_name(names, user=None)
        self.assertEqual(len(result), 1)
        self.assertEqual(result[0].name, "Test Tag 1")

-    @patch("paperless_ai.matching.get_objects_for_user_owner_aware")
-    def test_match_correspondents_by_name(self, mock_get_objects) -> None:
-        mock_get_objects.return_value = Correspondent.objects.all()
+    def test_match_correspondents_by_name(self) -> None:
        names = ["Test Correspondent 1", "Nonexistent Correspondent"]
        result = match_correspondents_by_name(names, user=None)
        self.assertEqual(len(result), 1)
        self.assertEqual(result[0].name, "Test Correspondent 1")

-    @patch("paperless_ai.matching.get_objects_for_user_owner_aware")
-    def test_match_document_types_by_name(self, mock_get_objects) -> None:
-        mock_get_objects.return_value = DocumentType.objects.all()
+    def test_match_document_types_by_name(self) -> None:
        names = ["Test Document Type 1", "Nonexistent Document Type"]
        result = match_document_types_by_name(names, user=None)
        self.assertEqual(len(result), 1)
        self.assertEqual(result[0].name, "Test Document Type 1")

-    @patch("paperless_ai.matching.get_objects_for_user_owner_aware")
-    def test_match_storage_paths_by_name(self, mock_get_objects) -> None:
-        mock_get_objects.return_value = StoragePath.objects.all()
+    def test_match_storage_paths_by_name(self) -> None:
        names = ["Test Storage Path 1", "Nonexistent Storage Path"]
        result = match_storage_paths_by_name(names, user=None)
        self.assertEqual(len(result), 1)
@@ -83,16 +74,12 @@ class TestAIMatching(TestCase):
        unmatched_names = extract_unmatched_names(llm_names, matched_objects)
        self.assertEqual(unmatched_names, ["Nonexistent Tag"])

-    @patch("paperless_ai.matching.get_objects_for_user_owner_aware")
-    def test_match_tags_by_name_with_empty_names(self, mock_get_objects) -> None:
-        mock_get_objects.return_value = Tag.objects.all()
+    def test_match_tags_by_name_with_empty_names(self) -> None:
        names = [None, "", "   "]
        result = match_tags_by_name(names, user=None)
        self.assertEqual(result, [])

-    @patch("paperless_ai.matching.get_objects_for_user_owner_aware")
-    def test_match_tags_with_fuzzy_matching(self, mock_get_objects) -> None:
-        mock_get_objects.return_value = Tag.objects.all()
+    def test_match_tags_with_fuzzy_matching(self) -> None:
        names = ["Test Taag 1", "Teest Tag 2"]
        result = match_tags_by_name(names, user=None)
        self.assertEqual(len(result), 2)
@@ -1,4 +1,5 @@
 import json
+from types import SimpleNamespace

 import pytest
 import pytest_mock
@@ -10,7 +11,6 @@ from documents.tests.factories import StoragePathFactory
 from documents.tests.factories import TagFactory
 from documents.tests.factories import UserFactory
 from paperless_ai.taxonomy import AssignedMetadata
-from paperless_ai.taxonomy import SimilarDocument
 from paperless_ai.taxonomy import TaxonomyCandidates
 from paperless_ai.taxonomy import build_taxonomy_candidates
 from paperless_ai.taxonomy import format_taxonomy_for_prompt
@@ -132,8 +132,9 @@ class TestGetAssignedMetadata:
        assert result["tags"] == ["Owned By Someone Else"]


-def make_similar(document_id: int, weight: float) -> SimilarDocument:
-    return SimilarDocument(document_id=document_id, weight=weight)
+def make_node(document_id: int, score: float) -> SimpleNamespace:
+    """A stand-in for NodeWithScore: only ``.metadata``/``.score`` are read."""
+    return SimpleNamespace(metadata={"document_id": str(document_id)}, score=score)


@pytest.mark.django_db
@@ -169,9 +170,9 @@ class TestBuildTaxonomyCandidates:
        doc_a.tags.add(tag)
        doc_b = DocumentFactory.create()
        doc_b.tags.add(tag)
-        similar_documents = [make_similar(doc_a.pk, 0.9), make_similar(doc_b.pk, 0.4)]
+        nodes = [make_node(doc_a.pk, 0.9), make_node(doc_b.pk, 0.4)]

-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)

        assert len(result["tags"]) == 1
        assert result["tags"][0]["id"] == tag.pk
@@ -196,9 +197,9 @@ class TestBuildTaxonomyCandidates:
        document.tags.add(tag)
        tag.name = "New Name"
        tag.save()
-        similar_documents = [make_similar(document.pk, 0.5)]
+        nodes = [make_node(document.pk, 0.5)]

-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)

        assert result["tags"][0]["name"] == "New Name"

@@ -218,9 +219,9 @@ class TestBuildTaxonomyCandidates:
        document = DocumentFactory.create()
        document.tags.add(tag)
        tag.delete()
-        similar_documents = [make_similar(document.pk, 0.5)]
+        nodes = [make_node(document.pk, 0.5)]

-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)

        assert result["tags"] == []

@@ -239,12 +240,9 @@ class TestBuildTaxonomyCandidates:
        strong_doc.tags.add(strong_tag)
        weak_doc = DocumentFactory.create()
        weak_doc.tags.add(weak_tag)
-        similar_documents = [
-            make_similar(strong_doc.pk, 0.9),
-            make_similar(weak_doc.pk, 0.1),
-        ]
+        nodes = [make_node(strong_doc.pk, 0.9), make_node(weak_doc.pk, 0.1)]

-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)

        assert [c["name"] for c in result["tags"]] == ["Strong", "Weak"]

@@ -260,9 +258,9 @@ class TestBuildTaxonomyCandidates:
        document = DocumentFactory.create()
        for i in range(15):
            document.tags.add(TagFactory.create(name=f"Tag{i}"))
-        similar_documents = [make_similar(document.pk, 0.5)]
+        nodes = [make_node(document.pk, 0.5)]

-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)

        assert len(result["tags"]) == 10

@@ -276,12 +274,12 @@ class TestBuildTaxonomyCandidates:
            - Only 5 correspondents are returned
        """
        correspondents = CorrespondentFactory.create_batch(7)
-        similar_documents = [
-            make_similar(DocumentFactory.create(correspondent=c).pk, 0.5)
+        nodes = [
+            make_node(DocumentFactory.create(correspondent=c).pk, 0.5)
            for c in correspondents
        ]

-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)

        assert len(result["correspondents"]) == 5

@@ -296,9 +294,9 @@ class TestBuildTaxonomyCandidates:
        """
        document_type = DocumentTypeFactory.create(name="Invoice")
        document = DocumentFactory.create(document_type=document_type)
-        similar_documents = [make_similar(document.pk, 0.5)]
+        nodes = [make_node(document.pk, 0.5)]

-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)

        assert len(result["document_types"]) == 1
        assert result["document_types"][0]["id"] == document_type.pk
@@ -314,12 +312,12 @@ class TestBuildTaxonomyCandidates:
            - Only 5 document_types are returned
        """
        document_types = DocumentTypeFactory.create_batch(7)
-        similar_documents = [
-            make_similar(DocumentFactory.create(document_type=dt).pk, 0.5)
+        nodes = [
+            make_node(DocumentFactory.create(document_type=dt).pk, 0.5)
            for dt in document_types
        ]

-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)

        assert len(result["document_types"]) == 5

@@ -334,9 +332,9 @@ class TestBuildTaxonomyCandidates:
        """
        storage_path = StoragePathFactory.create(name="Invoices")
        document = DocumentFactory.create(storage_path=storage_path)
-        similar_documents = [make_similar(document.pk, 0.5)]
+        nodes = [make_node(document.pk, 0.5)]

-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)

        assert len(result["storage_paths"]) == 1
        assert result["storage_paths"][0]["id"] == storage_path.pk
@@ -352,12 +350,12 @@ class TestBuildTaxonomyCandidates:
            - Only 5 storage_paths are returned
        """
        storage_paths = StoragePathFactory.create_batch(7)
-        similar_documents = [
-            make_similar(DocumentFactory.create(storage_path=sp).pk, 0.5)
+        nodes = [
+            make_node(DocumentFactory.create(storage_path=sp).pk, 0.5)
            for sp in storage_paths
        ]

-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)

        assert len(result["storage_paths"]) == 5

@@ -377,14 +375,14 @@ class TestBuildTaxonomyCandidates:
        tag = TagFactory.create(name="Restricted")
        document = DocumentFactory.create()
        document.tags.add(tag)
-        similar_documents = [make_similar(document.pk, 0.5)]
+        nodes = [make_node(document.pk, 0.5)]
        user = UserFactory.create()
        mocker.patch(
            "documents.permissions.permitted_object_ids",
            return_value=[],  # user cannot see this tag
        )

-        result = build_taxonomy_candidates(similar_documents, user=user)
+        result = build_taxonomy_candidates(nodes, user=user)

        assert result["tags"] == []

@@ -414,10 +412,10 @@ class TestBuildTaxonomyCandidates:
        tag.save()
        document = DocumentFactory.create()
        document.tags.add(tag)
-        similar_documents = [make_similar(document.pk, 0.5)]
+        nodes = [make_node(document.pk, 0.5)]
        spy = mocker.patch("documents.permissions.permitted_object_ids")

-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)

        assert result["tags"][0]["name"] == "Owned"
        spy.assert_not_called()
Author	SHA1	Message	Date
shamoon	a13e8a59dd	Fix: annotate effective content in the filters rather than relying on callers	2026-08-26 09:54:18 -07:00
GitHub Actions	cafed919a1	Auto translate strings	2026-08-26 16:50:29 +00:00
shamoonandGitHub	549afde1fd	Enhancement: Apply AI suggestions workflow action (#13639 )	2026-08-26 16:41:21 +00:00
GitHub Actions	a8576d827e	Auto translate strings	2026-08-26 14:29:05 +00:00
shamoonandGitHub	7330c4d9cb	Fix: exclude version documents from bulk edit "all" (#13791 )	2026-08-26 14:27:31 +00:00