Collapse near-duplicate outbound guard tests into parametrized cases

Several tests in test_network.py and test_network_integration.py had grown into near-copies of each other, differing only in the URL or host they fed the guard while asserting the same thing. That made the suite noisier to read and to extend than the behaviour it covers warrants. Merge the invalid-host rejection tests into one parametrized test over all seven unusual host forms, and likewise for their allow_internal counterparts that assert nothing is rejected. Merge the resolve_public_addresses tests that only check the host reaches the resolver unchanged, and the ones that check a private answer blocks regardless of how the host was spelled, keeping the stricter resolver call assertion on both merged cases. Add localhost as a fourth case to the numeric-host-forms test in the transport integration tests, so the name and the non-canonical numeric spellings of loopback are covered by one test. No behaviour or assertion strength changes; this only reduces duplicated test bodies while keeping every original case addressable by its own parametrize id. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Fold the outbound guard's public-address patch into a named fixture
2026-09-28 20:30:33 +00:00 · 2026-09-22 13:49:35 -07:00 · 2026-09-22 12:56:55 -07:00 · 2026-09-22 12:49:06 -07:00 · 2026-09-22 12:49:06 -07:00 · 2026-09-22 12:49:06 -07:00
59 changed files with 3505 additions and 1279 deletions
@@ -59,6 +59,7 @@ updates:
          - "drf-*"
          - "djangorestframework"
          - "whitenoise"
+          - "bleach"
          - "jinja2"
      # Async, Task Queuing & Caching
      async-tasks:
@@ -30,7 +30,7 @@ RUN set -eux \
 # Purpose: Installs s6-overlay and rootfs
 # Comments:
 #  - Don't leave anything extra in here either
-FROM ghcr.io/astral-sh/uv:0.12.19-python3.14-trixie-slim AS s6-overlay-base
+FROM ghcr.io/astral-sh/uv:0.12.16-python3.14-trixie-slim AS s6-overlay-base

 WORKDIR /usr/src/s6

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

 WORKDIR /usr/src/paperless/src/

@@ -1,18 +0,0 @@
-#!/bin/sh
-# Source s6 container environment for interactive shells.
-# Ensures variables resolved from *_FILE secret injection are visible
-# when using 'docker exec bash'. Does not affect s6 services (those
-# use with-contenv directly). Has no effect in non-container contexts
-# because the directory will not exist.
-# Note: sh/dash shells opened via 'docker exec sh' are not covered;
-# only bash-based sessions benefit from this file.
-_pngx_contenv="/run/s6/container_environment"
-if [ -d "${_pngx_contenv}" ]; then
-	for _pngx_f in "${_pngx_contenv}"/*; do
-		[ -f "${_pngx_f}" ] || continue
-		_pngx_name=$(basename "${_pngx_f}")
-		_pngx_val=$(cat "${_pngx_f}")
-		export "${_pngx_name}=${_pngx_val}"
-	done
-fi
-unset _pngx_contenv _pngx_f _pngx_name _pngx_val
@@ -153,11 +153,8 @@ in similar existing documents, and the document chat can retrieve relevant conte

 Enable it by setting
 [`PAPERLESS_AI_LLM_EMBEDDING_BACKEND`](configuration.md#PAPERLESS_AI_LLM_EMBEDDING_BACKEND)
-(`huggingface` for fully-local embeddings, or `ollama` / `openai-like`). By default, the main
-LLM API key and endpoint are used, but an optional embedding-specific[API key](configuration.md#PAPERLESS_AI_LLM_EMBEDDING_API_KEY)
-and [endpoint](configuration.md#PAPERLESS_AI_LLM_EMBEDDING_ENDPOINT) can be configured.
-
-The index is only built when AI is enabled **and** an embedding backend is set.
+(`huggingface` for fully-local embeddings, or `ollama` / `openai-like`). The index is only
+built when AI is enabled **and** an embedding backend is set.

 The index is updated automatically on a schedule controlled by
 [`PAPERLESS_LLM_INDEX_TASK_CRON`](configuration.md#PAPERLESS_LLM_INDEX_TASK_CRON) (daily by
@@ -1576,6 +1576,9 @@ ports.
 #### [`PAPERLESS_WEBHOOKS_ALLOW_INTERNAL_REQUESTS=<bool>`](#PAPERLESS_WEBHOOKS_ALLOW_INTERNAL_REQUESTS) {#PAPERLESS_WEBHOOKS_ALLOW_INTERNAL_REQUESTS}

 : If set to false, webhooks cannot be sent to internal URLs (e.g., localhost).
+A hostname is blocked if any of the addresses it resolves to is non-public.
+Webhook requests connect directly, without using the `HTTP_PROXY` or
+`HTTPS_PROXY` environment variables, and never follow redirects.

    Defaults to true, which allows internal requests.

@@ -1584,7 +1587,7 @@ ports.
 #### [`PAPERLESS_EMAIL_ALLOW_INTERNAL_HOSTS=<bool>`](#PAPERLESS_EMAIL_ALLOW_INTERNAL_HOSTS) {#PAPERLESS_EMAIL_ALLOW_INTERNAL_HOSTS}

 : If set to false, incoming mail account connections are blocked when the
-configured IMAP hostname resolves to a non-public address (for example,
+configured IMAP hostname resolves to any non-public address (for example,
 localhost, link-local, or RFC1918 private ranges).

    Defaults to true, which allows internal hosts.
@@ -2133,13 +2136,6 @@ for language and resource considerations.

    Defaults to None.

-#### [`PAPERLESS_AI_LLM_EMBEDDING_API_KEY=<str>`](#PAPERLESS_AI_LLM_EMBEDDING_API_KEY) {#PAPERLESS_AI_LLM_EMBEDDING_API_KEY}
-
-: The API key to use for the embedding backend. If not supplied, embeddings use
-`PAPERLESS_AI_LLM_API_KEY`.
-
-    Defaults to None.
-
 #### [`PAPERLESS_AI_LLM_EMBEDDING_ENDPOINT=<str>`](#PAPERLESS_AI_LLM_EMBEDDING_ENDPOINT) {#PAPERLESS_AI_LLM_EMBEDDING_ENDPOINT}

 : The endpoint / url to use for the embedding backend. If not supplied, embeddings use
@@ -2221,22 +2217,11 @@ used with the OpenAI-compatible backend to target a custom provider or local gat
 #### [`PAPERLESS_AI_LLM_ALLOW_INTERNAL_ENDPOINTS=<bool>`](#PAPERLESS_AI_LLM_ALLOW_INTERNAL_ENDPOINTS) {#PAPERLESS_AI_LLM_ALLOW_INTERNAL_ENDPOINTS}

 : If set to false, Paperless blocks AI endpoint URLs that resolve to non-public addresses (e.g., localhost, etc).
+A hostname is blocked if any of the addresses it resolves to is non-public, and redirects are checked the same way.
+Requests to a configured AI endpoint connect directly, without using the `HTTP_PROXY` or `HTTPS_PROXY` environment variables.

    Defaults to true, which allows internal endpoints.

-#### [`PAPERLESS_AI_LLM_EXTRA_PARAMS=<json>`](#PAPERLESS_AI_LLM_EXTRA_PARAMS) {#PAPERLESS_AI_LLM_EXTRA_PARAMS}
-
-: A JSON object of extra parameters sent with every LLM request, for providers that require a parameter Paperless does not
-set itself. Values here override Paperless' own, and no validation is performed. Whatever you put here is passed to the
-backend as-is, so an invalid parameter will simply be rejected by your provider. For example, current OpenAI reasoning
-models refuse tool calls on the chat completions API unless reasoning is off:
-
-    ```
-    PAPERLESS_AI_LLM_EXTRA_PARAMS={"reasoning_effort": "none"}
-    ```
-
-    Defaults to empty, which adds nothing to requests.
-
 #### [`PAPERLESS_LLM_INDEX_TASK_CRON=<cron expression>`](#PAPERLESS_LLM_INDEX_TASK_CRON) {#PAPERLESS_LLM_INDEX_TASK_CRON}

 : Configures the schedule to update the AI embeddings of text content and metadata for all documents. Only performed if
@@ -613,7 +613,7 @@ The following workflow action types are available:
 - The request headers as key-value pairs

 For security reasons, webhooks can be limited to specific ports and disallowed from connecting to local URLs. See the relevant
-[configuration settings](configuration.md#workflow-webhooks) to change this behavior. If you are allowing non-admins to create workflows,
+[configuration settings](configuration.md#workflow-webhooks) to change this behavior. Webhook requests connect directly (proxy environment variables are not used) and do not follow redirects. If you are allowing non-admins to create workflows,
 you may want to adjust these settings to prevent abuse.

 ##### Move to Trash {#workflow-action-move-to-trash}
@@ -17,8 +17,10 @@ classifiers = [
 # TODO: Move certain things to groups and then utilize that further
 # This will allow testing to not install a webserver, mysql, etc
 dependencies = [
+  "anyio>=4.12",
  "azure-ai-documentintelligence>=1.0.2",
  "babel>=2.17",
+  "bleach~=6.4.0",
  "celery[redis]~=5.6.2",
  "channels~=4.2",
  "channels-redis~=4.2",
@@ -46,6 +48,8 @@ dependencies = [
  "filelock~=3.32.0",
  "flower>=2.0.1,<2.2",
  "gotenberg-client[httpx]~=1.0",
+  "httpcore~=1.0.9",
+  "httpx~=0.28.1",
  "httpx-oauth~=0.17",
  "ijson>=3.5.1",
  "imap-tools>=1.14,<1.16",
@@ -76,7 +80,6 @@ dependencies = [
  "tantivy~=0.26.0",
  "tika-client[httpx]~=1.0",
  "torch>=2.13,<2.15",
-  "turbohtml~=1.10.0",
  "watchfiles>=1.2",
  "whitenoise~=6.11",
  "whoosh-compat[tantivy]==0.3",
@@ -136,6 +139,7 @@ typing = [
  "mypy",
  "mypy-baseline",
  "pyrefly",
+  "types-bleach",
  "types-channels",
  "types-colorama",
  "types-dateparser",
@@ -9745,7 +9745,7 @@
        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/document-list/document-list.component.ts</context>
-          <context context-type="linenumber">351</context>
+          <context context-type="linenumber">348</context>
        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/manage/document-attributes/document-attributes.component.html</context>
@@ -9760,7 +9760,7 @@
        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/document-list/document-list.component.ts</context>
-          <context context-type="linenumber">344</context>
+          <context context-type="linenumber">341</context>
        </context-group>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/manage/document-attributes/document-attributes.component.html</context>
@@ -10016,56 +10016,56 @@
        <source>Reset filters / selection</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/document-list/document-list.component.ts</context>
-          <context context-type="linenumber">332</context>
+          <context context-type="linenumber">329</context>
        </context-group>
      </trans-unit>
      <trans-unit id="4135055128446167640" datatype="html">
        <source>Open first [selected] document</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/document-list/document-list.component.ts</context>
-          <context context-type="linenumber">360</context>
+          <context context-type="linenumber">357</context>
        </context-group>
      </trans-unit>
      <trans-unit id="3629960544875360046" datatype="html">
        <source>Previous page</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/document-list/document-list.component.ts</context>
-          <context context-type="linenumber">376</context>
+          <context context-type="linenumber">373</context>
        </context-group>
      </trans-unit>
      <trans-unit id="3337301694210287595" datatype="html">
        <source>Next page</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/document-list/document-list.component.ts</context>
-          <context context-type="linenumber">388</context>
+          <context context-type="linenumber">385</context>
        </context-group>
      </trans-unit>
      <trans-unit id="2155249406916744630" datatype="html">
        <source>View &quot;<x id="PH" equiv-text="this.list.activeSavedViewTitle"/>&quot; saved successfully.</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/document-list/document-list.component.ts</context>
-          <context context-type="linenumber">422</context>
+          <context context-type="linenumber">419</context>
        </context-group>
      </trans-unit>
      <trans-unit id="4646273665293421938" datatype="html">
        <source>Failed to save view &quot;<x id="PH" equiv-text="this.list.activeSavedViewTitle"/>&quot;.</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/document-list/document-list.component.ts</context>
-          <context context-type="linenumber">428</context>
+          <context context-type="linenumber">425</context>
        </context-group>
      </trans-unit>
      <trans-unit id="6837554170707123455" datatype="html">
        <source>View &quot;<x id="PH" equiv-text="savedView.name"/>&quot; created successfully.</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/document-list/document-list.component.ts</context>
-          <context context-type="linenumber">497</context>
+          <context context-type="linenumber">494</context>
        </context-group>
      </trans-unit>
      <trans-unit id="6028096992841030074" datatype="html">
        <source>View &quot;<x id="PH" equiv-text="savedView.name"/>&quot; created successfully, but could not update visibility settings.</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/components/document-list/document-list.component.ts</context>
-          <context context-type="linenumber">503</context>
+          <context context-type="linenumber">500</context>
        </context-group>
      </trans-unit>
      <trans-unit id="739880801667335279" datatype="html">
@@ -12018,95 +12018,81 @@
          <context context-type="linenumber">351</context>
        </context-group>
      </trans-unit>
-      <trans-unit id="861068592166833023" datatype="html">
-        <source>LLM Embedding API Key</source>
-        <context-group purpose="location">
-          <context context-type="sourcefile">src/app/data/paperless-config.ts</context>
-          <context context-type="linenumber">358</context>
-        </context-group>
-      </trans-unit>
-      <trans-unit id="2929108042259892948" datatype="html">
-        <source>Used for embeddings when set, otherwise LLM API key is used.</source>
-        <context-group purpose="location">
-          <context context-type="sourcefile">src/app/data/paperless-config.ts</context>
-          <context context-type="linenumber">360</context>
-        </context-group>
-      </trans-unit>
      <trans-unit id="3554114880473286122" datatype="html">
        <source>LLM Embedding Endpoint</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/data/paperless-config.ts</context>
-          <context context-type="linenumber">366</context>
+          <context context-type="linenumber">358</context>
        </context-group>
      </trans-unit>
      <trans-unit id="1044242175651289991" datatype="html">
        <source>LLM Embedding Chunk Size</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/data/paperless-config.ts</context>
-          <context context-type="linenumber">373</context>
+          <context context-type="linenumber">365</context>
        </context-group>
      </trans-unit>
      <trans-unit id="7218245223139363113" datatype="html">
        <source>LLM Context Size</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/data/paperless-config.ts</context>
-          <context context-type="linenumber">380</context>
+          <context context-type="linenumber">372</context>
        </context-group>
      </trans-unit>
      <trans-unit id="4234495692726214397" datatype="html">
        <source>LLM Backend</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/data/paperless-config.ts</context>
-          <context context-type="linenumber">387</context>
+          <context context-type="linenumber">379</context>
        </context-group>
      </trans-unit>
      <trans-unit id="7935234833834000002" datatype="html">
        <source>LLM Model</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/data/paperless-config.ts</context>
-          <context context-type="linenumber">395</context>
+          <context context-type="linenumber">387</context>
        </context-group>
      </trans-unit>
      <trans-unit id="1980550530387803165" datatype="html">
        <source>LLM API Key</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/data/paperless-config.ts</context>
-          <context context-type="linenumber">402</context>
+          <context context-type="linenumber">394</context>
        </context-group>
      </trans-unit>
      <trans-unit id="6126617860376156501" datatype="html">
        <source>LLM Endpoint</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/data/paperless-config.ts</context>
-          <context context-type="linenumber">409</context>
+          <context context-type="linenumber">401</context>
        </context-group>
      </trans-unit>
      <trans-unit id="6572826277249350975" datatype="html">
        <source>LLM Output Language</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/data/paperless-config.ts</context>
-          <context context-type="linenumber">416</context>
+          <context context-type="linenumber">408</context>
        </context-group>
      </trans-unit>
      <trans-unit id="3284403507172415792" datatype="html">
        <source>Language to use for generated AI suggestions. When unset, AI suggestions use the user&apos;s display language if explicitly set.</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/data/paperless-config.ts</context>
-          <context context-type="linenumber">420</context>
+          <context context-type="linenumber">412</context>
        </context-group>
      </trans-unit>
      <trans-unit id="4493921125434706859" datatype="html">
        <source>LLM Request Timeout</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/data/paperless-config.ts</context>
-          <context context-type="linenumber">424</context>
+          <context context-type="linenumber">416</context>
        </context-group>
      </trans-unit>
      <trans-unit id="483994032066441287" datatype="html">
        <source>Timeout in seconds for LLM requests.</source>
        <context-group purpose="location">
          <context context-type="sourcefile">src/app/data/paperless-config.ts</context>
-          <context context-type="linenumber">428</context>
+          <context context-type="linenumber">420</context>
        </context-group>
      </trans-unit>
      <trans-unit id="1055686627716339120" datatype="html">
@@ -154,29 +154,11 @@
  & section {
    position: absolute;
    text-align: initial;
-    pointer-events: auto;
    box-sizing: border-box;
    transform-origin: 0 0;
  }

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

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

-  it('opens external links in a new tab', () => {
-    const linkService = (component as any).linkService
-    expect(linkService.options).toEqual(
-      expect.objectContaining({
-        externalLinkTarget: LinkTarget.BLANK,
-        externalLinkRel: 'noopener noreferrer nofollow',
-      })
-    )
-  })
-
  it('resolves the worker source relative to the document base URI', async () => {
    setBaseHref('/paperless/')
    const getDocumentSpy = jest.spyOn(pdfjs, 'getDocument')
@@ -21,7 +21,6 @@ import {
 } from 'pdfjs-dist/legacy/build/pdf.mjs'
 import {
  EventBus,
-  LinkTarget,
  PDFFindController,
  PDFLinkService,
  PDFSinglePageViewer,
@@ -76,11 +75,7 @@ export class PngxPdfViewerComponent
  private lastViewerPage?: number

  private readonly eventBus = new EventBus()
-  private readonly linkService = new PDFLinkService({
-    eventBus: this.eventBus,
-    externalLinkTarget: LinkTarget.BLANK,
-    externalLinkRel: 'noopener noreferrer nofollow',
-  })
+  private readonly linkService = new PDFLinkService({ eventBus: this.eventBus })
  private readonly findController = new PDFFindController({
    eventBus: this.eventBus,
    linkService: this.linkService,
@@ -146,19 +146,6 @@ describe('DocumentListComponent', () => {
    expect(reloadSpy).toHaveBeenCalled()
  })

-  it('should stop reloading on document deleted after destroy', () => {
-    const reloadSpy = jest.spyOn(documentListService, 'reload')
-    const documentDeletedSubject = new Subject<boolean>()
-    jest
-      .spyOn(websocketStatusService, 'onDocumentDeleted')
-      .mockReturnValue(documentDeletedSubject)
-    fixture.detectChanges()
-    fixture.destroy()
-    reloadSpy.mockClear()
-    documentDeletedSubject.next(true)
-    expect(reloadSpy).not.toHaveBeenCalled()
-  })
-
  it('should show score sort fields on fulltext queries', () => {
    documentListService.setFilterRules([
      {
@@ -270,12 +270,9 @@ export class DocumentListComponent
        this.list.reload()
      })

-    this.websocketStatusService
-      .onDocumentDeleted()
-      .pipe(takeUntil(this.unsubscribeNotifier))
-      .subscribe(() => {
-        this.list.reload()
-      })
+    this.websocketStatusService.onDocumentDeleted().subscribe(() => {
+      this.list.reload()
+    })

    this.route.paramMap
      .pipe(
@@ -353,14 +353,6 @@ export const PaperlessConfigOptions: ConfigOption[] = [
    config_key: 'PAPERLESS_AI_LLM_EMBEDDING_MODEL',
    category: ConfigCategory.AI,
  },
-  {
-    key: 'llm_embedding_api_key',
-    title: $localize`LLM Embedding API Key`,
-    type: ConfigOptionType.Password,
-    note: $localize`Used for embeddings when set, otherwise LLM API key is used.`,
-    config_key: 'PAPERLESS_AI_LLM_EMBEDDING_API_KEY',
-    category: ConfigCategory.AI,
-  },
  {
    key: 'llm_embedding_endpoint',
    title: $localize`LLM Embedding Endpoint`,
@@ -465,7 +457,6 @@ export interface PaperlessConfig extends ObjectWithId {
  ai_enabled: boolean
  llm_embedding_backend: string
  llm_embedding_model: string
-  llm_embedding_api_key: string
  llm_embedding_endpoint: string
  llm_embedding_chunk_size: number
  llm_context_size: number
@@ -25,20 +25,10 @@ export class PDFFindController {
  onIsPageVisible?: () => boolean
 }

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

-  constructor(readonly options: Record<string, unknown> = {}) {}
-
  setDocument(document: unknown): void {
    this.document = document
  }
@@ -17,10 +17,14 @@ if TYPE_CHECKING:

    from django.contrib.auth.models import User
    from pytest_django.fixtures import Settings
+    from pytest_mock import MockerFixture
    from rest_framework.test import APIClient

    from paperless_testing.dirs import PaperlessDirs
    from paperless_testing.fakes.progress import FakeProgressManager
+    from paperless_testing.outbound import DialRecorder
+    from paperless_testing.outbound import FakeDNS
+    from paperless_testing.outbound import LocalHTTPServer


@pytest.fixture(scope="session", autouse=True)
@@ -149,3 +153,41 @@ def fake_progress_manager(

    monkeypatch.setattr("documents.tasks.ProgressManager", FakeProgressManager)
    return FakeProgressManager
+
+
+@pytest.fixture
+def local_http_server() -> Generator[LocalHTTPServer, None, None]:
+    """A recording HTTP server on 127.0.0.1, for outbound connection tests."""
+    from paperless_testing.outbound import running_http_server
+
+    with running_http_server() as server:
+        yield server
+
+
+@pytest.fixture
+def fake_dns(mocker: MockerFixture) -> FakeDNS:
+    """Per-hostname answers for the outbound guard's resolver hooks."""
+    from paperless_testing.outbound import install_fake_dns
+
+    return install_fake_dns(mocker)
+
+
+@pytest.fixture
+def dial_recorder(mocker: MockerFixture) -> DialRecorder:
+    """Records which addresses the outbound guard actually dialled."""
+    from paperless_testing.outbound import install_dial_recorder
+
+    return install_dial_recorder(mocker)
+
+
+@pytest.fixture
+def every_address_is_public(mocker: MockerFixture) -> None:
+    """Disable the outbound guard's address policy: every address passes.
+
+    For tests that are not themselves exercising which addresses the guard
+    accepts, so loopback and other private addresses dial just like a
+    public one.
+    """
+    from paperless_testing.outbound import allow_all_addresses
+
+    allow_all_addresses(mocker)
@@ -26,6 +26,7 @@ class DocumentsConfig(AppConfig):
        document_consumption_finished.connect(set_document_type)
        document_consumption_finished.connect(set_tags)
        document_consumption_finished.connect(set_storage_path)
+        document_consumption_finished.connect(add_to_index)
        document_consumption_finished.connect(run_workflows_added)
        document_consumption_finished.connect(add_to_index)
        document_consumption_finished.connect(add_or_update_document_in_llm_index)
@@ -857,9 +857,8 @@ class ConsumerPlugin(
            self.log.debug(f"Creation date from parse_date: {create_date}")
        else:
            stats = Path(self.input_doc.original_file).stat()
-            create_date = datetime.datetime.fromtimestamp(
-                stats.st_mtime,
-                tz=timezone.get_current_timezone(),
+            create_date = timezone.make_aware(
+                datetime.datetime.fromtimestamp(stats.st_mtime),
            )
            self.log.debug(f"Creation date from st_mtime: {create_date}")

@@ -56,7 +56,6 @@ from documents.permissions import get_objects_for_user_owner_aware
 from documents.plugins.helpers import DocumentsStatusManager
 from documents.templating.utils import convert_format_str_to_template_format
 from documents.utils import compute_checksum
-from documents.utils import copy_file_with_basic_stats
 from documents.workflows.actions import build_workflow_action_context
 from documents.workflows.actions import execute_email_action
 from documents.workflows.actions import execute_move_to_trash_action
@@ -364,11 +363,7 @@ def cleanup_document_deletion(sender, instance, **kwargs) -> None:

            logger.debug(f"Moving {instance.source_path} to trash at {new_file_path}")
            try:
-                shutil.move(
-                    instance.source_path,
-                    new_file_path,
-                    copy_function=copy_file_with_basic_stats,
-                )
+                shutil.move(instance.source_path, new_file_path)
            except OSError as e:
                logger.error(
                    f"Failed to move {instance.source_path} to trash at "
@@ -81,7 +81,6 @@ class TestApiAppConfig(DirectoriesMixin, APITestCase):
                "ai_enabled": None,
                "llm_embedding_backend": None,
                "llm_embedding_model": None,
-                "llm_embedding_api_key": None,
                "llm_embedding_endpoint": None,
                "llm_embedding_chunk_size": None,
                "llm_context_size": None,
@@ -923,49 +922,6 @@ class TestApiAppConfig(DirectoriesMixin, APITestCase):
        self.assertEqual(response.status_code, status.HTTP_405_METHOD_NOT_ALLOWED)
        self.assertEqual(ApplicationConfiguration.objects.count(), 1)

-    def test_update_llm_embedding_api_key(self) -> None:
-        """
-        GIVEN:
-            - Existing config with llm_embedding_api_key specified
-        WHEN:
-            - API to update llm_embedding_api_key is called with all *s
-            - API to update llm_embedding_api_key is called with empty string
-        THEN:
-            - llm_embedding_api_key is unchanged
-            - llm_embedding_api_key is set to None
-        """
-        config = ApplicationConfiguration.objects.first()
-        assert config is not None
-        config.llm_embedding_api_key = "1234567890"
-        config.save()
-
-        # Test with all *
-        response = self.client.patch(
-            f"{self.ENDPOINT}1/",
-            json.dumps(
-                {
-                    "llm_embedding_api_key": "*" * 32,
-                },
-            ),
-            content_type="application/json",
-        )
-        self.assertEqual(response.status_code, status.HTTP_200_OK)
-        config.refresh_from_db()
-        self.assertEqual(config.llm_embedding_api_key, "1234567890")
-        # Test with empty string
-        response = self.client.patch(
-            f"{self.ENDPOINT}1/",
-            json.dumps(
-                {
-                    "llm_embedding_api_key": "",
-                },
-            ),
-            content_type="application/json",
-        )
-        self.assertEqual(response.status_code, status.HTTP_200_OK)
-        config.refresh_from_db()
-        self.assertEqual(config.llm_embedding_api_key, None)
-
    def test_update_llm_api_key(self) -> None:
        """
        GIVEN:
@@ -166,15 +166,7 @@ class TestBulkDownload(DirectoriesMixin, SampleDirMixin, APITestCase):
            ),
            content_type="application/json",
        )
-
-        self.assertEqual(response.status_code, status.HTTP_200_OK)
-        self.assertEqual(response["Content-Type"], "application/zip")
-
-        with zipfile.ZipFile(io.BytesIO(read_streaming_response(response))) as zipf:
-            self.assertEqual(zipf.infolist()[0].compress_type, zipfile.ZIP_LZMA)
-
-            with self.doc2.source_file as f:
-                self.assertEqual(f.read(), zipf.read("2021-01-01 document A.pdf"))
+        response.close()

    @override_settings(FILENAME_FORMAT="{correspondent}/{title}")
    def test_formatted_download_originals(self) -> None:
@@ -9,7 +9,6 @@ from rest_framework.test import APITestCase

 from documents.models import Correspondent
 from documents.models import CustomField
-from documents.models import CustomFieldInstance
 from documents.models import Document
 from documents.models import DocumentType
 from documents.models import StoragePath
@@ -2526,7 +2525,7 @@ class TestBulkEditAPI(DirectoriesMixin, APITestCase):
        WHEN:
            - API to bulk edit documents is called
        THEN:
-            - Audit log is created with the old and new correspondent
+            - Audit log is created
        """
        LogEntry.objects.all().delete()
        response = self.client.post(
@@ -2542,8 +2541,7 @@ class TestBulkEditAPI(DirectoriesMixin, APITestCase):
        )

        self.assertEqual(response.status_code, status.HTTP_200_OK)
-        entry = LogEntry.objects.get_for_object(self.doc1).get()
-        self.assertEqual(entry.changes, {"correspondent": [None, self.c2.id]})
+        self.assertEqual(LogEntry.objects.filter(object_pk=self.doc1.id).count(), 1)

    @override_settings(AUDIT_LOG_ENABLED=True)
    def test_bulk_edit_audit_log_enabled_tags(self) -> None:
@@ -2551,18 +2549,16 @@ class TestBulkEditAPI(DirectoriesMixin, APITestCase):
        GIVEN:
            - Audit log is enabled
        WHEN:
-            - API to bulk edit tags is called on an untagged document and a
-              document with several tags
+            - API to bulk edit tags is called
        THEN:
-            - Audit log is created for each document with its full tag list
-              before and after the edit
+            - Audit log is created
        """
        LogEntry.objects.all().delete()
        response = self.client.post(
            "/api/documents/bulk_edit/",
            json.dumps(
                {
-                    "documents": [self.doc1.id, self.doc4.id],
+                    "documents": [self.doc1.id],
                    "method": "modify_tags",
                    "parameters": {
                        "add_tags": [self.t1.id],
@@ -2574,32 +2570,18 @@ class TestBulkEditAPI(DirectoriesMixin, APITestCase):
        )

        self.assertEqual(response.status_code, status.HTTP_200_OK)
-        entry = LogEntry.objects.get_for_object(self.doc1).get()
-        self.assertEqual(entry.changes, {"tags": [[], [self.t1.id]]})
-        entry = LogEntry.objects.get_for_object(self.doc4).get()
-        self.assertEqual(
-            entry.changes,
-            {"tags": [[self.t1.id, self.t2.id], [self.t1.id]]},
-        )
+        self.assertEqual(LogEntry.objects.filter(object_pk=self.doc1.id).count(), 1)

    @override_settings(AUDIT_LOG_ENABLED=True)
    def test_bulk_edit_audit_log_enabled_custom_fields(self) -> None:
        """
        GIVEN:
            - Audit log is enabled
-            - A document with two custom fields
        WHEN:
-            - API to bulk edit custom fields is called to add a third
+            - API to bulk edit custom fields is called
        THEN:
-            - Audit log is created with every custom field instance before and
-              after the edit
-            - Audit log is created for the new custom field instance
+            - Audit log is created
        """
-        cf3 = CustomField.objects.create(name="cf3", data_type="string")
-        existing = [
-            CustomFieldInstance.objects.create(document=self.doc1, field=field)
-            for field in (self.cf2, cf3)
-        ]
        LogEntry.objects.all().delete()
        response = self.client.post(
            "/api/documents/bulk_edit/",
@@ -2617,14 +2599,7 @@ class TestBulkEditAPI(DirectoriesMixin, APITestCase):
        )

        self.assertEqual(response.status_code, status.HTTP_200_OK)
-        added = CustomFieldInstance.objects.get(document=self.doc1, field=self.cf1)
-        existing_ids = [instance.id for instance in existing]
-        entry = LogEntry.objects.get_for_object(self.doc1).get()
-        self.assertEqual(
-            entry.changes,
-            {"custom_fields": [existing_ids, [*existing_ids, added.id]]},
-        )
-        self.assertEqual(LogEntry.objects.get_for_object(added).count(), 1)
+        self.assertEqual(LogEntry.objects.filter(object_pk=self.doc1.id).count(), 2)

    def test_api_bulk_edit_with_bad_search_query_returns_400(self) -> None:
        """
@@ -64,15 +64,16 @@ class TestApiProfile(DirectoriesMixin, APITestCase):
        )
        self.client.force_authenticate(user=self.user)

-    def setupSocialAccount(self) -> SocialAccount:
+    def setupSocialAccount(self) -> None:
        SocialApp.objects.create(
            name="Keycloak",
            provider="openid_connect",
            provider_id="keycloak-test",
        )
-        social_account = SocialAccount(uid="123456789", provider="keycloak-test")
-        self.user.socialaccount_set.add(social_account, bulk=False)
-        return social_account
+        self.user.socialaccount_set.add(
+            SocialAccount(uid="123456789", provider="keycloak-test"),
+            bulk=False,
+        )

    def test_get_profile(self) -> None:
        """
@@ -110,17 +111,19 @@ class TestApiProfile(DirectoriesMixin, APITestCase):
        THEN:
            - Profile is returned with social accounts
        """
-        social_account = self.setupSocialAccount()
+        self.setupSocialAccount()

-        openid_provider = MockOpenIDConnectProvider(
-            app=SocialApp.objects.get(provider_id="keycloak-test"),
+        openid_provider = (
+            MockOpenIDConnectProvider(
+                app=SocialApp.objects.get(provider_id="keycloak-test"),
+            ),
        )
        mock_list_providers.return_value = [
            openid_provider,
        ]
        mock_get_provider_account.return_value = MockOpenIDConnectProviderAccount(
            mock_social_account_dict={
-                "name": openid_provider.name,
+                "name": openid_provider[0].name,
            },
        )

@@ -132,7 +135,7 @@ class TestApiProfile(DirectoriesMixin, APITestCase):
            response.data["social_accounts"],
            [
                {
-                    "id": social_account.pk,
+                    "id": 1,
                    "provider": "keycloak-test",
                    "name": "Keycloak",
                },
@@ -149,7 +152,7 @@ class TestApiProfile(DirectoriesMixin, APITestCase):
        THEN:
            - Profile is returned with "Unknown App" as name
        """
-        social_account = self.setupSocialAccount()
+        self.setupSocialAccount()

        # Remove the social app
        SocialApp.objects.get(provider_id="keycloak-test").delete()
@@ -162,7 +165,7 @@ class TestApiProfile(DirectoriesMixin, APITestCase):
            response.data["social_accounts"],
            [
                {
-                    "id": social_account.pk,
+                    "id": 1,
                    "provider": "keycloak-test",
                    "name": "Unknown App",
                },
@@ -677,13 +677,12 @@ class TestExportImport(
        THEN:
            - Error is raised
        """
-        with tempfile.TemporaryDirectory() as tmp_dir:
-            args = ["document_exporter", str(Path(tmp_dir) / "does-not-exist")]
+        args = ["document_exporter", "/tmp/foo/bar"]

-            with self.assertRaises(CommandError) as e:
-                call_command(*args, skip_checks=True)
+        with self.assertRaises(CommandError) as e:
+            call_command(*args, skip_checks=True)

-            self.assertEqual("That path doesn't exist", str(e.exception))
+        self.assertEqual("That path doesn't exist", str(e.exception))

    def test_export_target_exists_but_is_file(self) -> None:
        """
@@ -123,14 +123,14 @@ class TestFuzzyMatchCommand(TestCase):
            - Output contains clickable links to the documents instead of titles
        """
        # Content similarity is 86.667
-        doc1 = Document.objects.create(
+        Document.objects.create(
            checksum="BEEFCAFE",
            title="A",
            content="first document scanned by bob",
            mime_type="application/pdf",
            filename="test.pdf",
        )
-        doc2 = Document.objects.create(
+        Document.objects.create(
            checksum="DEADBEAF",
            title="A",
            content="first document scanned by alice",
@@ -145,8 +145,8 @@ class TestFuzzyMatchCommand(TestCase):
                "http://localhost:8000",
            )
        self.assertIn("Found 1 matching pair(s)", stdout)
-        self.assertIn(f"http://localhost:8000/documents/{doc1.pk}/details", stdout)
-        self.assertIn(f"http://localhost:8000/documents/{doc2.pk}/details", stdout)
+        self.assertIn("http://localhost:8000/documents/1/details", stdout)
+        self.assertIn("http://localhost:8000/documents/2/details", stdout)

    def test_with_3_matches(self) -> None:
        """
@@ -198,14 +198,14 @@ class TestFuzzyMatchCommand(TestCase):
            - Documents 1 and 2 remain
        """
        # Content similarity is 86.667
-        doc1 = Document.objects.create(
+        Document.objects.create(
            checksum="BEEFCAFE",
            title="A",
            content="first document scanned by bob",
            mime_type="application/pdf",
            filename="test.pdf",
        )
-        doc2 = Document.objects.create(
+        Document.objects.create(
            checksum="DEADBEAF",
            title="A",
            content="second document scanned by alice",
@@ -235,8 +235,8 @@ class TestFuzzyMatchCommand(TestCase):
        self.assertIn("Deleting 1 document(s)", stdout)

        self.assertEqual(Document.objects.count(), 2)
-        self.assertIsNotNone(Document.objects.get(pk=doc1.pk))
-        self.assertIsNotNone(Document.objects.get(pk=doc2.pk))
+        self.assertIsNotNone(Document.objects.get(pk=1))
+        self.assertIsNotNone(Document.objects.get(pk=2))

    def test_document_deletion_cancelled(self) -> None:
        """
@@ -14,7 +14,7 @@ from paperless_testing.dirs import DirectoriesMixin
 class TestManageSuperUser(DirectoriesMixin, TestCase):
    def call_command(self, environ):
        out = StringIO()
-        with mock.patch.dict(os.environ, environ, clear=True):
+        with mock.patch.dict(os.environ, environ):
            call_command(
                "manage_superuser",
                "--no-color",
@@ -430,53 +430,6 @@ class TestBulkDownloadPermissionChecksRootDocument:
        )  # version-only grant must not substitute for root permission


-@pytest.mark.django_db
-class TestDocumentOperationPermissionChecksRootDocument:
-    @pytest.mark.parametrize(
-        ("endpoint", "payload"),
-        [
-            pytest.param("/api/documents/merge/", {}, id="merge"),
-            pytest.param("/api/documents/rotate/", {"degrees": 90}, id="rotate"),
-        ],
-    )
-    @pytest.mark.parametrize("version_owner", ["none", "requester"])
-    def test_version_operation_acts_on_root(
-        self,
-        rest_api_client: APIClient,
-        endpoint: str,
-        payload: dict,
-        version_owner: str,
-    ) -> None:
-        owner = UserFactory(username="owner")
-        requester = UserFactory(username="requester")
-        grant_global(requester, "change_document")
-        grant_global(requester, "add_document")
-        rest_api_client.force_authenticate(user=requester)
-        root = DocumentFactory(owner=owner)
-        # A version whose owner went stale, e.g. created before the root changed hands
-        version = DocumentFactory(
-            owner=requester if version_owner == "requester" else None,
-            root_document=root,
-            version_index=1,
-        )
-
-        with (
-            patch("documents.views.bulk_edit.merge") as mock_merge,
-            patch("documents.views.bulk_edit.rotate") as mock_rotate,
-        ):
-            mock_merge.__name__ = "merge"
-            mock_rotate.__name__ = "rotate"
-            response = rest_api_client.post(
-                endpoint,
-                {"documents": [version.pk], **payload},
-                format="json",
-            )
-
-        assert response.status_code == HTTPStatus.FORBIDDEN
-        mock_merge.assert_not_called()
-        mock_rotate.assert_not_called()
-
-
@pytest.mark.django_db
@pytest.mark.usefixtures("_search_index")
 class TestTrashRestorePermissionBoundary:
@@ -339,6 +339,15 @@ class ShareLinkBundleBuildTaskTests(DirectoriesMixin, APITestCase):
        )
        self.document.archive_checksum = ""
        self.document.save()
+        self.addCleanup(
+            setattr,
+            settings,
+            "SHARE_LINK_BUNDLE_DIR",
+            settings.SHARE_LINK_BUNDLE_DIR,
+        )
+        settings.SHARE_LINK_BUNDLE_DIR = (
+            Path(settings.MEDIA_ROOT) / "documents" / "share_link_bundles"
+        )

    def _write_document_file(self, *, archive: bool, content: bytes) -> Path:
        if archive:
@@ -18,6 +18,7 @@ from documents.models import WorkflowAction
 from documents.sanity_checker import SanityCheckFailedException
 from documents.sanity_checker import SanityCheckMessages
 from documents.tests.helpers import dummy_preprocess
+from paperless_ai.exceptions import LLMBlockedError
 from paperless_testing.assertions import FileSystemAssertsMixin
 from paperless_testing.dirs import DirectoriesMixin

@@ -555,3 +556,37 @@ class TestApplyAISuggestionsTask(DirectoriesMixin, TestCase):

        apply_suggestions.assert_not_called()
        self.assertIn("no longer exists", "".join(cm.output))
+
+    @override_settings(AI_ENABLED=True)
+    def test_blocked_request_fails_without_retry(self) -> None:
+        """
+        GIVEN:
+            - AI enabled and a document with content
+            - The AI classification call blocked by the outbound request policy
+        WHEN:
+            - The task runs through Celery
+        THEN:
+            - The workflow code does not swallow the block
+            - The task fails with LLMBlockedError and is never retried
+        """
+        with (
+            mock.patch(
+                "documents.workflows.ai.get_ai_document_classification",
+                side_effect=LLMBlockedError(
+                    "AI backend request was blocked by the outbound request "
+                    "policy: detail",
+                ),
+            ),
+            mock.patch.object(
+                tasks.apply_ai_suggestions,
+                "retry",
+                wraps=tasks.apply_ai_suggestions.retry,
+            ) as retry,
+        ):
+            result = tasks.apply_ai_suggestions.apply(
+                args=(self.action.pk, self.doc.pk),
+            )
+
+        self.assertTrue(result.failed())
+        self.assertIsInstance(result.result, LLMBlockedError)
+        retry.assert_not_called()
@@ -29,6 +29,7 @@ from documents.models import Tag
 from documents.models import UiSettings
 from documents.signals.handlers import update_llm_suggestions_cache
 from paperless.models import ApplicationConfiguration
+from paperless_ai.exceptions import LLMBlockedError
 from paperless_ai.exceptions import LLMProviderError
 from paperless_ai.exceptions import LLMTimeoutError
 from paperless_testing.dirs import DirectoriesMixin
@@ -770,6 +771,48 @@ class TestAISuggestions(DirectoriesMixin, TestCase):
            get_llm_suggestion_cache(self.document.pk, backend="openai-like"),
        )

+    @patch("documents.views.get_ai_document_classification")
+    @override_settings(
+        AI_ENABLED=True,
+        LLM_BACKEND="openai-like",
+    )
+    def test_ai_suggestions_with_blocked_llm_request(
+        self,
+        mock_get_ai_classification,
+    ) -> None:
+        """
+        GIVEN:
+            - An AI backend request blocked by the outbound request policy
+        WHEN:
+            - AI suggestions are requested
+        THEN:
+            - 502 is returned with a generic message and nothing is cached
+        """
+        mock_get_ai_classification.side_effect = LLMBlockedError(
+            "AI backend request was blocked by the outbound request policy: detail",
+        )
+
+        self.client.force_login(user=self.user)
+        response = self.client.get(
+            f"/api/documents/{self.document.pk}/ai_suggestions/",
+        )
+
+        self.assertEqual(response.status_code, status.HTTP_502_BAD_GATEWAY)
+        self.assertEqual(
+            response.json(),
+            {
+                "ai": [
+                    (
+                        "AI backend request was blocked by the outbound request "
+                        "policy. Check logs for details."
+                    ),
+                ],
+            },
+        )
+        self.assertIsNone(
+            get_llm_suggestion_cache(self.document.pk, backend="openai-like"),
+        )
+
    @patch("documents.views.get_ai_document_classification")
    @override_settings(
        AI_ENABLED=True,
@@ -1,9 +1,7 @@
 import datetime
 import json
 import shutil
-import socket
 import tempfile
-from collections.abc import Callable
 from datetime import timedelta
 from pathlib import Path
 from typing import TYPE_CHECKING
@@ -19,11 +17,11 @@ from django.test import override_settings
 from django.utils import timezone
 from guardian.shortcuts import get_groups_with_perms
 from guardian.shortcuts import get_users_with_perms
-from httpx import ConnectError
 from httpx import HTTPError
 from httpx import HTTPStatusError
 from pytest_django.fixtures import Settings
 from pytest_httpx import HTTPXMock
+from pytest_mock import MockerFixture
 from rest_framework.test import APIClient
 from rest_framework.test import APITestCase

@@ -33,8 +31,12 @@ 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.network import OutboundRequestBlockedError
 from paperless_ai.base_model import ClassificationSuggestions
 from paperless_ai.exceptions import LLMTimeoutError
+from paperless_testing.outbound import DialRecorder
+from paperless_testing.outbound import FakeDNS
+from paperless_testing.outbound import LocalHTTPServer

 if TYPE_CHECKING:
    from django.db.models import QuerySet
@@ -69,9 +71,7 @@ from paperless_mail.models import MailAccount
 from paperless_mail.models import MailRule
 from paperless_testing.assertions import FileSystemAssertsMixin
 from paperless_testing.dirs import DirectoriesMixin
-from paperless_testing.factories import DocumentFactory
 from paperless_testing.factories import UserFactory
-from paperless_testing.permissions import grant_global
 from paperless_testing.permissions import grant_object


@@ -1061,41 +1061,6 @@ class TestWorkflows(
        self.assertEqual(doc.correspondent, self.c2)
        self.assertEqual(doc.title, f"Doc created in {created.year}")

-    @pytest.mark.usefixtures("_search_index")
-    def test_document_added_workflow_indexes_final_title(self) -> None:
-        trigger = WorkflowTrigger.objects.create(
-            type=WorkflowTrigger.WorkflowTriggerType.DOCUMENT_ADDED,
-            filter_filename="*sample*",
-        )
-        action = WorkflowAction.objects.create(
-            assign_title="Linked document",
-            assign_owner=self.user2,
-        )
-        link_field = CustomField.objects.create(
-            name="Related documents",
-            data_type=CustomField.FieldDataType.DOCUMENTLINK,
-        )
-        action.assign_custom_fields.add(link_field)
-        workflow = Workflow.objects.create(name="Link workflow", order=0)
-        workflow.triggers.add(trigger)
-        workflow.actions.add(action)
-
-        doc = DocumentFactory.create()
-        document_consumption_finished.send(sender=self.__class__, document=doc)
-
-        self.assertTrue(doc.custom_fields.filter(field=link_field).exists())
-        doc.refresh_from_db()
-        self.assertEqual(doc.title, "Linked document")
-
-        grant_global(self.user2, "view_document")
-        self.client.force_authenticate(user=self.user2)
-        response = self.client.get("/api/documents/?title_search=linked")
-        self.assertEqual(response.status_code, 200)
-        self.assertEqual(
-            [result["id"] for result in response.data["results"]],
-            [doc.pk],
-        )
-
    def test_document_added_no_match_filename(self) -> None:
        trigger = WorkflowTrigger.objects.create(
            type=WorkflowTrigger.WorkflowTriggerType.DOCUMENT_ADDED,
@@ -4459,18 +4424,18 @@ class TestWorkflows(
        )

    @mock.patch("documents.bulk_edit.remove_password")
-    def test_password_removal_action_skips_blank_and_whitespace_passwords(
+    def test_password_removal_action_fails_without_correct_password(
        self,
        mock_remove_password,
    ) -> None:
        """
        GIVEN:
            - Workflow password removal action
-            - Only blank and whitespace-only passwords configured
+            - No correct password provided
        WHEN:
            - Document updated triggering the workflow
        THEN:
-            - Password removal is not attempted
+            - Password removal is attempted for all passwords and fails
        """
        doc = Document.objects.create(
            title="Protected",
@@ -4491,60 +4456,6 @@ class TestWorkflows(

        mock_remove_password.assert_not_called()

-    @mock.patch("documents.bulk_edit.remove_password")
-    def test_password_removal_action_fails_without_correct_password(
-        self,
-        mock_remove_password,
-    ) -> None:
-        """
-        GIVEN:
-            - Workflow password removal action
-            - No configured password is correct
-        WHEN:
-            - Document updated triggering the workflow
-        THEN:
-            - Password removal is attempted for every configured password and fails
-        """
-        doc = Document.objects.create(
-            title="Protected",
-            checksum="pw-checksum-3",
-        )
-        trigger = WorkflowTrigger.objects.create(
-            type=WorkflowTrigger.WorkflowTriggerType.DOCUMENT_UPDATED,
-        )
-        action = WorkflowAction.objects.create(
-            type=WorkflowAction.WorkflowActionType.PASSWORD_REMOVAL,
-            passwords=["wrong", "also-wrong"],
-        )
-        workflow = Workflow.objects.create(name="Password workflow wrong passwords")
-        workflow.triggers.add(trigger)
-        workflow.actions.add(action)
-
-        mock_remove_password.side_effect = ValueError("wrong password")
-
-        with self.assertLogs("paperless.workflows.actions", level="ERROR"):
-            run_workflows(trigger.type, doc)
-
-        assert mock_remove_password.call_count == 2
-        mock_remove_password.assert_has_calls(
-            [
-                mock.call(
-                    [doc.id],
-                    password="wrong",
-                    update_document=True,
-                    user=doc.owner,
-                    source_paths_by_id=None,
-                ),
-                mock.call(
-                    [doc.id],
-                    password="also-wrong",
-                    update_document=True,
-                    user=doc.owner,
-                    source_paths_by_id=None,
-                ),
-            ],
-        )
-
    @mock.patch("documents.bulk_edit.remove_password")
    def test_password_removal_action_skips_without_passwords(
        self,
@@ -5160,25 +5071,6 @@ class TestWebhookSend:
        assert httpx_mock.get_request().headers["Content-Type"] == "application/json"


-@pytest.fixture
-def resolve_to(monkeypatch: pytest.MonkeyPatch) -> Callable[[str], None]:
-    """
-    Force DNS resolution to a specific IP for any hostname.
-    """
-
-    def _set(ip: str) -> None:
-        def fake_getaddrinfo(
-            host: str,
-            *_args: object,
-            **_kwargs: object,
-        ) -> list[tuple[Any, ...]]:
-            return [(socket.AF_INET, None, None, "", (ip, 0))]
-
-        monkeypatch.setattr(socket, "getaddrinfo", fake_getaddrinfo)
-
-    return _set
-
-
 class TestWebhookSecurity:
    def test_blocks_invalid_scheme_or_hostname(self, httpx_mock: HTTPXMock) -> None:
        """
@@ -5228,60 +5120,145 @@ class TestWebhookSecurity:

        assert httpx_mock.get_request() is None

+    @pytest.mark.parametrize(
+        "address",
+        [
+            pytest.param("127.0.0.1", id="loopback"),
+            pytest.param("10.0.0.1", id="private"),
+            pytest.param("169.254.169.254", id="link-local-metadata"),
+            pytest.param("::ffff:127.0.0.1", id="ipv4-mapped-loopback"),
+            pytest.param("64:ff9b::7f00:1", id="nat64-wrapping-loopback"),
+        ],
+    )
    @override_settings(WEBHOOKS_ALLOW_INTERNAL_REQUESTS=False)
    def test_blocks_private_loopback_linklocal(
        self,
-        httpx_mock: HTTPXMock,
-        resolve_to,
+        local_http_server: LocalHTTPServer,
+        fake_dns: FakeDNS,
+        dial_recorder: DialRecorder,
+        address: str,
    ) -> None:
        """
        GIVEN:
-            - URL with a private, loopback, or link-local IP address
+            - A webhook host resolving to a non-public address
            - WEBHOOKS_ALLOW_INTERNAL_REQUESTS is False
        WHEN:
-            - send_webhook is called with such URL
+            - send_webhook is called
        THEN:
-            - ValueError is raised
+            - The request is blocked before any connection is opened
        """
-        resolve_to("127.0.0.1")
-        with pytest.raises(ConnectError):
+        fake_dns.add("webhook.test", address)
+
+        with pytest.raises(OutboundRequestBlockedError):
            send_webhook(
-                "http://paperless-ngx.com",
+                f"http://webhook.test:{local_http_server.port}",
                data="",
                headers={},
                files=None,
                as_json=False,
            )

-    def test_allows_public_ip_and_sends(
+        assert local_http_server.connections == 0
+        assert dial_recorder.hosts() == []
+
+    @override_settings(WEBHOOKS_ALLOW_INTERNAL_REQUESTS=False)
+    @pytest.mark.usefixtures("every_address_is_public")
+    def test_sends_to_validated_address(
        self,
-        httpx_mock: HTTPXMock,
-        resolve_to,
+        local_http_server: LocalHTTPServer,
+        fake_dns: FakeDNS,
    ) -> None:
        """
        GIVEN:
-            - URL with a public IP address
+            - A webhook host resolving to an address the policy accepts
+            - WEBHOOKS_ALLOW_INTERNAL_REQUESTS is False
        WHEN:
-            - send_webhook is called with such URL
+            - send_webhook is called
        THEN:
-            - Request is sent successfully
+            - The payload arrives with the webhook hostname in the Host header
        """
-        resolve_to("52.207.186.75")
-        httpx_mock.add_response(content=b"ok")
+        fake_dns.add("webhook.test", "127.0.0.1")

        send_webhook(
-            url="http://paperless-ngx.com",
+            url=f"http://webhook.test:{local_http_server.port}",
            data="hi",
            headers={},
            files=None,
            as_json=False,
        )

-        req = httpx_mock.get_request()
-        assert req.url.host == "52.207.186.75"
-        assert req.headers["host"] == "paperless-ngx.com"
+        received = local_http_server.requests[0]
+        assert received.body == b"hi"
+        assert received.headers["host"] == f"webhook.test:{local_http_server.port}"

-    def test_follow_redirects_disabled(self, httpx_mock: HTTPXMock, resolve_to) -> None:
+    @override_settings(WEBHOOKS_ALLOW_INTERNAL_REQUESTS=True)
+    def test_allow_internal_sends_to_internal_address(
+        self,
+        local_http_server: LocalHTTPServer,
+        fake_dns: FakeDNS,
+    ) -> None:
+        """
+        GIVEN:
+            - A webhook to localhost
+            - WEBHOOKS_ALLOW_INTERNAL_REQUESTS is True
+        WHEN:
+            - send_webhook is called
+        THEN:
+            - The payload arrives at the internal address
+            - The guard does not resolve the host, leaving it to the stock
+              connection path
+        """
+        send_webhook(
+            url=f"http://localhost:{local_http_server.port}",
+            data="hi",
+            headers={},
+            files=None,
+            as_json=False,
+        )
+
+        received = local_http_server.requests[0]
+        assert received.body == b"hi"
+        assert fake_dns.lookups == []
+
+    @override_settings(WEBHOOKS_ALLOW_INTERNAL_REQUESTS=False)
+    def test_block_is_an_expected_task_failure(
+        self,
+        mocker: MockerFixture,
+        local_http_server: LocalHTTPServer,
+        fake_dns: FakeDNS,
+        dial_recorder: DialRecorder,
+    ) -> None:
+        """
+        GIVEN:
+            - A webhook host resolving to a loopback address
+            - WEBHOOKS_ALLOW_INTERNAL_REQUESTS is False
+        WHEN:
+            - The webhook task runs through Celery
+        THEN:
+            - The task fails with the original block error, not a wrapper,
+              so it matches the task's expected errors, and is not retried
+        """
+        fake_dns.add("webhook.test", "127.0.0.1")
+        retry = mocker.spy(send_webhook, "retry")
+
+        result = send_webhook.apply(
+            kwargs={
+                "url": f"http://webhook.test:{local_http_server.port}",
+                "data": "",
+                "headers": {},
+                "files": None,
+                "as_json": False,
+            },
+        )
+
+        assert result.failed()
+        assert isinstance(result.result, OutboundRequestBlockedError)
+        assert isinstance(result.result, send_webhook.throws)
+        retry.assert_not_called()
+        assert local_http_server.connections == 0
+        assert dial_recorder.hosts() == []
+
+    def test_follow_redirects_disabled(self, httpx_mock: HTTPXMock) -> None:
        """
        GIVEN:
            - A URL that redirects
@@ -5290,7 +5267,6 @@ class TestWebhookSecurity:
        THEN:
            - Request is made to the original URL and does not follow the redirect
        """
-        resolve_to("52.207.186.75")
        # Return a redirect and ensure we don't follow it (only one request recorded)
        httpx_mock.add_response(
            status_code=302,
@@ -5312,7 +5288,6 @@ class TestWebhookSecurity:
    def test_strips_user_supplied_host_header(
        self,
        httpx_mock: HTTPXMock,
-        resolve_to: Callable[[str], None],
    ) -> None:
        """
        GIVEN:
@@ -5320,9 +5295,8 @@ class TestWebhookSecurity:
        WHEN:
            - send_webhook is called with a malicious Host header
        THEN:
-            - The Host header is stripped and replaced with the resolved hostname
+            - The Host header is stripped and set from the URL hostname
        """
-        resolve_to("52.207.186.75")
        httpx_mock.add_response(content=b"ok")

        send_webhook(
@@ -49,6 +49,7 @@ from django.db.models import Sum
 from django.db.models import When
 from django.db.models.functions import Coalesce
 from django.db.models.functions import Lower
+from django.db.models.manager import Manager
 from django.http import FileResponse
 from django.http import Http404
 from django.http import HttpRequest
@@ -255,6 +256,7 @@ 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 LLMBlockedError
 from paperless_ai.exceptions import LLMProviderError
 from paperless_ai.exceptions import LLMTimeoutError
 from paperless_ai.matching import extract_unmatched_names
@@ -1696,6 +1698,23 @@ class DocumentViewSet(
                    },
                    status=status.HTTP_502_BAD_GATEWAY,
                )
+            except LLMBlockedError as exc:
+                logger.warning(
+                    "AI backend request for document %s was blocked: %s",
+                    doc.pk,
+                    exc,
+                )
+                return Response(
+                    {
+                        "ai": [
+                            _(
+                                "AI backend request was blocked by the outbound "
+                                "request policy. Check logs for details.",
+                            ),
+                        ],
+                    },
+                    status=status.HTTP_502_BAD_GATEWAY,
+                )
            set_llm_suggestions_cache(
                doc.pk,
                llm_suggestions,
@@ -2967,15 +2986,11 @@ class DocumentOperationPermissionMixin(PassUserMixin, DocumentSelectionMixin):
        if user.is_superuser:
            return True

-        root_docs = {
-            get_root_document(doc)
-            for doc in Document.objects.select_related(
-                "owner",
-                "root_document__owner",
-            ).filter(pk__in=documents)
-        }
+        document_objs = Document.objects.select_related("owner").filter(
+            pk__in=documents,
+        )
        user_is_owner_of_all_documents = all(
-            (doc.owner == user or doc.owner is None) for doc in root_docs
+            (doc.owner == user or doc.owner is None) for doc in document_objs
        )

        # check global and object permissions for all documents
@@ -2983,13 +2998,9 @@ class DocumentOperationPermissionMixin(PassUserMixin, DocumentSelectionMixin):
            user.has_perm(
                "documents.change_document",
            )
-            and not Document.global_objects.filter(
-                pk__in=[doc.pk for doc in root_docs],
-            )
-            .exclude(
+            and not document_objs.exclude(
                pk__in=permitted_document_ids(user, perm="change_document"),
-            )
-            .exists()
+            ).exists()
        )

        # check ownership for methods that change original document
@@ -3148,38 +3159,6 @@ class BulkEditView(DocumentOperationPermissionMixin):

    serializer_class = BulkEditSerializer

-    @staticmethod
-    def _snapshot_field(doc_ids: list[int], field: str) -> dict[int, Any]:
-        """
-        Returns each document's current value of field, for the audit log.
-
-        Tags and custom fields are one row per value, so they are gathered
-        into a sorted list of pks per document (empty when there are none).
-        Reading them through Document.values() instead would join those rows
-        and return one arbitrary value per document.
-        """
-        if field == "tags":
-            rows = (
-                Document.tags.through.objects.filter(document_id__in=doc_ids)
-                .order_by("tag_id")
-                .values_list("document_id", "tag_id")
-            )
-        elif field == "custom_fields":
-            rows = (
-                CustomFieldInstance.objects.filter(document_id__in=doc_ids)
-                .order_by("pk")
-                .values_list("document_id", "pk")
-            )
-        else:
-            return dict(
-                Document.objects.filter(pk__in=doc_ids).values_list("pk", field),
-            )
-
-        values: dict[int, list[int]] = {doc_id: [] for doc_id in doc_ids}
-        for doc_id, pk in rows:
-            values[doc_id].append(pk)
-        return values
-
    def post(self, request, *args, **kwargs):
        request_method = request.data.get("method")
        api_version = int(request.version or settings.REST_FRAMEWORK["DEFAULT_VERSION"])
@@ -3226,19 +3205,41 @@ class BulkEditView(DocumentOperationPermissionMixin):
        try:
            modified_field = self.MODIFIED_FIELD_BY_METHOD.get(method.__name__, None)
            if settings.AUDIT_LOG_ENABLED and modified_field:
-                old_values = self._snapshot_field(documents, modified_field)
+                old_documents = {
+                    obj["pk"]: obj
+                    for obj in Document.objects.filter(pk__in=documents).values(
+                        "pk",
+                        "correspondent",
+                        "document_type",
+                        "storage_path",
+                        "tags",
+                        "custom_fields",
+                        "deleted_at",
+                        "checksum",
+                    )
+                }

            result = method(documents, **parameters)

            if settings.AUDIT_LOG_ENABLED and modified_field:
-                new_values = self._snapshot_field(documents, modified_field)
-                for doc in Document.objects.filter(pk__in=documents):
+                new_documents = Document.objects.filter(pk__in=documents)
+                for doc in new_documents:
+                    old_value = old_documents[doc.pk][modified_field]
+                    new_value = getattr(doc, modified_field)
+
+                    if isinstance(new_value, Model):
+                        # correspondent, document type, etc.
+                        new_value = new_value.pk
+                    elif isinstance(new_value, Manager):
+                        # tags, custom fields
+                        new_value = list(new_value.values_list("pk", flat=True))
+
                    LogEntry.objects.log_create(
                        instance=doc,
                        changes={
                            modified_field: [
-                                old_values[doc.pk],
-                                new_values[doc.pk],
+                                old_value,
+                                new_value,
                            ],
                        },
                        action=LogEntry.Action.UPDATE,
@@ -4,7 +4,8 @@ import httpx
 from celery import shared_task
 from django.conf import settings

-from paperless.network import PinnedHostHTTPTransport
+from paperless.network import GuardedHTTPTransport
+from paperless.network import OutboundRequestBlockedError
 from paperless.network import validate_outbound_http_url

 logger = logging.getLogger("paperless.workflows.webhooks")
@@ -14,7 +15,7 @@ logger = logging.getLogger("paperless.workflows.webhooks")
    retry_backoff=True,
    autoretry_for=(httpx.HTTPStatusError,),
    max_retries=3,
-    throws=(httpx.HTTPError,),
+    throws=(httpx.HTTPError, OutboundRequestBlockedError),
 )
 def send_webhook(
    url: str,
@@ -29,14 +30,15 @@ def send_webhook(
            url,
            allowed_schemes=settings.WEBHOOKS_ALLOWED_SCHEMES,
            allowed_ports=settings.WEBHOOKS_ALLOWED_PORTS,
-            # Internal-address checks happen in transport to preserve ConnectError behavior.
+            # Scheme and port only; the transport enforces the internal-address
+            # policy at connect time, on the address actually dialled.
            allow_internal=True,
        )
    except ValueError as e:
        logger.warning("Webhook blocked: %s", e)
        raise

-    transport = PinnedHostHTTPTransport(
+    transport = GuardedHTTPTransport(
        allow_internal=settings.WEBHOOKS_ALLOW_INTERNAL_REQUESTS,
    )

@@ -2,7 +2,7 @@ msgid ""
 msgstr ""
 "Project-Id-Version: paperless-ngx\n"
 "Report-Msgid-Bugs-To: \n"
-"POT-Creation-Date: 2026-09-27 14:07+0000\n"
+"POT-Creation-Date: 2026-09-21 19:00+0000\n"
 "PO-Revision-Date: 2022-02-17 04:17\n"
 "Last-Translator: \n"
 "Language-Team: English\n"
@@ -1632,7 +1632,7 @@ msgid "workflow runs"
 msgstr ""

 #: documents/serialisers.py:514 documents/serialisers.py:871
-#: documents/serialisers.py:2885 documents/views.py:342 documents/views.py:2725
+#: documents/serialisers.py:2885 documents/views.py:343 documents/views.py:2726
 #: paperless_mail/serialisers.py:156
 msgid "Insufficient permissions."
 msgstr ""
@@ -1673,7 +1673,7 @@ msgstr ""
 msgid "Duplicate document identifiers are not allowed."
 msgstr ""

-#: documents/serialisers.py:2971 documents/views.py:4780
+#: documents/serialisers.py:2971 documents/views.py:4763
 #, python-format
 msgid "Documents not found: %(ids)s"
 msgstr ""
@@ -1941,40 +1941,40 @@ msgstr ""
 msgid "Unable to parse URI {value}"
 msgstr ""

-#: documents/views.py:335 documents/views.py:2722
+#: documents/views.py:336 documents/views.py:2723
 msgid "Invalid more_like_id"
 msgstr ""

-#: documents/views.py:1669
+#: documents/views.py:1670
 msgid "Invalid AI configuration."
 msgstr ""

-#: documents/views.py:1680
+#: documents/views.py:1681
 msgid "AI backend request timed out."
 msgstr ""

-#: documents/views.py:1692
+#: documents/views.py:1693
 msgid "AI backend rejected the request. Check logs for details."
 msgstr ""

-#: documents/views.py:2547 documents/views.py:2863
+#: documents/views.py:2548 documents/views.py:2864
 msgid "Specify only one of text, title_search, query, or more_like_id."
 msgstr ""

-#: documents/views.py:4793
+#: documents/views.py:4776
 #, python-format
 msgid "Insufficient permissions to share document %(id)s."
 msgstr ""

-#: documents/views.py:4839
+#: documents/views.py:4822
 msgid "Bundle is already being processed."
 msgstr ""

-#: documents/views.py:4903
+#: documents/views.py:4886
 msgid "The share link bundle is still being prepared. Please try again later."
 msgstr ""

-#: documents/views.py:4917
+#: documents/views.py:4900
 msgid "The share link bundle is unavailable."
 msgstr ""

@@ -2219,198 +2219,194 @@ msgid "Sets the LLM embedding model"
 msgstr ""

 #: paperless/models.py:369
-msgid "Sets the LLM embedding API key"
-msgstr ""
-
-#: paperless/models.py:376
 msgid "Sets the LLM embedding endpoint, optional"
 msgstr ""

-#: paperless/models.py:383
+#: paperless/models.py:376
 msgid "Sets the LLM embedding chunk size"
 msgstr ""

-#: paperless/models.py:389
+#: paperless/models.py:382
 msgid "Sets the LLM context size"
 msgstr ""

-#: paperless/models.py:395
+#: paperless/models.py:388
 msgid "Sets the LLM backend"
 msgstr ""

-#: paperless/models.py:403
+#: paperless/models.py:396
 msgid "Sets the LLM model"
 msgstr ""

-#: paperless/models.py:410
+#: paperless/models.py:403
 msgid "Sets the LLM API key"
 msgstr ""

-#: paperless/models.py:417
+#: paperless/models.py:410
 msgid "Sets the LLM endpoint, optional"
 msgstr ""

-#: paperless/models.py:424
+#: paperless/models.py:417
 msgid "Sets the LLM output language"
 msgstr ""

-#: paperless/models.py:431
+#: paperless/models.py:424
 msgid "Sets the LLM timeout in seconds"
 msgstr ""

-#: paperless/models.py:437
+#: paperless/models.py:430
 msgid "paperless application settings"
 msgstr ""

-#: paperless/settings/__init__.py:559
+#: paperless/settings/__init__.py:558
 msgid "English (US)"
 msgstr ""

-#: paperless/settings/__init__.py:560
+#: paperless/settings/__init__.py:559
 msgid "Arabic"
 msgstr ""

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

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

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

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

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

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

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

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

-#: paperless/settings/__init__.py:569
+#: paperless/settings/__init__.py:568
 msgid "English (GB)"
 msgstr ""

-#: paperless/settings/__init__.py:570
+#: paperless/settings/__init__.py:569
 msgid "Spanish"
 msgstr ""

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

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

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

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

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

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

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

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

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

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

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

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

-#: paperless/settings/__init__.py:583
+#: paperless/settings/__init__.py:582
 msgid "Portuguese (Brazil)"
 msgstr ""

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

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

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

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

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

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

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

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

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

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

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

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

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

@@ -1,6 +1,5 @@
 import dataclasses
 import json
-from typing import Any

 from django.conf import settings

@@ -245,7 +244,6 @@ class AIConfig(BaseConfig):
    ai_enabled: bool = dataclasses.field(init=False)
    llm_embedding_backend: str = dataclasses.field(init=False)
    llm_embedding_model: str = dataclasses.field(init=False)
-    llm_embedding_api_key: str = dataclasses.field(init=False)
    llm_embedding_endpoint: str = dataclasses.field(init=False)
    llm_embedding_chunk_size: int = dataclasses.field(init=False)
    llm_context_size: int = dataclasses.field(init=False)
@@ -256,7 +254,6 @@ class AIConfig(BaseConfig):
    llm_endpoint: str = dataclasses.field(init=False)
    llm_output_language: str = dataclasses.field(init=False)
    llm_allow_internal_endpoints: bool = dataclasses.field(init=False)
-    llm_extra_params: dict[str, Any] = dataclasses.field(init=False)

    def __post_init__(self) -> None:
        app_config = self._get_config_instance()
@@ -272,9 +269,6 @@ class AIConfig(BaseConfig):
        self.llm_embedding_model = (
            app_config.llm_embedding_model or settings.LLM_EMBEDDING_MODEL
        )
-        self.llm_embedding_api_key = (
-            app_config.llm_embedding_api_key or settings.LLM_EMBEDDING_API_KEY
-        )
        self.llm_embedding_endpoint = (
            app_config.llm_embedding_endpoint or settings.LLM_EMBEDDING_ENDPOINT
        )
@@ -293,7 +287,6 @@ class AIConfig(BaseConfig):
            app_config.llm_output_language or settings.LLM_OUTPUT_LANGUAGE
        )
        self.llm_allow_internal_endpoints = settings.LLM_ALLOW_INTERNAL_ENDPOINTS
-        self.llm_extra_params = settings.LLM_EXTRA_PARAMS

    @property
    def llm_index_enabled(self) -> bool:
@@ -1,23 +0,0 @@
-# Generated by Django 5.2.16 on 2026-09-11 09:32
-
-from django.db import migrations
-from django.db import models
-
-
-class Migration(migrations.Migration):
-    dependencies = [
-        ("paperless", "0016_alter_applicationconfiguration_ai_enabled"),
-    ]
-
-    operations = [
-        migrations.AddField(
-            model_name="applicationconfiguration",
-            name="llm_embedding_api_key",
-            field=models.CharField(
-                blank=True,
-                max_length=1024,
-                null=True,
-                verbose_name="Sets the LLM embedding API key",
-            ),
-        ),
-    ]
@@ -365,13 +365,6 @@ class ApplicationConfiguration(AbstractSingletonModel):
        max_length=128,
    )

-    llm_embedding_api_key = models.CharField(
-        verbose_name=_("Sets the LLM embedding API key"),
-        blank=True,
-        null=True,
-        max_length=1024,
-    )
-
    llm_embedding_endpoint = models.CharField(
        verbose_name=_("Sets the LLM embedding endpoint, optional"),
        blank=True,
@@ -1,61 +1,533 @@
+import functools
 import ipaddress
+import logging
+import math
+import re
 import socket
+import time
+from collections.abc import Callable
 from collections.abc import Collection
+from collections.abc import Iterable
+from enum import StrEnum
+from typing import Any
+from typing import Final
+from typing import Self
+from typing import TypeAlias
 from urllib.parse import ParseResult
 from urllib.parse import urlparse

+import anyio
+import httpcore
 import httpx

-# Ranges ipaddress does not report as private, but which routinely front
-# internal infrastructure.
+# Not exported by httpcore; the guard asserts it is still the async default.
+from httpcore._backends.auto import AutoBackend
+
+logger = logging.getLogger("paperless.network")
+
+# requires-python is >=3.11, so no PEP 695 `type` statement.
+IPAddress: TypeAlias = ipaddress.IPv4Address | ipaddress.IPv6Address
+
+# Ranges that ipaddress reports as global but which still reach internal hosts.
 _NON_PUBLIC_NETWORKS = (
-    # RFC 6598 shared address space: ISP CGNAT, and the default pod/service
-    # CIDR on several managed Kubernetes offerings.
-    ipaddress.ip_network("100.64.0.0/10"),
    # RFC 6052 NAT64 well-known prefix: 64:ff9b::7f00:1 is 127.0.0.1 wherever
-    # a NAT64 gateway exists.
+    # a NAT64 gateway exists, yet ipaddress classifies the prefix as global.
    ipaddress.ip_network("64:ff9b::/96"),
 )


-def is_public_ip(ip: str | int) -> bool:
-    try:
-        obj = ipaddress.ip_address(ip)
-        return not (
-            obj.is_private
-            or obj.is_loopback
-            or obj.is_link_local
-            or obj.is_multicast
-            or obj.is_unspecified
-            or any(obj in network for network in _NON_PUBLIC_NETWORKS)
+class BlockReason(StrEnum):
+    NON_PUBLIC_ADDRESS = "non_public_address"
+    UNIX_SOCKET = "unix_socket"
+
+
+class OutboundRequestBlockedError(Exception):
+    """
+    An outbound connection was refused by policy before any socket was opened.
+
+    For NON_PUBLIC_ADDRESS, ``host`` is the name or literal being connected to
+    and ``address`` the first offending address. For UNIX_SOCKET, ``host`` is
+    the socket path and ``port`` and ``address`` are None.
+
+    ``address`` is deliberately left out of the message: the message is logged
+    and stored on failed tasks, and must not disclose internal addresses.
+    """
+
+    def __init__(
+        self,
+        *,
+        host: str,
+        port: int | None,
+        reason: BlockReason,
+        address: IPAddress | None = None,
+    ) -> None:
+        self.host = host
+        self.port = port
+        self.reason = reason
+        self.address = address
+        target = host if port is None else f"{host}:{port}"
+        super().__init__(f"Outbound connection to {target} blocked ({reason})")
+
+    def __reduce__(self) -> tuple[Callable[..., Self], tuple[object, ...]]:
+        # Celery rebuilds failed-task exceptions by pickling; keyword-only
+        # fields cannot be recovered from ``args`` alone.
+        return (
+            functools.partial(
+                type(self),
+                host=self.host,
+                port=self.port,
+                reason=self.reason,
+                address=self.address,
+            ),
+            (),
        )
-    except ValueError:  # pragma: no cover
-        return False


-def resolve_hostname_ips(hostname: str) -> list[str]:
-    try:
-        addr_info = socket.getaddrinfo(hostname, None)
-    except socket.gaierror as e:
-        raise ValueError(f"Could not resolve hostname: {hostname}") from e
+class HostResolutionError(Exception):
+    """The resolver returned no usable addresses for a host."""

-    ips = [info[4][0] for info in addr_info if info and info[4]]
-    if not ips:
-        raise ValueError(f"Could not resolve hostname: {hostname}")
-    return ips
+    def __init__(self, *, host: str, detail: str) -> None:
+        self.host = host
+        self.detail = detail
+        super().__init__(f"Could not resolve {host}: {detail}")
+
+    def __reduce__(self) -> tuple[Callable[..., Self], tuple[object, ...]]:
+        return (
+            functools.partial(type(self), host=self.host, detail=self.detail),
+            (),
+        )


-def format_host_for_url(host: str) -> str:
+def blocked_message(exc: OutboundRequestBlockedError | HostResolutionError) -> str:
+    """User-facing text for validation errors, kept stable for existing callers."""
+    if isinstance(exc, HostResolutionError):
+        return f"Could not resolve hostname: {exc.host}"
+    if exc.reason is BlockReason.UNIX_SOCKET:
+        return "Connection blocked: unix sockets are not permitted"
+    return f"Connection blocked: {exc.host} resolves to a non-public address"
+
+
+def is_public_ip(ip: IPAddress) -> bool:
    """
-    Format IP address for URL use (wrap IPv6 in brackets).
+    True when ``ip`` is globally routable unicast and not in a range that
+    ipaddress reports as global but which still reaches internal hosts.
+    """
+    return (
+        ip.is_global
+        and not ip.is_multicast
+        and not any(ip in network for network in _NON_PUBLIC_NETWORKS)
+    )
+
+
+# Resolver and clock indirection so tests can fake DNS and time for this module
+# without changing how the stock httpcore backends resolve the literals the
+# guard dials.
+_getaddrinfo = socket.getaddrinfo
+_agetaddrinfo = anyio.getaddrinfo
+# The clock is a seam because time-machine does not mock monotonic clocks, and
+# patching time.monotonic globally would also replace the asyncio event loop's
+# own clock, hanging or misfiring its timers for the rest of the test.
+_monotonic = time.monotonic
+
+
+def _collect_addresses(
+    host: str,
+    infos: Iterable[tuple[Any, ...]],
+) -> tuple[IPAddress, ...]:
+    # Resolver output is always an address, but a scoped IPv6 answer carries a
+    # zone id ("fe80::1%1"), which is dropped before classification.
+    # dict keys keep the first occurrence and resolver order
+    addresses: dict[IPAddress, None] = {}
+    for info in infos:
+        address = ipaddress.ip_address(str(info[4][0]).split("%", 1)[0])
+        addresses.setdefault(address, None)
+    if not addresses:
+        raise HostResolutionError(host=host, detail="no addresses returned")
+    return tuple(addresses)
+
+
+def _require_public(
+    host: str,
+    port: int | None,
+    addresses: tuple[IPAddress, ...],
+) -> tuple[IPAddress, ...]:
+    for address in addresses:
+        if not is_public_ip(address):
+            raise OutboundRequestBlockedError(
+                host=host,
+                port=port,
+                reason=BlockReason.NON_PUBLIC_ADDRESS,
+                address=address,
+            )
+    return addresses
+
+
+def resolve_public_addresses(host: str, port: int | None) -> tuple[IPAddress, ...]:
+    """
+    Resolve ``host`` and return its addresses in resolver order, or raise if
+    any of them is non-public. A name is rejected as a whole; offending
+    addresses are never filtered out.
+
+    IP literals go through the resolver too: getaddrinfo answers them without
+    a lookup, and validating only its answer means no second parser can read
+    the host differently from the one that connects.
    """
    try:
-        ip_obj = ipaddress.ip_address(host)
-        if ip_obj.version == 6:
-            return f"[{host}]"
-        return host
-    except ValueError:
-        return host
+        infos = _getaddrinfo(host, port, type=socket.SOCK_STREAM)
+    except (OSError, UnicodeError) as e:
+        raise HostResolutionError(host=host, detail=str(e)) from e
+    return _require_public(host, port, _collect_addresses(host, infos))
+
+
+async def aresolve_public_addresses(
+    host: str,
+    port: int | None,
+) -> tuple[IPAddress, ...]:
+    """Async variant of resolve_public_addresses."""
+    try:
+        infos = await _agetaddrinfo(host, port, type=socket.SOCK_STREAM)
+    except (OSError, UnicodeError) as e:
+        raise HostResolutionError(host=host, detail=str(e)) from e
+    return _require_public(host, port, _collect_addresses(host, infos))
+
+
+MAX_ADDRESSES_TRIED: Final = 8
+MIN_ATTEMPT_TIMEOUT: Final = 2.0
+MAX_ATTEMPT_TIMEOUT: Final = 10.0
+
+
+def _require_positive_timeout(host: str, timeout: float | None) -> None:
+    # A zero timeout makes the socket non-blocking and a negative one is
+    # rejected by settimeout; neither can produce a useful connection attempt.
+    if timeout is not None and timeout <= 0:
+        raise httpcore.ConnectTimeout(
+            f"Connect timeout for {host} must be positive, got {timeout}",
+        )
+
+
+def _deadline(timeout: float | None) -> float:
+    return math.inf if timeout is None else _monotonic() + timeout
+
+
+def _attempt_order(addresses: tuple[IPAddress, ...]) -> list[IPAddress]:
+    # Alternate address families, starting with the resolver's first family
+    # (RFC 8305 section 4), so one unreachable family cannot delay the other.
+    first_version = addresses[0].version
+    primary = [a for a in addresses if a.version == first_version]
+    secondary = [a for a in addresses if a.version != first_version]
+    ordered: list[IPAddress] = []
+    for index in range(max(len(primary), len(secondary))):
+        ordered.extend(primary[index : index + 1])
+        ordered.extend(secondary[index : index + 1])
+    return ordered[:MAX_ADDRESSES_TRIED]
+
+
+def _attempt_timeout(remaining: float, attempts_left: int) -> float:
+    """
+    Budget for the next attempt. Once the budget is too small to split, or on
+    the last address, the attempt gets everything left. Otherwise it gets an
+    equal share clamped to [MIN, MAX], always leaving MIN for a later attempt.
+    The floor survives one lost SYN; the ceiling bounds how long a black-holed
+    address delays the next one.
+    """
+    if attempts_left == 1 or remaining < 2 * MIN_ATTEMPT_TIMEOUT:
+        return remaining
+    share = remaining / attempts_left
+    return min(
+        MAX_ATTEMPT_TIMEOUT,
+        max(MIN_ATTEMPT_TIMEOUT, share),
+        remaining - MIN_ATTEMPT_TIMEOUT,
+    )
+
+
+def _as_httpcore_timeout(seconds: float) -> float | None:
+    return None if math.isinf(seconds) else seconds
+
+
+def _log_block(error: OutboundRequestBlockedError) -> None:
+    logger.warning("Blocked outbound connection: %s", error)
+
+
+def _budget_exhausted(host: str, tried: int, total: int) -> httpcore.ConnectTimeout:
+    return httpcore.ConnectTimeout(
+        f"Timed out connecting to {host} after trying {tried} of {total} addresses",
+    )
+
+
+def _next_attempt_budget(
+    host: str,
+    deadline: float,
+    candidates: list[IPAddress],
+    index: int,
+) -> float:
+    """Budget for the attempt at index, or a timeout if none is left."""
+    remaining = deadline - _monotonic()
+    if remaining <= 0:
+        raise _budget_exhausted(host, index, len(candidates))
+    return _attempt_timeout(remaining, len(candidates) - index)
+
+
+def _resolve_for_connect(host: str, port: int) -> tuple[IPAddress, ...]:
+    try:
+        return resolve_public_addresses(host, port)
+    except OutboundRequestBlockedError as e:
+        _log_block(e)
+        raise
+    except HostResolutionError as e:
+        raise httpcore.ConnectError(str(e)) from e
+
+
+async def _aresolve_for_connect(
+    host: str,
+    port: int,
+    timeout: float | None,
+) -> tuple[IPAddress, ...]:
+    # The scope closes before dialling; attempts are not nested inside it.
+    try:
+        with anyio.fail_after(timeout):
+            return await aresolve_public_addresses(host, port)
+    except TimeoutError as e:
+        raise httpcore.ConnectTimeout(f"Timed out resolving {host}") from e
+    except OutboundRequestBlockedError as e:
+        _log_block(e)
+        raise
+    except HostResolutionError as e:
+        raise httpcore.ConnectError(str(e)) from e
+
+
+class _GuardedSyncBackend(httpcore.NetworkBackend):
+    """
+    Wraps httpcore's sync backend. With internal addresses disallowed, it
+    resolves the origin host itself, rejects the name if any address is
+    non-public, and dials the validated literals so the checked address is
+    the connected one. TLS still verifies against the origin hostname.
+    """
+
+    def __init__(self, inner: httpcore.NetworkBackend, *, allow_internal: bool) -> None:
+        self._inner = inner
+        self._allow_internal = allow_internal
+
+    def connect_tcp(
+        self,
+        host: str,
+        port: int,
+        timeout: float | None = None,
+        local_address: str | None = None,
+        socket_options: Iterable[httpcore.SOCKET_OPTION] | None = None,
+    ) -> httpcore.NetworkStream:
+        if self._allow_internal:
+            return self._inner.connect_tcp(
+                host,
+                port,
+                timeout=timeout,
+                local_address=local_address,
+                socket_options=socket_options,
+            )
+        _require_positive_timeout(host, timeout)
+        # Resolution is not charged to the budget, matching the stock backend.
+        candidates = _attempt_order(_resolve_for_connect(host, port))
+        deadline = _deadline(timeout)
+        last_error: httpcore.ConnectError | httpcore.ConnectTimeout | None = None
+        for index, address in enumerate(candidates):
+            budget = _next_attempt_budget(host, deadline, candidates, index)
+            try:
+                return self._inner.connect_tcp(
+                    str(address),
+                    port,
+                    timeout=_as_httpcore_timeout(budget),
+                    local_address=local_address,
+                    socket_options=socket_options,
+                )
+            except (httpcore.ConnectError, httpcore.ConnectTimeout) as e:
+                logger.debug("Connecting to %s via %s failed: %s", host, address, e)
+                last_error = e
+        # candidates is never empty, so every address was tried and failed
+        raise last_error or _budget_exhausted(host, len(candidates), len(candidates))
+
+    def connect_unix_socket(
+        self,
+        path: str,
+        timeout: float | None = None,
+        socket_options: Iterable[httpcore.SOCKET_OPTION] | None = None,
+    ) -> httpcore.NetworkStream:
+        error = OutboundRequestBlockedError(
+            host=path,
+            port=None,
+            reason=BlockReason.UNIX_SOCKET,
+        )
+        _log_block(error)
+        raise error
+
+    def sleep(self, seconds: float) -> None:
+        self._inner.sleep(seconds)
+
+
+class _GuardedAsyncBackend(httpcore.AsyncNetworkBackend):
+    """Async twin of _GuardedSyncBackend."""
+
+    def __init__(
+        self,
+        inner: httpcore.AsyncNetworkBackend,
+        *,
+        allow_internal: bool,
+    ) -> None:
+        self._inner = inner
+        self._allow_internal = allow_internal
+
+    async def connect_tcp(
+        self,
+        host: str,
+        port: int,
+        timeout: float | None = None,
+        local_address: str | None = None,
+        socket_options: Iterable[httpcore.SOCKET_OPTION] | None = None,
+    ) -> httpcore.AsyncNetworkStream:
+        if self._allow_internal:
+            return await self._inner.connect_tcp(
+                host,
+                port,
+                timeout=timeout,
+                local_address=local_address,
+                socket_options=socket_options,
+            )
+        _require_positive_timeout(host, timeout)
+        # Resolution counts against the budget, matching the stock backend.
+        deadline = _deadline(timeout)
+        candidates = _attempt_order(await _aresolve_for_connect(host, port, timeout))
+        last_error: httpcore.ConnectError | httpcore.ConnectTimeout | None = None
+        for index, address in enumerate(candidates):
+            budget = _next_attempt_budget(host, deadline, candidates, index)
+            try:
+                return await self._inner.connect_tcp(
+                    str(address),
+                    port,
+                    timeout=_as_httpcore_timeout(budget),
+                    local_address=local_address,
+                    socket_options=socket_options,
+                )
+            except (httpcore.ConnectError, httpcore.ConnectTimeout) as e:
+                logger.debug("Connecting to %s via %s failed: %s", host, address, e)
+                last_error = e
+        raise last_error or _budget_exhausted(host, len(candidates), len(candidates))
+
+    async def connect_unix_socket(
+        self,
+        path: str,
+        timeout: float | None = None,
+        socket_options: Iterable[httpcore.SOCKET_OPTION] | None = None,
+    ) -> httpcore.AsyncNetworkStream:
+        error = OutboundRequestBlockedError(
+            host=path,
+            port=None,
+            reason=BlockReason.UNIX_SOCKET,
+        )
+        _log_block(error)
+        raise error
+
+    async def sleep(self, seconds: float) -> None:
+        await self._inner.sleep(seconds)
+
+
+_LAYOUT_ERROR = (
+    "Unexpected httpx transport layout; refusing to create a transport "
+    "without the outbound connection guard"
+)
+
+
+class GuardedHTTPTransport(httpx.HTTPTransport):
+    """
+    httpx transport whose connections pass through the outbound guard.
+
+    Deliberately accepts no proxy, uds or retries options: a proxy would be
+    dialled instead of the destination, and a unix socket bypasses TCP
+    entirely. Adding an option here is a reviewed change, not a pass-through.
+    """
+
+    def __init__(self, *, allow_internal: bool) -> None:
+        super().__init__()
+        # httpx has no public hook for the network backend. Check the exact
+        # layout before swapping so an httpx or httpcore change fails loudly.
+        pool = self._pool
+        if (
+            type(pool) is not httpcore.ConnectionPool
+            or type(pool._network_backend) is not httpcore.SyncBackend
+        ):
+            raise RuntimeError(_LAYOUT_ERROR)
+        pool._network_backend = _GuardedSyncBackend(
+            pool._network_backend,
+            allow_internal=allow_internal,
+        )
+
+
+class GuardedAsyncHTTPTransport(httpx.AsyncHTTPTransport):
+    """Async twin of GuardedHTTPTransport."""
+
+    def __init__(self, *, allow_internal: bool) -> None:
+        super().__init__()
+        pool = self._pool
+        if (
+            type(pool) is not httpcore.AsyncConnectionPool
+            or type(pool._network_backend) is not AutoBackend
+        ):
+            raise RuntimeError(_LAYOUT_ERROR)
+        pool._network_backend = _GuardedAsyncBackend(
+            pool._network_backend,
+            allow_internal=allow_internal,
+        )
+
+
+def create_guarded_httpx_client(
+    url: str,
+    *,
+    allow_internal: bool,
+    timeout: float,
+) -> httpx.Client:
+    """
+    Validate ``url`` up front, then build a client that re-checks at connect
+    time. The up-front check turns static misconfiguration into a ValueError
+    before any retry layer sees it.
+    """
+    validate_outbound_http_url(url, allow_internal=allow_internal)
+    return httpx.Client(
+        transport=GuardedHTTPTransport(allow_internal=allow_internal),
+        timeout=timeout,
+    )
+
+
+def create_guarded_async_httpx_client(
+    url: str,
+    *,
+    allow_internal: bool,
+    timeout: float,
+) -> httpx.AsyncClient:
+    """Async twin of create_guarded_httpx_client."""
+    validate_outbound_http_url(url, allow_internal=allow_internal)
+    return httpx.AsyncClient(
+        transport=GuardedAsyncHTTPTransport(allow_internal=allow_internal),
+        timeout=timeout,
+    )
+
+
+# urllib3 treats a backslash as ending the authority while urlparse and httpx do
+# not, so the host checked here could differ from the one that is dialled.
+# Control and whitespace characters are refused for the same reason.
+_UNSAFE_URL_CHARS = re.compile(r"[\\\x00-\x1f\x7f\s]")
+
+
+def _dns_name(url: str) -> str:
+    """
+    The ASCII hostname that httpx and urllib3 look up for ``url``.
+
+    urlparse keeps a non-ASCII hostname as typed, and getaddrinfo would then
+    encode it with the stdlib IDNA 2003 codec. That maps some characters
+    differently from the IDNA 2008 encoding the HTTP clients use ("faß"
+    becomes "fass" instead of "xn--fa-hia"), so the check would resolve a
+    different name from the one that is connected to.
+    """
+    try:
+        return httpx.URL(url).raw_host.decode("ascii")
+    except (httpx.InvalidURL, UnicodeError) as e:
+        raise ValueError("Invalid URL scheme or hostname.") from e


 def validate_outbound_http_url(
@@ -81,128 +553,17 @@ def validate_outbound_http_url(
        raise ValueError("Destination port not permitted.")

    if not allow_internal:
-        for ip_str in resolve_hostname_ips(parsed.hostname):
-            if not is_public_ip(ip_str):
-                raise ValueError(
-                    f"Connection blocked: {parsed.hostname} resolves to a non-public address",
-                )
+        if _UNSAFE_URL_CHARS.search(url):
+            raise ValueError("Invalid URL scheme or hostname.")
+        host = _dns_name(url)
+        # HTTP clients may percent-decode the host before resolving it, so the
+        # checked name could differ from the dialled one. An IPv6 zone id is the
+        # only legitimate use, and link-local addresses are non-public anyway.
+        if "%" in host:
+            raise ValueError("Invalid URL scheme or hostname.")
+        try:
+            resolve_public_addresses(host, port)
+        except (OutboundRequestBlockedError, HostResolutionError) as e:
+            raise ValueError(blocked_message(e)) from e

    return parsed
-
-
-def _rewrite_request_to_pinned_ip(
-    request: httpx.Request,
-    *,
-    allow_internal: bool,
-) -> httpx.Request:
-    hostname = request.url.host
-
-    if not hostname:
-        raise httpx.ConnectError("No hostname in request URL")
-
-    try:
-        ips = resolve_hostname_ips(hostname)
-    except ValueError as e:
-        raise httpx.ConnectError(str(e)) from e
-
-    if not allow_internal:
-        for ip_str in ips:
-            if not is_public_ip(ip_str):
-                raise httpx.ConnectError(
-                    f"Connection blocked: {hostname} resolves to a non-public address",
-                )
-
-    ip_str = ips[0]
-    formatted_ip = format_host_for_url(ip_str)
-
-    new_headers = httpx.Headers(request.headers)
-    if "host" in new_headers:
-        del new_headers["host"]
-    host_header = format_host_for_url(hostname)
-    default_port = 443 if request.url.scheme == "https" else 80
-    if request.url.port and request.url.port != default_port:
-        host_header = f"{host_header}:{request.url.port}"
-    new_headers["Host"] = host_header
-    new_url = request.url.copy_with(host=formatted_ip)
-
-    rewritten_request = httpx.Request(
-        method=request.method,
-        url=new_url,
-        headers=new_headers,
-        stream=request.stream,
-        extensions=request.extensions,
-    )
-    rewritten_request.extensions["sni_hostname"] = hostname
-
-    return rewritten_request
-
-
-class PinnedHostHTTPTransport(httpx.HTTPTransport):
-    """
-    HTTP transport that resolves/validates hostnames per request and connects to
-    a vetted IP while preserving the original Host header and TLS SNI hostname.
-    """
-
-    def __init__(
-        self,
-        *args,
-        allow_internal: bool = False,
-        **kwargs,
-    ) -> None:
-        super().__init__(*args, **kwargs)
-        self.allow_internal = allow_internal
-
-    def handle_request(self, request: httpx.Request) -> httpx.Response:
-        request = _rewrite_request_to_pinned_ip(
-            request,
-            allow_internal=self.allow_internal,
-        )
-        return super().handle_request(request)
-
-
-class PinnedHostAsyncHTTPTransport(httpx.AsyncHTTPTransport):
-    """
-    Async variant of PinnedHostHTTPTransport.
-    """
-
-    def __init__(
-        self,
-        *args,
-        allow_internal: bool = False,
-        **kwargs,
-    ) -> None:
-        super().__init__(*args, **kwargs)
-        self.allow_internal = allow_internal
-
-    async def handle_async_request(self, request: httpx.Request) -> httpx.Response:
-        request = _rewrite_request_to_pinned_ip(
-            request,
-            allow_internal=self.allow_internal,
-        )
-        return await super().handle_async_request(request)
-
-
-def create_pinned_httpx_client(
-    url: str,
-    *,
-    allow_internal: bool = False,
-    **kwargs,
-) -> httpx.Client:
-    validate_outbound_http_url(url, allow_internal=allow_internal)
-    return httpx.Client(
-        transport=PinnedHostHTTPTransport(allow_internal=allow_internal),
-        **kwargs,
-    )
-
-
-def create_pinned_async_httpx_client(
-    url: str,
-    *,
-    allow_internal: bool = False,
-    **kwargs,
-) -> httpx.AsyncClient:
-    validate_outbound_http_url(url, allow_internal=allow_internal)
-    return httpx.AsyncClient(
-        transport=PinnedHostAsyncHTTPTransport(allow_internal=allow_internal),
-        **kwargs,
-    )
@@ -22,6 +22,8 @@ from pathlib import Path
 from typing import TYPE_CHECKING
 from typing import Self

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

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

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

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

@@ -216,11 +216,6 @@ class ApplicationConfigurationSerializer(
    externally_configured_variables = serializers.SerializerMethodField()
    user_args = serializers.JSONField(binary=True, allow_null=True)
    barcode_tag_mapping = serializers.JSONField(binary=True, allow_null=True)
-    llm_embedding_api_key = ObfuscatedPasswordField(
-        required=False,
-        allow_null=True,
-        max_length=1024,
-    )
    llm_api_key = ObfuscatedPasswordField(
        required=False,
        allow_null=True,
@@ -232,11 +227,7 @@ class ApplicationConfigurationSerializer(
        max_length=1024,
    )

-    OBFUSCATED_FIELDS = (
-        "llm_embedding_api_key",
-        "llm_api_key",
-        "remote_ocr_api_key",
-    )
+    OBFUSCATED_FIELDS = ("llm_api_key", "remote_ocr_api_key")

    def get_externally_configured_variables(
        self,
@@ -7,7 +7,6 @@ import multiprocessing
 import os
 import tempfile
 from pathlib import Path
-from typing import Any
 from typing import Final
 from urllib.parse import urlparse

@@ -1082,25 +1081,6 @@ CLASSIFIER_LANGUAGES: Final[dict[str, str]] = {
 }


-def _get_llm_extra_params() -> dict[str, Any]:
-    """
-    Parse PAPERLESS_AI_LLM_EXTRA_PARAMS, a JSON object passed straight through
-    to the LLM backend's request body.
-    """
-    raw = os.getenv("PAPERLESS_AI_LLM_EXTRA_PARAMS", "{}")
-    try:
-        parsed = json.loads(raw)
-    except json.JSONDecodeError as e:
-        raise ImproperlyConfigured(
-            "PAPERLESS_AI_LLM_EXTRA_PARAMS must be valid JSON",
-        ) from e
-    if not isinstance(parsed, dict):
-        raise ImproperlyConfigured(
-            "PAPERLESS_AI_LLM_EXTRA_PARAMS must be a JSON object",
-        )
-    return parsed
-
-
 def _get_classifier_language_setting(ocr_lang: str) -> str | None:
    """
    Maps the primary Tesseract language to the classifier's stemming
@@ -1236,7 +1216,6 @@ LLM_EMBEDDING_BACKEND = get_choice_from_env(
    {"huggingface", "openai-like", "ollama"},
 )
 LLM_EMBEDDING_MODEL = os.getenv("PAPERLESS_AI_LLM_EMBEDDING_MODEL")
-LLM_EMBEDDING_API_KEY = os.getenv("PAPERLESS_AI_LLM_EMBEDDING_API_KEY")
 LLM_EMBEDDING_ENDPOINT = os.getenv("PAPERLESS_AI_LLM_EMBEDDING_ENDPOINT")
 LLM_EMBEDDING_CHUNK_SIZE = get_int_from_env(
    "PAPERLESS_AI_LLM_EMBEDDING_CHUNK_SIZE",
@@ -1262,4 +1241,3 @@ LLM_ALLOW_INTERNAL_ENDPOINTS = get_bool_from_env(
    "PAPERLESS_AI_LLM_ALLOW_INTERNAL_ENDPOINTS",
    "true",
 )
-LLM_EXTRA_PARAMS = _get_llm_extra_params()
@@ -735,24 +735,33 @@ class TestParser:

            assert expected_html == actual_html

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

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

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

    def test_generate_pdf_from_mail(
        self,
@@ -17,24 +17,6 @@ class TestRemoteUser(DirectoriesMixin, APITestCase):

        self.user = UserFactory(username="temp_admin", superuser=True)

-        # _parse_remote_user_settings() mutates these shared lists in place,
-        # so undo that after the test instead of leaking remote-user auth
-        # into every test that runs afterward.
-        original_middleware = list(settings.MIDDLEWARE)
-        original_auth_backends = list(settings.AUTHENTICATION_BACKENDS)
-        original_auth_classes = list(
-            settings.REST_FRAMEWORK["DEFAULT_AUTHENTICATION_CLASSES"],
-        )
-
-        def _restore_remote_user_settings() -> None:
-            settings.MIDDLEWARE[:] = original_middleware
-            settings.AUTHENTICATION_BACKENDS[:] = original_auth_backends
-            settings.REST_FRAMEWORK["DEFAULT_AUTHENTICATION_CLASSES"][:] = (
-                original_auth_classes
-            )
-
-        self.addCleanup(_restore_remote_user_settings)
-
    def test_remote_user(self) -> None:
        """
        GIVEN:
@@ -7,7 +7,6 @@ from django.core.exceptions import ImproperlyConfigured

 from paperless.settings import _get_allauth_trusted_proxy_count
 from paperless.settings import _get_classifier_language_setting
-from paperless.settings import _get_llm_extra_params
 from paperless.settings import _get_search_language_setting
 from paperless.settings import _parse_paperless_url
 from paperless.settings import default_threads_per_worker
@@ -167,45 +166,3 @@ class TestPaperlessURLSettings(TestCase):

            self.assertIn(url, settings.CSRF_TRUSTED_ORIGINS)
            self.assertIn(url, settings.CORS_ALLOWED_ORIGINS)
-
-
-class TestLlmExtraParams:
-    @pytest.mark.parametrize(
-        ("env_value", "expected"),
-        [
-            pytest.param(None, {}, id="unset"),
-            pytest.param(
-                '{"reasoning_effort": "none"}',
-                {"reasoning_effort": "none"},
-                id="json-object",
-            ),
-        ],
-    )
-    def test_parses(
-        self,
-        monkeypatch,
-        env_value,
-        expected,
-    ):
-        if env_value is None:
-            monkeypatch.delenv("PAPERLESS_AI_LLM_EXTRA_PARAMS", raising=False)
-        else:
-            monkeypatch.setenv("PAPERLESS_AI_LLM_EXTRA_PARAMS", env_value)
-        assert _get_llm_extra_params() == expected
-
-    @pytest.mark.parametrize(
-        ("env_value", "match"),
-        [
-            pytest.param("reasoning_effort=none", "valid JSON", id="invalid-json"),
-            pytest.param('["none"]', "JSON object", id="not-an-object"),
-        ],
-    )
-    def test_invalid_raises(
-        self,
-        monkeypatch,
-        env_value,
-        match,
-    ):
-        monkeypatch.setenv("PAPERLESS_AI_LLM_EXTRA_PARAMS", env_value)
-        with pytest.raises(ImproperlyConfigured, match=match):
-            _get_llm_extra_params()
@@ -1,68 +0,0 @@
-import time
-
-from allauth.mfa import app_settings as mfa_settings
-from allauth.mfa.totp.internal import auth as totp_auth
-from django.test import TestCase
-from django.urls import reverse
-
-from paperless_testing.factories import UserFactory
-
-
-class TestAdminAuth(TestCase):
-    def test_admin_login_redirects_to_allauth(self):
-        user = UserFactory(staff=True, password="testpassword")
-        admin_url = reverse("admin:index")
-        login_url = reverse("admin:login")
-        expected_url = f"{reverse('account_login')}?next={admin_url}"
-
-        response = self.client.get(login_url, {"next": admin_url})
-        self.assertRedirects(response, expected_url)
-
-        response = self.client.post(
-            login_url,
-            {"username": user.username, "password": "testpassword", "next": admin_url},
-        )
-        self.assertRedirects(response, expected_url)
-        self.assertNotIn("_auth_user_id", self.client.session)
-
-    def test_admin_access_requires_totp_for_enrolled_staff(self):
-        user = UserFactory(staff=True, password="testpassword")
-        secret = totp_auth.generate_totp_secret()
-        totp_auth.TOTP.activate(user, secret)
-        admin_url = reverse("admin:index")
-        mfa_url = reverse("mfa_authenticate")
-
-        response = self.client.post(
-            reverse("account_login"),
-            {"login": user.username, "password": "testpassword", "next": admin_url},
-        )
-        self.assertRedirects(response, mfa_url)
-        self.assertNotIn("_auth_user_id", self.client.session)
-        self.assertRedirects(
-            self.client.get(admin_url),
-            f"{reverse('admin:login')}?next={admin_url}",
-            fetch_redirect_response=False,
-        )
-
-        response = self.client.post(mfa_url, {"code": "invalid"})
-        self.assertEqual(response.status_code, 200)
-        self.assertNotIn("_auth_user_id", self.client.session)
-
-        code = totp_auth.format_hotp_value(
-            totp_auth.hotp_value(secret, int(time.time()) // mfa_settings.TOTP_PERIOD),
-        )
-        response = self.client.post(mfa_url, {"code": code})
-        self.assertRedirects(response, admin_url)
-        self.assertEqual(self.client.session["_auth_user_id"], str(user.pk))
-
-    def test_staff_without_totp_can_still_log_in(self):
-        user = UserFactory(staff=True, password="testpassword")
-        admin_url = reverse("admin:index")
-
-        response = self.client.post(
-            reverse("account_login"),
-            {"login": user.username, "password": "testpassword", "next": admin_url},
-        )
-
-        self.assertRedirects(response, admin_url)
-        self.assertEqual(self.client.session["_auth_user_id"], str(user.pk))
@@ -30,27 +30,3 @@ class TestBooleanConfigPrecedence(TestCase):
        config.save()

        self.assertTrue(AIConfig().ai_enabled)
-
-
-class TestAIConfigPrecedence(TestCase):
-    @override_settings(LLM_EMBEDDING_API_KEY="environment-embedding-key")
-    def test_database_embedding_api_key_overrides_environment_setting(self) -> None:
-        config, _ = ApplicationConfiguration.objects.get_or_create()
-        config.llm_embedding_api_key = "database-embedding-key"
-        config.save()
-
-        self.assertEqual(
-            AIConfig().llm_embedding_api_key,
-            "database-embedding-key",
-        )
-
-    @override_settings(LLM_EMBEDDING_API_KEY="environment-embedding-key")
-    def test_null_embedding_api_key_uses_environment_setting(self) -> None:
-        config, _ = ApplicationConfiguration.objects.get_or_create()
-        config.llm_embedding_api_key = None
-        config.save()
-
-        self.assertEqual(
-            AIConfig().llm_embedding_api_key,
-            "environment-embedding-key",
-        )
@@ -0,0 +1,375 @@
+import ipaddress
+import os
+
+import httpcore
+import httpx
+import pytest
+from pytest_mock import MockerFixture
+
+from paperless.network import GuardedAsyncHTTPTransport
+from paperless.network import GuardedHTTPTransport
+from paperless.network import OutboundRequestBlockedError
+from paperless.network import create_guarded_httpx_client
+from paperless_testing.outbound import DialRecorder
+from paperless_testing.outbound import FakeDNS
+from paperless_testing.outbound import LocalHTTPServer
+from paperless_testing.outbound import running_http_server
+
+
+class TestGuardedTransportSync:
+    @pytest.mark.usefixtures("every_address_is_public")
+    def test_pinned_connection_falls_back_to_next_address(
+        self,
+        local_http_server: LocalHTTPServer,
+        fake_dns: FakeDNS,
+        dial_recorder: DialRecorder,
+    ) -> None:
+        """
+        GIVEN:
+            - A hostname resolving to ::1 then 127.0.0.1
+            - A server listening on 127.0.0.1 only
+            - Internal addresses disallowed, with loopback treated as public
+        WHEN:
+            - A request is made
+        THEN:
+            - ::1 fails, 127.0.0.1 is dialled next and the request succeeds
+        """
+        fake_dns.add("dual-stack.test", "::1", "127.0.0.1")
+
+        with httpx.Client(
+            transport=GuardedHTTPTransport(allow_internal=False),
+            timeout=5.0,
+        ) as client:
+            response = client.get(f"http://dual-stack.test:{local_http_server.port}/")
+
+        assert response.status_code == 200
+        assert dial_recorder.hosts() == ["::1", "127.0.0.1"]
+
+    def test_allow_internal_uses_stock_resolution(
+        self,
+        local_http_server: LocalHTTPServer,
+        fake_dns: FakeDNS,
+    ) -> None:
+        """
+        GIVEN:
+            - Internal addresses allowed
+        WHEN:
+            - A request is made to localhost
+        THEN:
+            - It succeeds without the guard resolving anything
+        """
+        with httpx.Client(
+            transport=GuardedHTTPTransport(allow_internal=True),
+            timeout=5.0,
+        ) as client:
+            response = client.get(f"http://localhost:{local_http_server.port}/")
+
+        assert response.status_code == 200
+        assert fake_dns.lookups == []
+
+    @pytest.mark.usefixtures("every_address_is_public")
+    def test_host_header_is_the_hostname(
+        self,
+        local_http_server: LocalHTTPServer,
+        fake_dns: FakeDNS,
+    ) -> None:
+        """
+        GIVEN:
+            - A pinned connection to a named host
+        WHEN:
+            - A request is made
+        THEN:
+            - The server receives the hostname in Host, not the dialled IP
+        """
+        fake_dns.add("pinned.test", "127.0.0.1")
+
+        with httpx.Client(
+            transport=GuardedHTTPTransport(allow_internal=False),
+            timeout=5.0,
+        ) as client:
+            client.get(f"http://pinned.test:{local_http_server.port}/")
+
+        assert local_http_server.requests[0].headers["host"] == (
+            f"pinned.test:{local_http_server.port}"
+        )
+
+    def test_redirect_to_internal_host_is_blocked(
+        self,
+        mocker: MockerFixture,
+        local_http_server: LocalHTTPServer,
+        fake_dns: FakeDNS,
+        dial_recorder: DialRecorder,
+    ) -> None:
+        """
+        GIVEN:
+            - An allowed origin that redirects to a host resolving to a blocked
+              address, and a client that follows redirects
+        WHEN:
+            - The origin is requested
+        THEN:
+            - The redirect hop is blocked without dialling the blocked address
+        """
+        allowed = ipaddress.ip_address("127.0.0.1")
+        mocker.patch(
+            "paperless.network.is_public_ip",
+            side_effect=lambda address: address == allowed,
+        )
+        fake_dns.add("origin.test", "127.0.0.1")
+        fake_dns.add("internal.test", "127.0.0.2")
+        local_http_server.redirect_to = (
+            f"http://internal.test:{local_http_server.port}/"
+        )
+
+        with (
+            httpx.Client(
+                transport=GuardedHTTPTransport(allow_internal=False),
+                timeout=5.0,
+                follow_redirects=True,
+            ) as client,
+            pytest.raises(OutboundRequestBlockedError) as exc_info,
+        ):
+            client.get(f"http://origin.test:{local_http_server.port}/")
+
+        assert exc_info.value.address == ipaddress.ip_address("127.0.0.2")
+        assert dial_recorder.hosts() == ["127.0.0.1"]
+        assert len(local_http_server.requests) == 1
+
+    @pytest.mark.usefixtures("every_address_is_public")
+    def test_connections_are_not_shared_between_hosts_on_one_address(
+        self,
+        local_http_server: LocalHTTPServer,
+        fake_dns: FakeDNS,
+        dial_recorder: DialRecorder,
+    ) -> None:
+        """
+        GIVEN:
+            - Two hostnames resolving to the same address
+            - Internal addresses disallowed, with loopback treated as public
+        WHEN:
+            - One client requests the first host twice, then the second host
+        THEN:
+            - The first host's connection is reused for its second request
+            - The second host gets its own connection, so its certificate would
+              be checked rather than inheriting the first host's session
+        """
+        fake_dns.add("first.test", "127.0.0.1")
+        fake_dns.add("second.test", "127.0.0.1")
+
+        with httpx.Client(
+            transport=GuardedHTTPTransport(allow_internal=False),
+            timeout=5.0,
+        ) as client:
+            client.get(f"http://first.test:{local_http_server.port}/")
+            client.get(f"http://first.test:{local_http_server.port}/")
+            client.get(f"http://second.test:{local_http_server.port}/")
+
+        assert dial_recorder.hosts() == ["127.0.0.1", "127.0.0.1"]
+        assert local_http_server.connections == 2
+        assert [request.headers["host"] for request in local_http_server.requests] == [
+            f"first.test:{local_http_server.port}",
+            f"first.test:{local_http_server.port}",
+            f"second.test:{local_http_server.port}",
+        ]
+
+    @pytest.mark.usefixtures("every_address_is_public")
+    def test_tls_uses_the_hostname_not_the_dialled_address(
+        self,
+        mocker: MockerFixture,
+        local_http_server: LocalHTTPServer,
+        fake_dns: FakeDNS,
+        dial_recorder: DialRecorder,
+    ) -> None:
+        """
+        GIVEN:
+            - A pinned HTTPS connection to a named host
+            - A plain HTTP server, so the handshake itself fails
+        WHEN:
+            - A request is made
+        THEN:
+            - The validated address is dialled
+            - TLS is started with the hostname for SNI and certificate checks
+        """
+        fake_dns.add("pinned.test", "127.0.0.1")
+        start_tls = mocker.spy(httpcore._backends.sync.SyncStream, "start_tls")
+
+        with (
+            httpx.Client(
+                transport=GuardedHTTPTransport(allow_internal=False),
+                timeout=5.0,
+            ) as client,
+            pytest.raises(httpx.ConnectError),
+        ):
+            client.get(f"https://pinned.test:{local_http_server.port}/")
+
+        assert dial_recorder.hosts() == ["127.0.0.1"]
+        start_tls.assert_called_once()
+        assert start_tls.call_args.kwargs["server_hostname"] == "pinned.test"
+
+    @pytest.mark.parametrize(
+        "host",
+        [
+            pytest.param("localhost", id="name"),
+            pytest.param("2130706433", id="decimal"),
+            pytest.param("0x7f.1", id="hex-short"),
+            pytest.param("127.1", id="short-dotted"),
+        ],
+    )
+    def test_blocks_internal_host_without_connecting(
+        self,
+        local_http_server: LocalHTTPServer,
+        dial_recorder: DialRecorder,
+        host: str,
+    ) -> None:
+        """
+        GIVEN:
+            - Internal addresses disallowed
+            - A URL whose host reaches loopback, by name or by a
+              non-canonical spelling of 127.0.0.1
+        WHEN:
+            - A request is made through the transport
+        THEN:
+            - The resolved address is checked, the request is blocked and the
+              server never sees a connection
+        """
+        with (
+            httpx.Client(
+                transport=GuardedHTTPTransport(allow_internal=False),
+                timeout=5.0,
+            ) as client,
+            pytest.raises(OutboundRequestBlockedError),
+        ):
+            client.get(f"http://{host}:{local_http_server.port}/")
+
+        assert local_http_server.connections == 0
+        assert dial_recorder.hosts() == []
+
+    @pytest.mark.usefixtures("every_address_is_public")
+    def test_environment_proxy_is_not_used(
+        self,
+        mocker: MockerFixture,
+        local_http_server: LocalHTTPServer,
+        fake_dns: FakeDNS,
+        dial_recorder: DialRecorder,
+    ) -> None:
+        """
+        GIVEN:
+            - Proxy variables in the environment pointing at a second local server
+            - Internal addresses disallowed
+        WHEN:
+            - A request is made through the production client factory to an
+              allowed origin
+        THEN:
+            - The origin server receives the request directly and the proxy
+              server never sees a connection
+        """
+        with running_http_server() as proxy_server:
+            mocker.patch.dict(
+                os.environ,
+                {
+                    "HTTP_PROXY": f"http://127.0.0.1:{proxy_server.port}",
+                    "HTTPS_PROXY": f"http://127.0.0.1:{proxy_server.port}",
+                    "ALL_PROXY": f"http://127.0.0.1:{proxy_server.port}",
+                },
+            )
+            fake_dns.add("origin.test", "127.0.0.1")
+
+            url = f"http://origin.test:{local_http_server.port}/"
+            with create_guarded_httpx_client(
+                url,
+                allow_internal=False,
+                timeout=5.0,
+            ) as client:
+                response = client.get(url)
+
+            assert response.status_code == 200
+            assert len(local_http_server.requests) == 1
+            assert local_http_server.requests[0].headers["host"] == (
+                f"origin.test:{local_http_server.port}"
+            )
+            assert proxy_server.connections == 0
+            assert proxy_server.requests == []
+            assert dial_recorder.hosts() == ["127.0.0.1"]
+
+
+class TestGuardedTransportAsync:
+    @pytest.fixture(autouse=True)
+    def anyio_backend(self) -> str:
+        return "asyncio"
+
+    @pytest.mark.anyio
+    @pytest.mark.usefixtures("every_address_is_public")
+    async def test_pinned_connection_falls_back_to_next_address(
+        self,
+        local_http_server: LocalHTTPServer,
+        fake_dns: FakeDNS,
+        dial_recorder: DialRecorder,
+    ) -> None:
+        """
+        GIVEN:
+            - A hostname resolving to ::1 then 127.0.0.1
+            - A server listening on 127.0.0.1 only
+            - Internal addresses disallowed, with loopback treated as public
+        WHEN:
+            - An async request is made
+        THEN:
+            - ::1 fails, 127.0.0.1 is dialled next and the request succeeds
+        """
+        fake_dns.add("dual-stack.test", "::1", "127.0.0.1")
+
+        async with httpx.AsyncClient(
+            transport=GuardedAsyncHTTPTransport(allow_internal=False),
+            timeout=5.0,
+        ) as client:
+            response = await client.get(
+                f"http://dual-stack.test:{local_http_server.port}/",
+            )
+
+        assert response.status_code == 200
+        assert dial_recorder.hosts() == ["::1", "127.0.0.1"]
+
+    @pytest.mark.anyio
+    async def test_allow_internal_uses_stock_resolution(
+        self,
+        local_http_server: LocalHTTPServer,
+        fake_dns: FakeDNS,
+    ) -> None:
+        """
+        GIVEN:
+            - Internal addresses allowed
+        WHEN:
+            - An async request is made to localhost
+        THEN:
+            - It succeeds without the guard resolving anything
+        """
+        async with httpx.AsyncClient(
+            transport=GuardedAsyncHTTPTransport(allow_internal=True),
+            timeout=5.0,
+        ) as client:
+            response = await client.get(f"http://localhost:{local_http_server.port}/")
+
+        assert response.status_code == 200
+        assert fake_dns.lookups == []
+
+    @pytest.mark.anyio
+    async def test_blocks_internal_host_without_connecting(
+        self,
+        local_http_server: LocalHTTPServer,
+        dial_recorder: DialRecorder,
+    ) -> None:
+        """
+        GIVEN:
+            - Internal addresses disallowed
+        WHEN:
+            - An async request is made to localhost through the transport
+        THEN:
+            - It is blocked and the server never sees a connection
+        """
+        async with httpx.AsyncClient(
+            transport=GuardedAsyncHTTPTransport(allow_internal=False),
+            timeout=5.0,
+        ) as client:
+            with pytest.raises(OutboundRequestBlockedError):
+                await client.get(f"http://localhost:{local_http_server.port}/")
+
+        assert local_http_server.connections == 0
+        assert dial_recorder.hosts() == []
@@ -1,5 +1,4 @@
 from allauth.account import views as allauth_account_views
-from allauth.account.decorators import secure_admin_login
 from allauth.mfa.base import views as allauth_mfa_views
 from allauth.socialaccount import views as allauth_social_account_views
 from allauth.urls import build_provider_urlpatterns
@@ -69,8 +68,6 @@ from paperless_mail.views import MailRuleViewSet
 from paperless_mail.views import OauthCallbackView
 from paperless_mail.views import ProcessedMailViewSet

-admin.site.login = secure_admin_login(admin.site.login)
-
 api_router = DefaultRouter()
 api_router.register(r"correspondents", CorrespondentViewSet)
 api_router.register(r"document_types", DocumentTypeViewSet)
@@ -14,14 +14,16 @@ if TYPE_CHECKING:
    from llama_index.llms.openai_like import OpenAILike

 from paperless.config import AIConfig
-from paperless.network import PinnedHostAsyncHTTPTransport
-from paperless.network import PinnedHostHTTPTransport
-from paperless.network import create_pinned_async_httpx_client
-from paperless.network import create_pinned_httpx_client
+from paperless.network import GuardedAsyncHTTPTransport
+from paperless.network import GuardedHTTPTransport
+from paperless.network import OutboundRequestBlockedError
+from paperless.network import create_guarded_async_httpx_client
+from paperless.network import create_guarded_httpx_client
 from paperless.network import validate_outbound_http_url
 from paperless_ai.base_model import ClassificationSuggestions
 from paperless_ai.base_model import DocumentClassifierSchema
 from paperless_ai.base_model import model_to_classification_suggestions
+from paperless_ai.exceptions import LLMBlockedError
 from paperless_ai.exceptions import LLMProviderError
 from paperless_ai.exceptions import LLMTimeoutError

@@ -43,6 +45,19 @@ LLM_SYSTEM_PROMPT = (
 PLACEHOLDER_API_KEY: Final = "fake"


+def _find_blocked_cause(exc: BaseException) -> OutboundRequestBlockedError | None:
+    # The openai SDK wraps transport errors in APIConnectionError, so the
+    # block can sit anywhere in the __cause__ chain.
+    current: BaseException | None = exc
+    seen: set[int] = set()
+    while current is not None and id(current) not in seen:
+        if isinstance(current, OutboundRequestBlockedError):
+            return current
+        seen.add(id(current))
+        current = current.__cause__
+    return None
+
+
 class AIClient:
    """
    A client for interacting with an LLM backend.
@@ -63,10 +78,10 @@ class AIClient:
                endpoint,
                allow_internal=self.settings.llm_allow_internal_endpoints,
            )
-            transport = PinnedHostHTTPTransport(
+            transport = GuardedHTTPTransport(
                allow_internal=self.settings.llm_allow_internal_endpoints,
            )
-            async_transport = PinnedHostAsyncHTTPTransport(
+            async_transport = GuardedAsyncHTTPTransport(
                allow_internal=self.settings.llm_allow_internal_endpoints,
            )
            return Ollama(
@@ -75,7 +90,6 @@ class AIClient:
                context_window=self.settings.llm_context_size,
                request_timeout=self.settings.llm_request_timeout,
                system_prompt=LLM_SYSTEM_PROMPT,
-                additional_kwargs=self.settings.llm_extra_params,
                client=Client(
                    host=endpoint,
                    timeout=self.settings.llm_request_timeout,
@@ -94,12 +108,12 @@ class AIClient:
            http_client = None
            async_http_client = None
            if endpoint:
-                http_client = create_pinned_httpx_client(
+                http_client = create_guarded_httpx_client(
                    endpoint,
                    allow_internal=self.settings.llm_allow_internal_endpoints,
                    timeout=self.settings.llm_request_timeout,
                )
-                async_http_client = create_pinned_async_httpx_client(
+                async_http_client = create_guarded_async_httpx_client(
                    endpoint,
                    allow_internal=self.settings.llm_allow_internal_endpoints,
                    timeout=self.settings.llm_request_timeout,
@@ -112,7 +126,6 @@ class AIClient:
                is_chat_model=True,
                is_function_calling_model=True,
                system_prompt=LLM_SYSTEM_PROMPT,
-                additional_kwargs=self.settings.llm_extra_params,
                http_client=http_client,
                async_http_client=async_http_client,
            )
@@ -181,6 +194,12 @@ class AIClient:
        except httpx.TimeoutException as exc:
            raise LLMTimeoutError from exc
        except Exception as exc:
+            blocked = _find_blocked_cause(exc)
+            if blocked is not None:
+                raise LLMBlockedError(
+                    "AI backend request was blocked by the outbound request "
+                    f"policy: {blocked}",
+                ) from exc
            if self._is_openai_timeout(exc):
                raise LLMTimeoutError from exc
            if self._is_provider_error(exc):
@@ -9,10 +9,10 @@ if TYPE_CHECKING:
 from documents.models import Document
 from paperless.config import AIConfig
 from paperless.models import LLMEmbeddingBackend
-from paperless.network import PinnedHostAsyncHTTPTransport
-from paperless.network import PinnedHostHTTPTransport
-from paperless.network import create_pinned_async_httpx_client
-from paperless.network import create_pinned_httpx_client
+from paperless.network import GuardedAsyncHTTPTransport
+from paperless.network import GuardedHTTPTransport
+from paperless.network import create_guarded_async_httpx_client
+from paperless.network import create_guarded_httpx_client
 from paperless.network import validate_outbound_http_url
 from paperless_ai.client import PLACEHOLDER_API_KEY

@@ -29,21 +29,19 @@ def get_embedding_model(config: AIConfig) -> "BaseEmbedding":
            http_client = None
            async_http_client = None
            if endpoint:
-                http_client = create_pinned_httpx_client(
+                http_client = create_guarded_httpx_client(
                    endpoint,
                    allow_internal=config.llm_allow_internal_endpoints,
                    timeout=config.llm_request_timeout,
                )
-                async_http_client = create_pinned_async_httpx_client(
+                async_http_client = create_guarded_async_httpx_client(
                    endpoint,
                    allow_internal=config.llm_allow_internal_endpoints,
                    timeout=config.llm_request_timeout,
                )
            return OpenAILikeEmbedding(
                model_name=config.llm_embedding_model or "text-embedding-3-small",
-                api_key=config.llm_embedding_api_key
-                or config.llm_api_key
-                or PLACEHOLDER_API_KEY,
+                api_key=config.llm_api_key or PLACEHOLDER_API_KEY,
                api_base=endpoint,
                timeout=config.llm_request_timeout,
                http_client=http_client,
@@ -79,14 +77,14 @@ def get_embedding_model(config: AIConfig) -> "BaseEmbedding":
            embedding._client = Client(
                host=endpoint,
                timeout=config.llm_request_timeout,
-                transport=PinnedHostHTTPTransport(
+                transport=GuardedHTTPTransport(
                    allow_internal=config.llm_allow_internal_endpoints,
                ),
            )
            embedding._async_client = AsyncClient(
                host=endpoint,
                timeout=config.llm_request_timeout,
-                transport=PinnedHostAsyncHTTPTransport(
+                transport=GuardedAsyncHTTPTransport(
                    allow_internal=config.llm_allow_internal_endpoints,
                ),
            )
@@ -4,3 +4,7 @@ class LLMTimeoutError(Exception):

 class LLMProviderError(Exception):
    """The LLM backend rejected the request."""
+
+
+class LLMBlockedError(Exception):
+    """The outbound request policy refused the connection to the LLM backend."""
@@ -1,3 +1,4 @@
+import ipaddress
 import json
 from unittest.mock import ANY
 from unittest.mock import MagicMock
@@ -9,11 +10,15 @@ import openai
 import pytest
 from llama_index.core.llms.llm import ToolSelection

+from paperless.network import BlockReason
+from paperless.network import OutboundRequestBlockedError
 from paperless_ai.client import LLM_SYSTEM_PROMPT
 from paperless_ai.client import PLACEHOLDER_API_KEY
 from paperless_ai.client import AIClient
+from paperless_ai.exceptions import LLMBlockedError
 from paperless_ai.exceptions import LLMProviderError
 from paperless_ai.exceptions import LLMTimeoutError
+from paperless_testing.outbound import guard_of


@pytest.fixture
@@ -23,7 +28,6 @@ def mock_ai_config():
        mock_config.llm_allow_internal_endpoints = True
        mock_config.llm_context_size = 8192
        mock_config.llm_request_timeout = 120
-        mock_config.llm_extra_params = {}
        MockAIConfig.return_value = mock_config
        yield mock_config

@@ -53,7 +57,6 @@ def test_get_llm_ollama(mock_ai_config, mock_ollama_llm):
        context_window=8192,
        request_timeout=120,
        system_prompt=LLM_SYSTEM_PROMPT,
-        additional_kwargs={},
        client=ANY,
        async_client=ANY,
    )
@@ -76,7 +79,6 @@ def test_get_llm_openai(mock_ai_config, mock_openai_llm):
        is_chat_model=True,
        is_function_calling_model=True,
        system_prompt=LLM_SYSTEM_PROMPT,
-        additional_kwargs={},
        http_client=ANY,
        async_http_client=ANY,
    )
@@ -199,36 +201,6 @@ def test_run_llm_query_openai_uses_tools(mock_ai_config, mock_openai_llm):
    )


-@pytest.mark.parametrize(
-    ("backend", "llm_fixture"),
-    [
-        pytest.param("openai-like", "mock_openai_llm", id="openai-like"),
-        pytest.param("ollama", "mock_ollama_llm", id="ollama"),
-    ],
-)
-def test_get_llm_passes_extra_params(request, mock_ai_config, backend, llm_fixture):
-    """
-    GIVEN:
-        - Extra LLM params configured, e.g. for a provider that needs a
-          parameter we do not set ourselves
-    WHEN:
-        - The client builds the LLM
-    THEN:
-        - They are handed to the backend as additional_kwargs
-    """
-    llm_mock = request.getfixturevalue(llm_fixture)
-    mock_ai_config.llm_backend = backend
-    mock_ai_config.llm_model = "gpt-5.6-luna"
-    mock_ai_config.llm_endpoint = "http://test-url"
-    mock_ai_config.llm_extra_params = {"reasoning_effort": "none"}
-
-    AIClient()
-
-    assert llm_mock.call_args.kwargs["additional_kwargs"] == {
-        "reasoning_effort": "none",
-    }
-
-
 def test_run_llm_query_openai_timeout_raises_local_error(
    mock_ai_config,
    mock_openai_llm,
@@ -310,3 +282,142 @@ def test_run_llm_query_httpx_timeout_raises_local_error(

    with pytest.raises(LLMTimeoutError):
        client.run_llm_query("test_prompt")
+
+
+class TestGuardedLLMClients:
+    @pytest.mark.parametrize(
+        ("endpoint", "allow_internal"),
+        [
+            pytest.param("http://test-url", True, id="internal-allowed"),
+            pytest.param("http://93.184.216.34:11434", False, id="internal-blocked"),
+        ],
+    )
+    def test_ollama_clients_are_guarded(
+        self,
+        mock_ai_config: MagicMock,
+        mock_ollama_llm: MagicMock,
+        endpoint: str,
+        *,
+        allow_internal: bool,
+    ) -> None:
+        """
+        GIVEN:
+            - The Ollama backend
+        WHEN:
+            - The LLM is built
+        THEN:
+            - Its sync and async clients use guarded transports with the setting
+        """
+        mock_ai_config.llm_backend = "ollama"
+        mock_ai_config.llm_model = "test_model"
+        mock_ai_config.llm_endpoint = endpoint
+        mock_ai_config.llm_allow_internal_endpoints = allow_internal
+
+        AIClient()
+
+        kwargs = mock_ollama_llm.call_args.kwargs
+        assert guard_of(kwargs["client"]._client)._allow_internal is allow_internal
+        assert (
+            guard_of(kwargs["async_client"]._client)._allow_internal is allow_internal
+        )
+
+    @pytest.mark.parametrize(
+        ("endpoint", "allow_internal"),
+        [
+            pytest.param("http://test-url", True, id="internal-allowed"),
+            pytest.param("http://93.184.216.34:8080", False, id="internal-blocked"),
+        ],
+    )
+    def test_openai_like_clients_are_guarded(
+        self,
+        mock_ai_config: MagicMock,
+        mock_openai_llm: MagicMock,
+        endpoint: str,
+        *,
+        allow_internal: bool,
+    ) -> None:
+        """
+        GIVEN:
+            - The OpenAI-like backend with an endpoint
+        WHEN:
+            - The LLM is built
+        THEN:
+            - Its sync and async http clients use guarded transports
+        """
+        mock_ai_config.llm_backend = "openai-like"
+        mock_ai_config.llm_model = "test_model"
+        mock_ai_config.llm_api_key = "key"
+        mock_ai_config.llm_endpoint = endpoint
+        mock_ai_config.llm_allow_internal_endpoints = allow_internal
+
+        AIClient()
+
+        kwargs = mock_openai_llm.call_args.kwargs
+        assert guard_of(kwargs["http_client"])._allow_internal is allow_internal
+        assert guard_of(kwargs["async_http_client"])._allow_internal is allow_internal
+
+
+def _block() -> OutboundRequestBlockedError:
+    return OutboundRequestBlockedError(
+        host="llm.example",
+        port=443,
+        reason=BlockReason.NON_PUBLIC_ADDRESS,
+        address=ipaddress.ip_address("10.0.0.1"),
+    )
+
+
+class TestBlockedLLMRequests:
+    def test_ollama_block_becomes_llm_blocked_error(
+        self,
+        mock_ai_config: MagicMock,
+        mock_ollama_llm: MagicMock,
+    ) -> None:
+        """
+        GIVEN:
+            - The Ollama backend and a connection blocked by policy
+        WHEN:
+            - An LLM query runs
+        THEN:
+            - LLMBlockedError is raised with a message, chained to the block
+            - The message, which tracked tasks store, names the destination but
+              not the resolved internal address
+        """
+        mock_ai_config.llm_backend = "ollama"
+        mock_ai_config.llm_model = "test_model"
+        mock_ai_config.llm_endpoint = "http://test-url"
+        block = _block()
+        mock_ollama_llm.return_value.chat.side_effect = block
+
+        with pytest.raises(LLMBlockedError) as exc_info:
+            AIClient().run_llm_query("test_prompt")
+
+        assert exc_info.value.__cause__ is block
+        assert "llm.example:443" in str(exc_info.value)
+        assert "10.0.0.1" not in str(exc_info.value)
+
+    def test_openai_wrapped_block_becomes_llm_blocked_error(
+        self,
+        mock_ai_config: MagicMock,
+        mock_openai_llm: MagicMock,
+    ) -> None:
+        """
+        GIVEN:
+            - The OpenAI-like backend, whose SDK wraps the block in
+              APIConnectionError
+        WHEN:
+            - An LLM query runs
+        THEN:
+            - LLMBlockedError is raised
+        """
+        mock_ai_config.llm_backend = "openai-like"
+        mock_ai_config.llm_model = "test_model"
+        mock_ai_config.llm_api_key = "key"
+        mock_ai_config.llm_endpoint = "http://test-url"
+        wrapped = openai.APIConnectionError(
+            request=httpx.Request("POST", "http://test-url/v1/chat/completions"),
+        )
+        wrapped.__cause__ = _block()
+        mock_openai_llm.return_value.chat_with_tools.side_effect = wrapped
+
+        with pytest.raises(LLMBlockedError):
+            AIClient().run_llm_query("test_prompt")
@@ -1,9 +1,12 @@
+from typing import TYPE_CHECKING
+from typing import cast
 from unittest.mock import ANY
 from unittest.mock import MagicMock
 from unittest.mock import patch

 import pytest
 from django.conf import settings
+from pytest_mock import MockerFixture

 from documents.models import Document
 from paperless.models import LLMEmbeddingBackend
@@ -12,12 +15,15 @@ from paperless_ai.embedding import _normalize_llm_index_text
 from paperless_ai.embedding import build_llm_index_text
 from paperless_ai.embedding import get_configured_model_name
 from paperless_ai.embedding import get_embedding_model
+from paperless_testing.outbound import guard_of
+
+if TYPE_CHECKING:
+    from llama_index.embeddings.ollama import OllamaEmbedding


@pytest.fixture
 def mock_ai_config():
    with patch("paperless_ai.embedding.AIConfig") as MockAIConfig:
-        MockAIConfig.return_value.llm_embedding_api_key = None
        MockAIConfig.return_value.llm_embedding_endpoint = None
        MockAIConfig.return_value.llm_allow_internal_endpoints = True
        MockAIConfig.return_value.llm_context_size = 8192
@@ -64,7 +70,6 @@ def mock_document():
 def test_get_embedding_model_openai(mock_ai_config):
    mock_ai_config.return_value.llm_embedding_backend = LLMEmbeddingBackend.OPENAI_LIKE
    mock_ai_config.return_value.llm_embedding_model = "text-embedding-3-small"
-    mock_ai_config.return_value.llm_embedding_api_key = "test_embedding_api_key"
    mock_ai_config.return_value.llm_api_key = "test_api_key"
    mock_ai_config.return_value.llm_endpoint = "http://test-url"

@@ -74,7 +79,7 @@ def test_get_embedding_model_openai(mock_ai_config):
        model = get_embedding_model(mock_ai_config.return_value)
        MockOpenAIEmbedding.assert_called_once_with(
            model_name="text-embedding-3-small",
-            api_key="test_embedding_api_key",
+            api_key="test_api_key",
            api_base="http://test-url",
            timeout=120,
            http_client=ANY,
@@ -83,20 +88,6 @@ def test_get_embedding_model_openai(mock_ai_config):
        assert model == MockOpenAIEmbedding.return_value


-def test_get_embedding_model_openai_falls_back_to_llm_api_key(mock_ai_config):
-    mock_ai_config.return_value.llm_embedding_backend = LLMEmbeddingBackend.OPENAI_LIKE
-    mock_ai_config.return_value.llm_embedding_model = "text-embedding-3-small"
-    mock_ai_config.return_value.llm_api_key = "test_api_key"
-    mock_ai_config.return_value.llm_endpoint = "http://test-url"
-
-    with patch(
-        "llama_index.embeddings.openai_like.OpenAILikeEmbedding",
-    ) as MockOpenAIEmbedding:
-        get_embedding_model(mock_ai_config.return_value)
-
-    assert MockOpenAIEmbedding.call_args.kwargs["api_key"] == "test_api_key"
-
-
@pytest.mark.parametrize("configured_key", [None, ""])
 def test_get_embedding_model_openai_without_api_key_sends_placeholder(
    mock_ai_config,
@@ -105,7 +96,6 @@ def test_get_embedding_model_openai_without_api_key_sends_placeholder(
    """Same required key handling as the LLM client, see #13831."""
    mock_ai_config.return_value.llm_embedding_backend = LLMEmbeddingBackend.OPENAI_LIKE
    mock_ai_config.return_value.llm_embedding_model = "text-embedding-3-small"
-    mock_ai_config.return_value.llm_embedding_api_key = configured_key
    mock_ai_config.return_value.llm_api_key = configured_key
    mock_ai_config.return_value.llm_endpoint = "http://test-url"

@@ -300,3 +290,61 @@ def test_normalize_llm_index_text_collapses_ocr_leaders_without_joining_lines():

 def test_normalize_llm_index_text_collapses_non_breaking_spaces():
    assert _normalize_llm_index_text("A\u00a0........\u00a0B") == "A B"
+
+
+class TestGuardedEmbeddingClients:
+    def test_ollama_embedding_clients_are_guarded(
+        self,
+        mocker: MockerFixture,
+        mock_ai_config: MagicMock,
+    ) -> None:
+        """
+        GIVEN:
+            - The Ollama embedding backend
+        WHEN:
+            - The embedding model is built
+        THEN:
+            - The clients swapped onto it use guarded transports
+        """
+        config = mock_ai_config.return_value
+        config.llm_embedding_backend = LLMEmbeddingBackend.OLLAMA
+        config.llm_embedding_model = "embeddinggemma"
+        config.llm_endpoint = "http://93.184.216.34:11434"
+        config.llm_allow_internal_endpoints = False
+
+        mocker.patch("llama_index.embeddings.ollama.OllamaEmbedding")
+
+        model = cast("OllamaEmbedding", get_embedding_model(config))
+
+        assert guard_of(model._client._client)._allow_internal is False
+        assert guard_of(model._async_client._client)._allow_internal is False
+
+    def test_openai_like_embedding_clients_are_guarded(
+        self,
+        mocker: MockerFixture,
+        mock_ai_config: MagicMock,
+    ) -> None:
+        """
+        GIVEN:
+            - The OpenAI-like embedding backend with an endpoint
+        WHEN:
+            - The embedding model is built
+        THEN:
+            - Its http clients use guarded transports
+        """
+        config = mock_ai_config.return_value
+        config.llm_embedding_backend = LLMEmbeddingBackend.OPENAI_LIKE
+        config.llm_embedding_model = "text-embedding-3-small"
+        config.llm_api_key = "key"
+        config.llm_endpoint = "http://93.184.216.34:8080"
+        config.llm_allow_internal_endpoints = False
+
+        embedding_class = mocker.patch(
+            "llama_index.embeddings.openai_like.OpenAILikeEmbedding",
+        )
+
+        get_embedding_model(config)
+
+        kwargs = embedding_class.call_args.kwargs
+        assert guard_of(kwargs["http_client"])._allow_internal is False
+        assert guard_of(kwargs["async_http_client"])._allow_internal is False
@@ -45,8 +45,11 @@ from documents.models import Correspondent
 from documents.models import PaperlessTask
 from documents.parsers import is_mime_type_supported
 from documents.tasks import consume_file
-from paperless.network import is_public_ip
-from paperless.network import resolve_hostname_ips
+from paperless.network import HostResolutionError
+from paperless.network import IPAddress
+from paperless.network import OutboundRequestBlockedError
+from paperless.network import blocked_message
+from paperless.network import resolve_public_addresses
 from paperless_mail.models import MailAccount
 from paperless_mail.models import MailRule
 from paperless_mail.models import ProcessedMail
@@ -445,18 +448,34 @@ class PinnedIMAP4(imaplib.IMAP4):

    Without pinned addresses, and with the ssl_context of the matching imaplib
    class, this behaves exactly like imaplib.IMAP4 / imaplib.IMAP4_SSL.
+
+    ``pinned_ips`` of ``None`` means no pinning was requested and the stock
+    imaplib connection path is used. An empty tuple means pinning was requested
+    and yielded nothing, and the connection fails without opening a socket
+    rather than falling back to a hostname lookup.
    """

-    def __init__(self, host, port, pinned_ips, ssl_context=None, timeout=None) -> None:
+    def __init__(
+        self,
+        host: str,
+        port: int | None,
+        pinned_ips: tuple[IPAddress, ...] | None,
+        ssl_context: ssl.SSLContext | None = None,
+        timeout: float | None = None,
+    ) -> None:
        self._pinned_ips = pinned_ips
        self.ssl_context = ssl_context
        super().__init__(host, port, timeout=timeout)

-    def _connect_pinned(self, timeout):
+    def _connect_pinned(
+        self,
+        pinned_ips: tuple[IPAddress, ...],
+        timeout: float | None,
+    ) -> socket.socket:
        last_error: OSError | None = None
-        for ip_str in self._pinned_ips:
+        for ip in pinned_ips:
            try:
-                address = (ip_str, self.port)
+                address = (str(ip), self.port)
                if timeout is not None:
                    return socket.create_connection(address, timeout)
                return socket.create_connection(address)
@@ -464,9 +483,9 @@ class PinnedIMAP4(imaplib.IMAP4):
                last_error = e
        raise last_error or OSError(f"Could not connect to {self.host}")

-    def _create_socket(self, timeout):
-        if self._pinned_ips:
-            sock = self._connect_pinned(timeout)
+    def _create_socket(self, timeout: float | None) -> socket.socket:
+        if self._pinned_ips is not None:
+            sock = self._connect_pinned(self._pinned_ips, timeout)
        else:
            sock = super()._create_socket(timeout)
        if self.ssl_context is None:
@@ -477,7 +496,12 @@ class PinnedIMAP4(imaplib.IMAP4):
 class PinnedClientMixin:
    """Builds the imaplib client against the pre-resolved addresses, if any."""

-    def __init__(self, *args, pinned_ips: list[str] | None, **kwargs) -> None:
+    def __init__(
+        self,
+        *args,
+        pinned_ips: tuple[IPAddress, ...] | None,
+        **kwargs,
+    ) -> None:
        self._pinned_ips = pinned_ips
        super().__init__(*args, **kwargs)

@@ -515,22 +539,20 @@ class PinnedMailBoxStartTls(PinnedClientMixin, MailBoxStartTls):
        return client


-def get_mailbox(server, port, security) -> MailBox:
+def get_mailbox(
+    server: str,
+    port: int | None,
+    security: int,
+) -> MailBox:
    """
    Returns the correct MailBox instance for the given configuration.
    """
-    pinned_ips: list[str] | None = None
+    pinned_ips: tuple[IPAddress, ...] | None = None
    if not settings.EMAIL_ALLOW_INTERNAL_HOSTS:
        try:
-            pinned_ips = resolve_hostname_ips(server)
-        except ValueError as e:
-            raise MailError(str(e)) from e
-
-        for ip_str in pinned_ips:
-            if not is_public_ip(ip_str):
-                raise MailError(
-                    f"Connection blocked: {server} resolves to a non-public address",
-                )
+            pinned_ips = resolve_public_addresses(server, port)
+        except (OutboundRequestBlockedError, HostResolutionError) as e:
+            raise MailError(blocked_message(e)) from e

    ssl_context = ssl.create_default_context()
    if settings.EMAIL_CERTIFICATE_FILE is not None:  # pragma: no cover
@@ -1,9 +1,12 @@
 import dataclasses
+import ipaddress
+import socket
 import time
 import uuid
 from collections import namedtuple
 from datetime import timedelta
 from unittest import mock
+from unittest.mock import MagicMock

 import pytest
 from django.contrib.auth.models import Permission
@@ -25,6 +28,7 @@ from documents.models import MatchingModel
 from paperless_mail import tasks
 from paperless_mail.mail import MailAccountHandler
 from paperless_mail.mail import MailError
+from paperless_mail.mail import PinnedIMAP4
 from paperless_mail.mail import TagMailAction
 from paperless_mail.mail import apply_mail_action
 from paperless_mail.mail import error_callback
@@ -1565,12 +1569,7 @@ class TestMail(
            ("electronic", None, "invoices@mycompany.com", None, 1),
            (None, "amazon", "me@myselfandi.com", None, 1),
        ]:
-            with self.subTest(
-                f_body=f_body,
-                f_from=f_from,
-                f_to=f_to,
-                f_subject=f_subject,
-            ):
+            with self.subTest(f_body=f_body, f_from=f_from, f_subject=f_subject):
                MailRule.objects.all().delete()
                _ = MailRule.objects.create(
                    name="testrule3",
@@ -1811,7 +1810,7 @@ class TestPostConsumeAction(TestCase):

        with (
            self.assertRaises(errors.ImapToolsError),
-            self.assertLogs("paperless_mail", level="ERROR") as cm,
+            self.assertLogs("paperless.mail", level="ERROR") as cm,
        ):
            apply_mail_action(
                result=[],
@@ -1820,10 +1819,9 @@ class TestPostConsumeAction(TestCase):
                message_subject=self.message_subject,
                message_date=self.message_date,
            )
-
-        error_str = cm.output[0]
-        expected_str = "Error while processing mail action during post_consume"
-        self.assertIn(expected_str, error_str)
+            error_str = cm.output[0]
+            expected_str = "Error while processing mail action during post_consume"
+            self.assertIn(expected_str, error_str)

        processed_mail = ProcessedMail.objects.get(uid=self.message_uid)
        self.assertEqual(processed_mail.status, "FAILED")
@@ -2051,10 +2049,13 @@ class TestMailAccountTestView(APITestCase):
        self.assertEqual(response.content.decode(), "Unable to connect to server")

    @override_settings(EMAIL_ALLOW_INTERNAL_HOSTS=False)
-    @mock.patch("paperless_mail.mail.resolve_hostname_ips", return_value=["127.0.0.1"])
+    @mock.patch(
+        "paperless.network._getaddrinfo",
+        return_value=[(socket.AF_INET, socket.SOCK_STREAM, 6, "", ("127.0.0.1", 993))],
+    )
    def test_mail_account_test_view_blocks_internal_host_when_disabled(
        self,
-        _mock_resolve_hostname_ips,
+        _mock_getaddrinfo: MagicMock,
    ) -> None:
        data = {
            "imap_server": "internal.example",
@@ -2211,10 +2212,10 @@ class TestGetMailboxHostPinning(TestCase):

    @override_settings(EMAIL_ALLOW_INTERNAL_HOSTS=False)
    @mock.patch(
-        "paperless_mail.mail.resolve_hostname_ips",
-        return_value=["93.184.216.34"],
+        "paperless_mail.mail.resolve_public_addresses",
+        return_value=(ipaddress.ip_address("93.184.216.34"),),
    )
-    def test_connects_to_validated_ip(self, _mock_resolve) -> None:
+    def test_connects_to_validated_ip(self, _mock_resolve: MagicMock) -> None:
        with mock.patch(
            "paperless_mail.mail.socket.create_connection",
            side_effect=OSError("no connection in tests"),
@@ -2231,10 +2232,13 @@ class TestGetMailboxHostPinning(TestCase):

    @override_settings(EMAIL_ALLOW_INTERNAL_HOSTS=False)
    @mock.patch(
-        "paperless_mail.mail.resolve_hostname_ips",
-        return_value=["93.184.216.34"],
+        "paperless_mail.mail.resolve_public_addresses",
+        return_value=(ipaddress.ip_address("93.184.216.34"),),
    )
-    def test_ssl_pins_ip_but_keeps_hostname_for_sni(self, _mock_resolve) -> None:
+    def test_ssl_pins_ip_but_keeps_hostname_for_sni(
+        self,
+        _mock_resolve: MagicMock,
+    ) -> None:
        ssl_context = mock.MagicMock()
        ssl_context.wrap_socket.return_value.makefile.side_effect = OSError(
            "no connection in tests",
@@ -2265,13 +2269,51 @@ class TestGetMailboxHostPinning(TestCase):

    @override_settings(EMAIL_ALLOW_INTERNAL_HOSTS=False)
    @mock.patch(
-        "paperless_mail.mail.resolve_hostname_ips",
-        return_value=["93.184.216.34", "127.0.0.1"],
+        "paperless.network._getaddrinfo",
+        return_value=[
+            (socket.AF_INET, socket.SOCK_STREAM, 6, "", ("93.184.216.34", 993)),
+            (socket.AF_INET, socket.SOCK_STREAM, 6, "", ("127.0.0.1", 993)),
+        ],
    )
-    def test_blocks_when_any_resolved_address_is_internal(self, _mock_resolve) -> None:
-        with self.assertRaises(MailError):
+    def test_blocks_when_any_resolved_address_is_internal(
+        self,
+        _mock_resolve: MagicMock,
+    ) -> None:
+        """
+        GIVEN:
+            - A mail host resolving to one public and one loopback address
+            - EMAIL_ALLOW_INTERNAL_HOSTS is False
+        WHEN:
+            - A mailbox is requested
+        THEN:
+            - The whole host is blocked with the existing message
+        """
+        with self.assertRaisesMessage(
+            MailError,
+            "Connection blocked: mail.example.com resolves to a non-public address",
+        ):
            get_mailbox("mail.example.com", 993, MailAccount.ImapSecurity.SSL)

+    def test_empty_pin_list_never_falls_back_to_hostname_lookup(self) -> None:
+        """
+        GIVEN:
+            - A pinned IMAP client given an empty tuple of addresses
+        WHEN:
+            - It connects
+        THEN:
+            - It fails without opening any socket, rather than resolving the
+              hostname itself
+        """
+        with (
+            mock.patch("paperless_mail.mail.socket.create_connection") as pinned,
+            mock.patch("imaplib.IMAP4._create_socket") as unpinned,
+            self.assertRaises(OSError),
+        ):
+            PinnedIMAP4("mail.example.com", 143, ())
+
+        pinned.assert_not_called()
+        unpinned.assert_not_called()
+

 class TestMailAccountProcess(APITestCase):
    def setUp(self) -> None:
@@ -37,7 +37,6 @@ class PaperlessDirs:
    logging_dir: Path
    model_file: Path
    media_lock: Path
-    share_link_bundle_dir: Path


 class DirSettings(TypedDict):
@@ -55,7 +54,6 @@ class DirSettings(TypedDict):
    STATIC_ROOT: Path
    MODEL_FILE: Path
    MEDIA_LOCK: Path
-    SHARE_LINK_BUNDLE_DIR: Path


 def build_paperless_dirs(root: Path) -> PaperlessDirs:
@@ -77,7 +75,6 @@ def build_paperless_dirs(root: Path) -> PaperlessDirs:
        logging_dir=data_dir / "log",
        model_file=data_dir / "classification_model.pickle",
        media_lock=media_dir / "media.lock",
-        share_link_bundle_dir=documents_dir / "share_link_bundles",
    )

    for directory in (
@@ -112,7 +109,6 @@ def dirs_settings(dirs: PaperlessDirs) -> DirSettings:
        STATIC_ROOT=dirs.static_dir,
        MODEL_FILE=dirs.model_file,
        MEDIA_LOCK=dirs.media_lock,
-        SHARE_LINK_BUNDLE_DIR=dirs.share_link_bundle_dir,
    )


@@ -0,0 +1,218 @@
+"""
+Real-socket helpers for tests of the outbound connection guard in
+paperless.network: a local HTTP server, a per-hostname resolver fake and
+spies recording which addresses were actually dialled.
+
+The fixtures wrapping these live in the root conftest.
+"""
+
+from __future__ import annotations
+
+import http.server
+import socket
+import threading
+from contextlib import contextmanager
+from dataclasses import dataclass
+from dataclasses import field
+from typing import TYPE_CHECKING
+from typing import Any
+from typing import cast
+
+import anyio
+import httpcore
+
+from paperless.network import GuardedAsyncHTTPTransport
+from paperless.network import GuardedHTTPTransport
+from paperless.network import _GuardedAsyncBackend
+from paperless.network import _GuardedSyncBackend
+
+if TYPE_CHECKING:
+    from collections.abc import Iterator
+    from unittest.mock import MagicMock
+    from unittest.mock import _Call
+
+    import httpx
+    from pytest_mock import MockerFixture
+
+_REAL_GETADDRINFO = socket.getaddrinfo
+_REAL_AGETADDRINFO = anyio.getaddrinfo
+
+
+@dataclass
+class ReceivedRequest:
+    method: str
+    path: str
+    headers: dict[str, str]
+    body: bytes
+
+
+@dataclass
+class LocalHTTPServer:
+    """State of a threaded HTTP server bound to 127.0.0.1 on an ephemeral port."""
+
+    port: int
+    requests: list[ReceivedRequest] = field(default_factory=list)
+    connections: int = 0
+    redirect_to: str | None = None
+
+
+class _Handler(http.server.BaseHTTPRequestHandler):
+    # HTTP/1.1 keeps connections open, so tests can observe connection reuse.
+    # Every response sets Content-Length, which keep-alive requires.
+    protocol_version = "HTTP/1.1"
+
+    def _handle(self) -> None:
+        length = int(self.headers.get("Content-Length") or 0)
+        body = self.rfile.read(length) if length else b""
+        # BaseHTTPRequestHandler types server as the base socketserver.BaseServer;
+        # narrowing the attribute's declared type is a variance error, so the
+        # subclass is recovered here instead of on the class body.
+        server = cast("_RecordingHTTPServer", self.server)
+        state = server.state
+        state.requests.append(
+            ReceivedRequest(
+                method=self.command,
+                path=self.path,
+                headers={key.lower(): value for key, value in self.headers.items()},
+                body=body,
+            ),
+        )
+        if state.redirect_to is not None:
+            self.send_response(302)
+            self.send_header("Location", state.redirect_to)
+            self.send_header("Content-Length", "0")
+            self.end_headers()
+            return
+        self.send_response(200)
+        self.send_header("Content-Length", "2")
+        self.end_headers()
+        self.wfile.write(b"ok")
+
+    do_GET = _handle
+    do_POST = _handle
+
+    def log_message(self, format: str, *args: Any) -> None:
+        return None
+
+
+class _RecordingHTTPServer(http.server.ThreadingHTTPServer):
+    daemon_threads = True
+
+    def __init__(self) -> None:
+        super().__init__(("127.0.0.1", 0), _Handler)
+        self.state = LocalHTTPServer(port=self.socket.getsockname()[1])
+
+    def verify_request(self, request: Any, client_address: Any) -> bool:
+        self.state.connections += 1
+        return True
+
+
+@contextmanager
+def running_http_server() -> Iterator[LocalHTTPServer]:
+    """Serve on 127.0.0.1 in a background thread until the block exits."""
+    server = _RecordingHTTPServer()
+    thread = threading.Thread(target=server.serve_forever, daemon=True)
+    thread.start()
+    try:
+        yield server.state
+    finally:
+        server.shutdown()
+        server.server_close()
+        thread.join(timeout=5)
+
+
+def _addrinfo(address: str, port: int | None) -> tuple[Any, ...]:
+    if ":" in address:
+        return (socket.AF_INET6, socket.SOCK_STREAM, 6, "", (address, port or 0, 0, 0))
+    return (socket.AF_INET, socket.SOCK_STREAM, 6, "", (address, port or 0))
+
+
+class FakeDNS:
+    """
+    Answers the guard's resolver hooks for registered names and delegates
+    every other name to the real resolver. The stock httpcore backends keep
+    using the unpatched socket.getaddrinfo.
+    """
+
+    def __init__(self) -> None:
+        self._answers: dict[str, list[str]] = {}
+        self.lookups: list[str] = []
+
+    def add(self, hostname: str, *addresses: str) -> None:
+        self._answers[hostname] = list(addresses)
+
+    def getaddrinfo(
+        self,
+        host: str,
+        port: int | None,
+        *args: Any,
+        **kwargs: Any,
+    ) -> list[tuple[Any, ...]]:
+        self.lookups.append(host)
+        if host in self._answers:
+            return [_addrinfo(address, port) for address in self._answers[host]]
+        return list(_REAL_GETADDRINFO(host, port, *args, **kwargs))
+
+    async def agetaddrinfo(
+        self,
+        host: str,
+        port: int | None,
+        **kwargs: Any,
+    ) -> list[tuple[Any, ...]]:
+        self.lookups.append(host)
+        if host in self._answers:
+            return [_addrinfo(address, port) for address in self._answers[host]]
+        return list(await _REAL_AGETADDRINFO(host, port, **kwargs))
+
+
+def install_fake_dns(mocker: MockerFixture) -> FakeDNS:
+    """Patch the guard's resolver hooks with a FakeDNS for the current test."""
+    dns = FakeDNS()
+    mocker.patch("paperless.network._getaddrinfo", new=dns.getaddrinfo)
+    mocker.patch("paperless.network._agetaddrinfo", new=dns.agetaddrinfo)
+    return dns
+
+
+def _dialled_host(call: _Call) -> str:
+    # The spy sits on the class, so args[0] is the backend instance.
+    if "host" in call.kwargs:
+        return str(call.kwargs["host"])
+    return str(call.args[1])
+
+
+@dataclass
+class DialRecorder:
+    sync_spy: MagicMock
+    async_spy: MagicMock
+
+    def hosts(self) -> list[str]:
+        calls = [*self.sync_spy.call_args_list, *self.async_spy.call_args_list]
+        return [_dialled_host(call) for call in calls]
+
+
+def install_dial_recorder(mocker: MockerFixture) -> DialRecorder:
+    """Spy on the stock backends' connect_tcp for the current test."""
+    return DialRecorder(
+        sync_spy=mocker.spy(httpcore.SyncBackend, "connect_tcp"),
+        async_spy=mocker.spy(httpcore.AnyIOBackend, "connect_tcp"),
+    )
+
+
+def allow_all_addresses(mocker: MockerFixture) -> None:
+    """Patch the guard's public-address check to accept every address.
+
+    Loopback and other private addresses pass just like a public one, for
+    tests that exercise something other than the address policy itself.
+    """
+    mocker.patch("paperless.network.is_public_ip", return_value=True)
+
+
+def guard_of(
+    client: httpx.Client | httpx.AsyncClient,
+) -> _GuardedSyncBackend | _GuardedAsyncBackend:
+    """Return the guard installed on a client's transport."""
+    transport = client._transport
+    assert isinstance(transport, GuardedHTTPTransport | GuardedAsyncHTTPTransport)
+    backend = transport._pool._network_backend
+    assert isinstance(backend, _GuardedSyncBackend | _GuardedAsyncBackend)
+    return backend
@@ -220,7 +220,7 @@ wheels = [

 [[package]]
 name = "autobahn"
-version = "26.7.1"
+version = "25.12.2"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "cbor2" },
@@ -228,35 +228,24 @@ dependencies = [
    { name = "cryptography" },
    { name = "hyperlink" },
    { name = "msgpack", marker = "platform_python_implementation == 'CPython'" },
+    { name = "py-ubjson" },
    { name = "txaio" },
    { name = "u-msgpack-python", marker = "platform_python_implementation != 'CPython'" },
    { name = "ujson" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/de/73/f109f563c27e048e45d135d81af19e6ca391e24905550b06bd1c9d674c57/autobahn-26.7.1.tar.gz", hash = "sha256:c6949a2c6eb95fb1c218837dbda0a59abbbebafb8b11098551c01a7061dfd245", size = 14056542, upload-time = "2026-07-15T19:14:01.246Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/54/d5/9adf0f5b9eb244e58e898e9f3db4b00c09835ef4b6c37d491886e0376b4f/autobahn-25.12.2.tar.gz", hash = "sha256:754c06a54753aeb7e8d10c5cbf03249ad9e2a1a32bca8be02865c6f00628a98c", size = 13893652, upload-time = "2025-12-15T11:13:19.086Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/44/8c/381cdcab8016df2177adc93d25f84ca3a5fb8f8be4f9d784336416c7bee8/autobahn-26.7.1-cp311-cp311-macosx_15_0_arm64.whl", hash = "sha256:3fe80550707f0affb5cb10f3e0f66ec7e6e52abb29edc66dd76734c2d7d51bf4", size = 1997747, upload-time = "2026-07-15T19:13:21.998Z" },
-    { url = "https://files.pythonhosted.org/packages/79/a9/9293c6c6bc8970f42c9675942de78f306e18eafa47edba52fd27f9dc71bd/autobahn-26.7.1-cp311-cp311-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:00fb9acd8775eaa0e272f36b76db903f10de56478f6a72f0bd07ee882ae1f2b8", size = 2082284, upload-time = "2026-07-15T19:13:23.582Z" },
-    { url = "https://files.pythonhosted.org/packages/a9/ba/7396cb42a9c59df20c350ea05f75e6f25f582b474dee82b8e32823b2711e/autobahn-26.7.1-cp311-cp311-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:97c9674eddd55ad3ebd733824789175e5fb90c88afd523de507569ba0fcd6853", size = 2254260, upload-time = "2026-07-15T19:13:24.894Z" },
-    { url = "https://files.pythonhosted.org/packages/b7/64/19753442770662ff45c4fe48db6345ac6fa3100fbbd989241c074e38ea6f/autobahn-26.7.1-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:30fa714de5c9903ef64084d3a938d8a3bac0bb42f1532d5de22f34b04a1c4819", size = 3173653, upload-time = "2026-07-15T19:13:26.261Z" },
-    { url = "https://files.pythonhosted.org/packages/04/a4/b690f272427acf1e8ea03b146e559dc67ade10dd4e0cccacc1d4c011b141/autobahn-26.7.1-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:c3362197f3b9d5b0df7f3365bd00dedba7ee8b649d652941377abe690d1c8b14", size = 3402880, upload-time = "2026-07-15T19:13:27.683Z" },
-    { url = "https://files.pythonhosted.org/packages/6e/23/0769ef39e1cfb0bec15bacdd7f407aaedfda14c0ca3f7e818b856f2ed1a1/autobahn-26.7.1-cp312-cp312-macosx_15_0_arm64.whl", hash = "sha256:6c9013e9aa9ea8a561c89d7be2709546b51fc7ac8fdf6cd71bc12a634672d9a8", size = 2000605, upload-time = "2026-07-15T19:13:30.391Z" },
-    { url = "https://files.pythonhosted.org/packages/f9/ef/26833f38ecf3aef3ff0aa09feb12f5d472f7370104148a6b78e3c7afc286/autobahn-26.7.1-cp312-cp312-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:abd9ebe577dd1030f9a0c41d20dc00eca90d5cf338531abcb510d978825feeea", size = 2082844, upload-time = "2026-07-15T19:13:31.559Z" },
-    { url = "https://files.pythonhosted.org/packages/b2/4c/00553ee9d57ee11df47bc9867d120cfe721a73bec0b58fb3a3b91cd7c797/autobahn-26.7.1-cp312-cp312-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:de491baa4cf52fb6d7f542e445c72d94ce52dc7aabb47b0b4d5191e452dd2c74", size = 2254852, upload-time = "2026-07-15T19:13:32.8Z" },
-    { url = "https://files.pythonhosted.org/packages/c3/ca/7884f6ffb8410882df98cb939dea24225dd79e4f091ceb59f4b826e54f2f/autobahn-26.7.1-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:9723561c820ed032fe5a8f1530cb6d5f91dc595eea6009f3b2de7044a3df892a", size = 3174235, upload-time = "2026-07-15T19:13:34.134Z" },
-    { url = "https://files.pythonhosted.org/packages/1f/e7/c6704e8f6bef3aa552a851a34908a06d91317d35ca55ec04b5db14385c33/autobahn-26.7.1-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:e81c86cf41adca8a56ca5621ecd9ba40037f6d90d338c334bd529c8ac94bd7b6", size = 3403687, upload-time = "2026-07-15T19:13:35.633Z" },
-    { url = "https://files.pythonhosted.org/packages/36/92/2f6e57d9f9e6b86b9db362f58aaa6cfeadc2f3a6901ec95aab27ef232b5c/autobahn-26.7.1-cp313-cp313-macosx_15_0_arm64.whl", hash = "sha256:2ce48214b28f73338fabe0c7fd13d222cfab9e1dd2ef11660293522f64e76727", size = 1987052, upload-time = "2026-07-15T19:13:38.543Z" },
-    { url = "https://files.pythonhosted.org/packages/38/6d/f170134468e276fa9ea57eb1ae41f9cc0dcd0228e9d501f370fa50c0ee31/autobahn-26.7.1-cp313-cp313-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:a5f285dce9b3dff3eb2ef6c818ac8ede24d96bd1edac340855170fc9825c38a9", size = 2082813, upload-time = "2026-07-15T19:13:39.686Z" },
-    { url = "https://files.pythonhosted.org/packages/0d/ce/b735fa933e9ba4fa8c3f9aa9ae68b4e2d4aadb0a92b38ade30e37a7d4795/autobahn-26.7.1-cp313-cp313-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:66ab6e034e54f8c473df1a6b8031a3db46c16deebaf0c9b66db3e9137d2fab5e", size = 2254818, upload-time = "2026-07-15T19:13:40.951Z" },
-    { url = "https://files.pythonhosted.org/packages/df/3e/57855f4f52aa0ee64c6d8210637d0d9847de4c1082e03ddd2ffafd853d2d/autobahn-26.7.1-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:20b3eab7d483e93278f9d7345592eb6b465883a6e88c2823aad13c3f943452db", size = 682494, upload-time = "2026-07-15T19:13:42.193Z" },
-    { url = "https://files.pythonhosted.org/packages/3b/3c/3944f17dd2a06aee7d0d9f1c37b5a94518434d13ba2c38e738d33ca10daf/autobahn-26.7.1-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:cd840524ff190aee695a58e8acd8740ee66b4a0e6a58ccb41416ed2cbe48d43f", size = 3403666, upload-time = "2026-07-15T19:13:43.508Z" },
-    { url = "https://files.pythonhosted.org/packages/51/3e/200471878093a502f8e8078c1ca19fd82acc68ae9ac363e395170da6dbe2/autobahn-26.7.1-cp314-cp314-macosx_15_0_arm64.whl", hash = "sha256:f3d1be925e3fb33fff5280c1bd02027047519812c400d3efa5477d3968686c94", size = 1987070, upload-time = "2026-07-15T19:13:47.112Z" },
-    { url = "https://files.pythonhosted.org/packages/61/d1/704f881fd2c52b056dc0f14e6d0d640b1f3ff43f3b84cf85d3631e3243f4/autobahn-26.7.1-cp314-cp314-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:44f8094b0c0fa29a4963ce12130b7932469f89afa0242ff858d2b26541a81005", size = 2082939, upload-time = "2026-07-15T19:13:48.569Z" },
-    { url = "https://files.pythonhosted.org/packages/a8/27/84e76aec7abbcb502d4cd34ef5c859eaa19a3b707cda68ab7ce68478dd92/autobahn-26.7.1-cp314-cp314-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:a0fcf9c3ff6b9b2bc85d6e1814a94d1d941d62f7df36d350b523d77df85d66ea", size = 2254983, upload-time = "2026-07-15T19:13:49.802Z" },
-    { url = "https://files.pythonhosted.org/packages/e6/81/a810732a10342c5d6b90d19f83fa2bc9b6126e7c0cda7c4df867e311aa2e/autobahn-26.7.1-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:4f82e5a113f6c1ff14cec99aa411f7da8fceec3dcd4647d7ebdfc0278811e14d", size = 3174295, upload-time = "2026-07-15T19:13:51.227Z" },
-    { url = "https://files.pythonhosted.org/packages/a6/c6/4886fdaecfeda013e085288a9d83bad6f1ded9995b8088ca933d9ec37201/autobahn-26.7.1-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:919309cbe41b28b0a3028e7c6fed52aca9fd21639619f372d3270c44341b6bdc", size = 3403713, upload-time = "2026-07-15T19:13:52.744Z" },
-    { url = "https://files.pythonhosted.org/packages/94/14/6485c29ad06a6bd7b3017558f99f6f89dcaa6ef6641930b25d9247adba4c/autobahn-26.7.1-pp311-pypy311_pp73-macosx_15_0_arm64.whl", hash = "sha256:9088acf790caf8cfd86590cb2b749279256ee210198f41d9858a38d1346e56c9", size = 1981962, upload-time = "2026-07-15T19:13:55.79Z" },
-    { url = "https://files.pythonhosted.org/packages/8d/46/cb6d09604417beacdf485b414a05efa18511b0e78ac5451b3655bee711fa/autobahn-26.7.1-pp311-pypy311_pp73-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:ea4548ee15c6bdf8aa0a1e81bf47b42db350f2bef69f83bace27a95ed0d21276", size = 655967, upload-time = "2026-07-15T19:13:57.19Z" },
-    { url = "https://files.pythonhosted.org/packages/00/d9/b846bc5a37f25ac147879d6451c466968a54efbf9d0467f0732f37c6f3f3/autobahn-26.7.1-pp311-pypy311_pp73-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:d4ee0fe13a5218831d60863becd8c3cf6558e6c5ccc5d9d0e722012bdb1459bf", size = 2207401, upload-time = "2026-07-15T19:13:58.392Z" },
+    { url = "https://files.pythonhosted.org/packages/ae/23/923e4f11dc9d12b9f5a014f36d591c479d623d54dda3bdcbd688cd12f052/autobahn-25.12.2-cp311-cp311-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:16df879672c60f1f3fe452138c80f0fd221b3cb2ee5a14390c80f33b994104c1", size = 2053413, upload-time = "2025-12-15T11:12:58.167Z" },
+    { url = "https://files.pythonhosted.org/packages/b3/0d/3d39637a1e32f555ce5fabec4a723a035556ef918b14140faea05e7de902/autobahn-25.12.2-cp311-cp311-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:6ffe28048ef96eb0f925f24c2569bd72332e120f4cb31cd6c40dd66718a5f85e", size = 2224850, upload-time = "2025-12-15T11:13:00.089Z" },
+    { url = "https://files.pythonhosted.org/packages/64/8d/36452c06cbcad6d04587aeb87dfa987ef94be4a427b9f2155783d166bd97/autobahn-25.12.2-cp312-cp312-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:220748f21e91bd4a538d2d3de640cc17ee30b79f1c04a6c3dcdef321d531ee1c", size = 2225453, upload-time = "2025-12-15T11:13:02.865Z" },
+    { url = "https://files.pythonhosted.org/packages/83/30/ef9c47038e4e9257319d6e1b87668b3df360a0c488d66ccff9d11aaff6ba/autobahn-25.12.2-cp313-cp313-macosx_15_0_arm64.whl", hash = "sha256:bc17f6cab9438156d2701c293c76fd02a144f9be0a992c065dfee1935ce4845b", size = 1960447, upload-time = "2025-12-15T11:13:05.007Z" },
+    { url = "https://files.pythonhosted.org/packages/e2/e4/f3d5cb70bc0b9b5523d940734b2e0a251510d051a50d2e723f321e890859/autobahn-25.12.2-cp313-cp313-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:b5297a782fc7d0a26842438ef1342549ceee29496cda52672ac44635c79eeb94", size = 2053955, upload-time = "2025-12-15T11:13:06.052Z" },
+    { url = "https://files.pythonhosted.org/packages/ea/49/4e592a19ae58fd9c796821a882b22598fac295ede50f899cc9d14a0282b6/autobahn-25.12.2-cp313-cp313-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:c0c3f1d5dafda52f8dc962ab583b6f3473b7b7186cab082d05372ed43a8261a5", size = 2225441, upload-time = "2025-12-15T11:13:07.527Z" },
+    { url = "https://files.pythonhosted.org/packages/54/b7/0a0e3ecb2af7e452f5f359d19bdc647cbc8658f3f498bfa3bf8545cf4768/autobahn-25.12.2-cp314-cp314-macosx_15_0_arm64.whl", hash = "sha256:c840ee136bfaf6560467160129b0b25a0e33c9a51e2b251e98c5474f27583915", size = 1960463, upload-time = "2025-12-15T11:13:10.183Z" },
+    { url = "https://files.pythonhosted.org/packages/19/8b/4215ac49d6b793b592fb08698f3a0e21a59eb3520be7f7ed288fcb52d919/autobahn-25.12.2-cp314-cp314-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:9abda5cf817c0f8a19a55a67a031adf2fc70ed351719b5bd9e6fa0f5f4bc8f89", size = 2225590, upload-time = "2025-12-15T11:13:11.367Z" },
+    { url = "https://files.pythonhosted.org/packages/d6/99/b4a3da42471d3ec36e2dca0c1a5368a079fed9f73b159ce3f049c4a4983b/autobahn-25.12.2-pp311-pypy311_pp73-macosx_15_0_arm64.whl", hash = "sha256:0c226329ddec154c6f3b491ea3e4713035f0326c96ebfd6b305bf90f27a2fba1", size = 1955357, upload-time = "2025-12-15T11:13:13.581Z" },
+    { url = "https://files.pythonhosted.org/packages/89/81/67f19dd7395a9f1123a1f071314f8d1c4879c1869adeb8d99a236e756ac0/autobahn-25.12.2-pp311-pypy311_pp73-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:5f079393a7626eb448c8accf21151f5f206d02f8e9cee4313d62a5ca30a3aaed", size = 623173, upload-time = "2025-12-15T11:13:14.945Z" },
+    { url = "https://files.pythonhosted.org/packages/71/eb/857eab3d25e3b9cc9e7e741d6193808ad91de0befb38cf10658bd339c205/autobahn-25.12.2-pp311-pypy311_pp73-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:9b3a6c7d54a9f0434a435d88b86555510e5d0a84aa87042e292f29f707cab237", size = 2178008, upload-time = "2025-12-15T11:13:15.881Z" },
 ]

 [[package]]
@@ -330,6 +319,18 @@ wheels = [
    { url = "https://files.pythonhosted.org/packages/cb/87/8bab77b323f16d67be364031220069f79159117dd5e43eeb4be2fef1ac9b/billiard-4.2.4-py3-none-any.whl", hash = "sha256:525b42bdec68d2b983347ac312f892db930858495db601b5836ac24e6477cde5", size = 87070, upload-time = "2025-11-30T13:28:47.016Z" },
 ]

+[[package]]
+name = "bleach"
+version = "6.4.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "webencodings" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/48/3c/e12ac860709702bd5ebeb9b56a4fe334f1001246ee1b8f2b7ee28912df7d/bleach-6.4.0.tar.gz", hash = "sha256:4202482733d85cedd04e59fcb2f89f4e4c7c385a78d3c3c23c30446843a37452", size = 204857, upload-time = "2026-06-05T13:01:13.734Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/58/9d/40b6267367182187139a4000b82a3b287d84d745bccd808e75d916920e9d/bleach-6.4.0-py3-none-any.whl", hash = "sha256:4b6b6a54fff2e69a3dde9d21cc6301220bee3c3cb792187d11403fd795031081", size = 165109, upload-time = "2026-06-05T13:01:12.504Z" },
+]
+
 [[package]]
 name = "brotli"
 version = "1.2.0"
@@ -2873,8 +2874,10 @@ name = "paperless-ngx"
 version = "3.2.1"
 source = { virtual = "." }
 dependencies = [
+    { name = "anyio" },
    { name = "azure-ai-documentintelligence" },
    { name = "babel" },
+    { name = "bleach" },
    { name = "celery", extra = ["redis"] },
    { name = "channels" },
    { name = "channels-redis" },
@@ -2900,6 +2903,8 @@ dependencies = [
    { name = "filelock" },
    { name = "flower" },
    { name = "gotenberg-client", extra = ["httpx"] },
+    { name = "httpcore" },
+    { name = "httpx" },
    { name = "httpx-oauth" },
    { name = "ijson" },
    { name = "imap-tools" },
@@ -2931,7 +2936,6 @@ dependencies = [
    { name = "tika-client", extra = ["httpx"] },
    { name = "torch", version = "2.14.0", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "sys_platform != 'linux'" },
    { name = "torch", version = "2.14.0+cpu", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "sys_platform == 'linux'" },
-    { name = "turbohtml" },
    { name = "watchfiles" },
    { name = "whitenoise" },
    { name = "whoosh-compat", extra = ["tantivy"] },
@@ -3010,6 +3014,7 @@ typing = [
    { name = "mypy" },
    { name = "mypy-baseline" },
    { name = "pyrefly" },
+    { name = "types-bleach" },
    { name = "types-channels" },
    { name = "types-colorama" },
    { name = "types-dateparser" },
@@ -3024,8 +3029,10 @@ typing = [

 [package.metadata]
 requires-dist = [
+    { name = "anyio", specifier = ">=4.12" },
    { name = "azure-ai-documentintelligence", specifier = ">=1.0.2" },
    { name = "babel", specifier = ">=2.17" },
+    { name = "bleach", specifier = "~=6.4.0" },
    { name = "celery", extras = ["redis"], specifier = "~=5.6.2" },
    { name = "channels", specifier = "~=4.2" },
    { name = "channels-redis", specifier = "~=4.2" },
@@ -3052,6 +3059,8 @@ requires-dist = [
    { name = "flower", specifier = ">=2.0.1,<2.2" },
    { name = "gotenberg-client", extras = ["httpx"], specifier = "~=1.0" },
    { name = "granian", extras = ["uvloop"], marker = "extra == 'webserver'", specifier = ">=2.7,<2.9" },
+    { name = "httpcore", specifier = "~=1.0.9" },
+    { name = "httpx", specifier = "~=0.28.1" },
    { name = "httpx-oauth", specifier = "~=0.17" },
    { name = "ijson", specifier = ">=3.5.1" },
    { name = "imap-tools", specifier = ">=1.14,<1.16" },
@@ -3090,7 +3099,6 @@ requires-dist = [
    { name = "tantivy", specifier = "~=0.26.0" },
    { name = "tika-client", extras = ["httpx"], specifier = "~=1.0" },
    { name = "torch", specifier = ">=2.13,<2.15", index = "https://download.pytorch.org/whl/cpu" },
-    { name = "turbohtml", specifier = "~=1.10.0" },
    { name = "watchfiles", specifier = ">=1.2" },
    { name = "whitenoise", specifier = "~=6.11" },
    { name = "whoosh-compat", extras = ["tantivy"], specifier = "==0.3" },
@@ -3151,6 +3159,7 @@ typing = [
    { name = "mypy" },
    { name = "mypy-baseline" },
    { name = "pyrefly" },
+    { name = "types-bleach" },
    { name = "types-channels" },
    { name = "types-colorama" },
    { name = "types-dateparser" },
@@ -3208,45 +3217,40 @@ wheels = [

 [[package]]
 name = "pikepdf"
-version = "10.13.0.post1"
+version = "10.2.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
+    { name = "deprecated" },
    { name = "lxml" },
    { name = "packaging" },
    { name = "pillow" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/1c/0e/6e74dd213537b71c945743a4b3112dbb430896ad68b8a6ad22e4468455d4/pikepdf-10.13.0.post1.tar.gz", hash = "sha256:4b73f926ebae81f04bf14527af330bd00bb268be767e0f189f7c4c3e4ad7ae0a", size = 4973186, upload-time = "2026-09-05T06:49:20.825Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/6e/e9/a1462d6160805ca80c8f4aafc941aaf410a92d0fcc683706e94f499c2fac/pikepdf-10.2.0.tar.gz", hash = "sha256:0f398b0daeb2ffd2358f75c06f1dd47b9ba76f1a77dfe938cccf7080c58227d7", size = 4568506, upload-time = "2026-01-09T22:54:25.847Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/4d/c1/48c9c0ed2ed88ca5d9cdd7f16075385a05a812d781b61bfa7f2d5182b247/pikepdf-10.13.0.post1-cp311-cp311-macosx_14_0_arm64.whl", hash = "sha256:98a7305e330f797da02b543d3ad57a134c4a14c6ec6f8d86d91aa9dd130c425b", size = 1846300, upload-time = "2026-09-05T06:48:11.563Z" },
-    { url = "https://files.pythonhosted.org/packages/e9/20/484a3a61664132dc8c4bd97e0b8291fa79f9f4a3b1e1ffd5b67ac41ed98a/pikepdf-10.13.0.post1-cp311-cp311-macosx_15_0_x86_64.whl", hash = "sha256:f3dedd02795626f17ee42d5c02ec4ec94e28aa47046ef430d4478454a8fbd07f", size = 1944179, upload-time = "2026-09-05T06:48:14.01Z" },
-    { url = "https://files.pythonhosted.org/packages/66/38/797df7d60352fc5ec3943c425acaa15d032cd2e673b516861f449536db57/pikepdf-10.13.0.post1-cp311-cp311-manylinux_2_26_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:7ba09ef5a5f26e38ee558d2a08223fee08a5ef1868962ae2d8590d4de3c8c92f", size = 2105054, upload-time = "2026-09-05T06:48:15.547Z" },
-    { url = "https://files.pythonhosted.org/packages/a8/8a/1f003558c5c05cecf182af775839ad674fe23e06b314d0219b9d8422680a/pikepdf-10.13.0.post1-cp311-cp311-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:365b94f2be7e2857c6cb5445b56dc52dc7417ba9f06c8a4282d9f521cb2d0fb8", size = 2308617, upload-time = "2026-09-05T06:48:17.216Z" },
-    { url = "https://files.pythonhosted.org/packages/55/5b/0e7193ee8c7ca5b15f478918033a0fa78917b01249e1d4a4a65754644cd3/pikepdf-10.13.0.post1-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:f515a31c76cce043bbb7b781e77a343a4e26fa8f520ba337d30ddea0f7a0ce50", size = 3742337, upload-time = "2026-09-05T06:48:18.971Z" },
-    { url = "https://files.pythonhosted.org/packages/b6/f5/519e8728c04d05dcbd44d03b266a3f6acf8de3a0a6ed5ec0fa39ececddd7/pikepdf-10.13.0.post1-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:b63577c44fedf7ed6b971076f7c7ed0ff95a8bac627ac93b79e8b542214861a7", size = 3952988, upload-time = "2026-09-05T06:48:20.984Z" },
-    { url = "https://files.pythonhosted.org/packages/6c/a5/598e72c72ed46046e297f15763dffda88424870724a4a22f599b815cb774/pikepdf-10.13.0.post1-cp312-cp312-macosx_14_0_arm64.whl", hash = "sha256:2c6e83f8a1828ec79cdec4df8cc07209eaf10ed7e4f5a90a7356b254bacc07d5", size = 1845515, upload-time = "2026-09-05T06:48:24.477Z" },
-    { url = "https://files.pythonhosted.org/packages/f9/b3/29691a5e9ee915357c081730d8cc02f35f19b4155857556fbe562a4d83ba/pikepdf-10.13.0.post1-cp312-cp312-macosx_15_0_x86_64.whl", hash = "sha256:f2463f650efab46905b9e279f5c776faf96c65bb45c1acf9f2de0e8a6eec5fb7", size = 1944755, upload-time = "2026-09-05T06:48:26.332Z" },
-    { url = "https://files.pythonhosted.org/packages/04/4e/201f553b9405424d7aefa3be997f0a1c787ac3a089a844f1fc42f412fe0e/pikepdf-10.13.0.post1-cp312-cp312-manylinux_2_26_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:9613505f5b22203465d4224a3fd8cf69876ce8278442c6478ac6849f54724a30", size = 2102202, upload-time = "2026-09-05T06:48:28.816Z" },
-    { url = "https://files.pythonhosted.org/packages/3e/a3/bd7e7b321e8bfe4b8530da57d12c557a259bbd4b40e739960a1f2ea507cb/pikepdf-10.13.0.post1-cp312-cp312-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:01f80ca046d984752cf6f08093debc193884bee91c01c104aa0236767711a20f", size = 2307194, upload-time = "2026-09-05T06:48:31.558Z" },
-    { url = "https://files.pythonhosted.org/packages/23/b0/ca630f56015dfc6c4c8c81fdeb5f5099e7fca354f54a41dfe7f1382a5316/pikepdf-10.13.0.post1-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:1f76fcbe5d86f2ae6f231ba542cd04793d4c89e75bd5f62526b7412926ff2100", size = 3740157, upload-time = "2026-09-05T06:48:33.285Z" },
-    { url = "https://files.pythonhosted.org/packages/be/48/7a84adc2fd14ec35e4b3d007575914de1ce0518e8c69b35ebee62fa2b142/pikepdf-10.13.0.post1-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:092a9bf15739e931ecab15ec3baee5d9629dad90ea4e42b779f6b439d2d1e462", size = 3951601, upload-time = "2026-09-05T06:48:35.468Z" },
-    { url = "https://files.pythonhosted.org/packages/10/f4/3636368760840cbc3ee512330024dd6f518d583c1bbbb1b551ca8e18f5e8/pikepdf-10.13.0.post1-cp313-cp313-macosx_14_0_arm64.whl", hash = "sha256:87141ada970386ff6640db54f0bda734d3bde7960d3ba04a76768b48e25028ca", size = 1845524, upload-time = "2026-09-05T06:48:39.309Z" },
-    { url = "https://files.pythonhosted.org/packages/ce/dc/7bbfba253a0394a237b81be371a64f904f99636d579ec78ef5d92025fd2d/pikepdf-10.13.0.post1-cp313-cp313-macosx_15_0_x86_64.whl", hash = "sha256:55e53b4d8a4b1700f686f76e3a68411e421e962a4c8b1b90d00aab3f3e494a55", size = 1944829, upload-time = "2026-09-05T06:48:41.601Z" },
-    { url = "https://files.pythonhosted.org/packages/b3/a3/10367bfb93501a151cbb96b6973f7c049fef04040aea35cf6cedf579c3e4/pikepdf-10.13.0.post1-cp313-cp313-manylinux_2_26_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:8fb8f82dc43056a4b4f891e78ee1db4e3ced75ba3e87b836f8e28c8771228928", size = 2102269, upload-time = "2026-09-05T06:48:43.222Z" },
-    { url = "https://files.pythonhosted.org/packages/2d/bd/a68b5d9b4aef4d4b9c374cfdfd941623f52303d31adea13554568df42abe/pikepdf-10.13.0.post1-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:d3b58ccb30b93ba400e6a6a83315b4830eea49f6f19e18d78241fe6b1c49fec2", size = 2307160, upload-time = "2026-09-05T06:48:45.192Z" },
-    { url = "https://files.pythonhosted.org/packages/9a/59/47bd86d9e338d301c28416d52d0e154a0d6322cf6b957c9df5f3d7828aa2/pikepdf-10.13.0.post1-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:6b038cd5bcbb6c1952bcc271695eaf24c4199d72e45606ee5e467f837760820e", size = 3739722, upload-time = "2026-09-05T06:48:47.105Z" },
-    { url = "https://files.pythonhosted.org/packages/49/40/87fb6dddc9dce110429c72449e942174fd500f6fc4c9ff518a6b73057aa7/pikepdf-10.13.0.post1-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:b7b0cbb135de32ec3f41651a08ab294e3c18ae9fec32516a48d27e64a53a47f0", size = 3951549, upload-time = "2026-09-05T06:48:49.263Z" },
-    { url = "https://files.pythonhosted.org/packages/14/0e/86897bf5325824c1f2d9d8be89839baf11011b5392d92620cb0273fb9af5/pikepdf-10.13.0.post1-cp314-abi3-macosx_14_0_arm64.whl", hash = "sha256:51fae4a4a3c6549aa4c405896ff7010f3e43e0c4f407c0bcee071ef13d271202", size = 1845245, upload-time = "2026-09-05T06:48:53.575Z" },
-    { url = "https://files.pythonhosted.org/packages/f0/ed/923846b7511627f8564d09345e083f14674dfb193de92d27f1cc4650602e/pikepdf-10.13.0.post1-cp314-abi3-macosx_15_0_x86_64.whl", hash = "sha256:8cb976331cb8b03ec3465e06d9e7a3eadbadb7e622be888e70f918ac732a105e", size = 1943659, upload-time = "2026-09-05T06:48:55.823Z" },
-    { url = "https://files.pythonhosted.org/packages/58/bb/fcb09ad4bd227bbb37a7e5b24de86f9ce9d462aa0c7ee18781899bfa378a/pikepdf-10.13.0.post1-cp314-abi3-manylinux_2_26_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:3e18d5a009bbe5f3ab18f916fb9e28f0c5b0d920736e3a85cc10c627ec633596", size = 2099423, upload-time = "2026-09-05T06:48:57.671Z" },
-    { url = "https://files.pythonhosted.org/packages/3b/c9/707f9ba96727fa366a650237e46f0e20a73b244592eb6b97a49e401e2b43/pikepdf-10.13.0.post1-cp314-abi3-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:414f42c83e5e6029870de1a988625dafc95781ebff10e15d82caeb5e69a83c9c", size = 2303539, upload-time = "2026-09-05T06:48:59.506Z" },
-    { url = "https://files.pythonhosted.org/packages/22/15/79a2ccc354514a1321a161867a011fc917be158fc01a88da8c78ad18399d/pikepdf-10.13.0.post1-cp314-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:f0eb89f06cad9231b9db54d81a22592b03b63924824a6b850febd2b75daa6546", size = 3737772, upload-time = "2026-09-05T06:49:01.245Z" },
-    { url = "https://files.pythonhosted.org/packages/6e/85/a17440c2de64da71dc012b42e644b30ee4d98eb540c14d4e9f3538b73a9e/pikepdf-10.13.0.post1-cp314-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:544f1be1b1e5630a79cd182a8663c439504099eb9eae0d342a50173d9825bdf3", size = 3948342, upload-time = "2026-09-05T06:49:03.27Z" },
-    { url = "https://files.pythonhosted.org/packages/e9/65/15a796a3cf9fb17d41acc1ab6719e7d3dcdf1260e76909d0dd32ecc97ba7/pikepdf-10.13.0.post1-cp314-cp314t-macosx_14_0_arm64.whl", hash = "sha256:ef4ed47d40aa44deb063feb4a88e8bcf1c8fa0183ce526dc4295f7cbdb1292f8", size = 1853638, upload-time = "2026-09-05T06:49:07.184Z" },
-    { url = "https://files.pythonhosted.org/packages/aa/f3/5d49a511fd13b59b94c5ad673695d331fce5d2846ab1501646c2a3b35b5f/pikepdf-10.13.0.post1-cp314-cp314t-macosx_15_0_x86_64.whl", hash = "sha256:571efcd1d54e0dd817973c76c253feb6fb758c93bb0c16a893cc68f2a178d404", size = 1951768, upload-time = "2026-09-05T06:49:09.048Z" },
-    { url = "https://files.pythonhosted.org/packages/06/de/f6bbd9653695f6e2ed3494a439f11f3a89450c004fe9bf8ee583201ea759/pikepdf-10.13.0.post1-cp314-cp314t-manylinux_2_26_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:2db9a18074ba112e7c517e8c21dfd8894cc13b8a37ccf49a5841192170fb68eb", size = 2107137, upload-time = "2026-09-05T06:49:10.788Z" },
-    { url = "https://files.pythonhosted.org/packages/c0/aa/43b355681f05ea0b5808a26cba9e8e764686fa8dcebe0875db612664de40/pikepdf-10.13.0.post1-cp314-cp314t-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:f7962a75cf22d0d683b49ab19b8966e94dc7014d9aa6806f3c0d2b37fb9ae607", size = 2310991, upload-time = "2026-09-05T06:49:12.455Z" },
-    { url = "https://files.pythonhosted.org/packages/a9/46/77574e9c4bded01afd7a3fe538f5432e396c3772c9bc1eab5d287aed00df/pikepdf-10.13.0.post1-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:b948e11f7dd3710f939194f00b4d75b0df87a30d53b31414128768ac25da77d8", size = 3744358, upload-time = "2026-09-05T06:49:14.428Z" },
-    { url = "https://files.pythonhosted.org/packages/89/a8/4857df72cf4773553c2e6a82f93ee5e98c02f1c4e4877379e84d27384982/pikepdf-10.13.0.post1-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:99f6afccd6119233e7133bd4c2ade48461de3ddd4269cc28a4f90cdf7c1372f5", size = 3955964, upload-time = "2026-09-05T06:49:16.656Z" },
+    { url = "https://files.pythonhosted.org/packages/10/dc/aa7293763b603a9080ffab7ab87c7b571d637a389e9fb2ba839b864ca283/pikepdf-10.2.0-cp311-cp311-macosx_14_0_arm64.whl", hash = "sha256:fb93732127d5183a91300af39e1cda5ded309e8439daec93536331a472b5e190", size = 4727891, upload-time = "2026-01-09T22:53:26.262Z" },
+    { url = "https://files.pythonhosted.org/packages/d2/dc/700c31f2c14f94d92483b10e1918390948ed20f6f572d82beb78ac5f94d0/pikepdf-10.2.0-cp311-cp311-macosx_15_0_x86_64.whl", hash = "sha256:ab7bd4629539cf2136a799dc3eaa2dfda59937035a97b0c5e22a7a3a4033cc49", size = 5030510, upload-time = "2026-01-09T22:53:28.023Z" },
+    { url = "https://files.pythonhosted.org/packages/23/46/dc63364b05aa1913f2d7480cad62676bfb473065ba4b02d314dfd482f7dd/pikepdf-10.2.0-cp311-cp311-manylinux_2_26_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:2f5623a5ba456d69dfeb86dc3bb3ec31ec1d120382d8c24804d1b430fce715ea", size = 2439498, upload-time = "2026-01-09T22:53:30.122Z" },
+    { url = "https://files.pythonhosted.org/packages/59/b6/1f9b8ca588fd34d9e3df49a80c62016e0b42ce6e580146c46d9728fdb6e8/pikepdf-10.2.0-cp311-cp311-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:dec4d12f294df378d122ae441c27c1e76fb0d15b1e9d7374ae70c26604559bab", size = 2666945, upload-time = "2026-01-09T22:53:32.309Z" },
+    { url = "https://files.pythonhosted.org/packages/c1/31/b1e61fac59f0b807edde655a821ff83bb041ae1500234c52ed1a2403c44a/pikepdf-10.2.0-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:c0cebe3235232f1bd3c5f7956218ce92241c94223cb80eba837d372a40c61765", size = 3638109, upload-time = "2026-01-09T22:53:34.144Z" },
+    { url = "https://files.pythonhosted.org/packages/02/e9/a99bbf503c9d55e54553edff84ec67cac49d335fc33f3d5516c4746b6340/pikepdf-10.2.0-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:d0908e845c9140e245ad89a19fdfc6e5a6d82fcb505b8cc2c0ce81439ac4f064", size = 3829538, upload-time = "2026-01-09T22:53:36.341Z" },
+    { url = "https://files.pythonhosted.org/packages/73/18/598383493a0f0f0c4eecd09b8fe06dddb9d326a89e2623a134d43e051485/pikepdf-10.2.0-cp312-cp312-macosx_14_0_arm64.whl", hash = "sha256:18c35d00baff72bfae82d67028bedb02ea2b208e1af5545c23cd681f2487a279", size = 4737716, upload-time = "2026-01-09T22:53:39.973Z" },
+    { url = "https://files.pythonhosted.org/packages/a9/f5/bec04784ba07d44f03b52ea524bcb7409bf7185ee8abec7ae29e3ac9e9ae/pikepdf-10.2.0-cp312-cp312-macosx_15_0_x86_64.whl", hash = "sha256:dd849d033b95de15965c095ebc4d78983099a11bb7b7897801dfaf3cb4083a35", size = 5042152, upload-time = "2026-01-09T22:53:41.808Z" },
+    { url = "https://files.pythonhosted.org/packages/38/3e/148b3c8e101c8ac3a33f41e86c5739413575495e471bce45ee228aafcbd6/pikepdf-10.2.0-cp312-cp312-manylinux_2_26_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:e9910efdc7907af3da9e7b2a125a1f67d512165ffa623f62825deeb642669a7a", size = 2445796, upload-time = "2026-01-09T22:53:44.027Z" },
+    { url = "https://files.pythonhosted.org/packages/c2/ef/b06f8fd68c34fed631cb8e3520dd955e59987de0eee6960dbc94bed11711/pikepdf-10.2.0-cp312-cp312-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:f0ec947e6429d7a3306153d32a0142462fdd8f905c5fe08c8a8e8c53b9c28a5c", size = 2693908, upload-time = "2026-01-09T22:53:46.039Z" },
+    { url = "https://files.pythonhosted.org/packages/3e/bf/e5c40e9210e2ae8da7cad2cf6ae7d1db3b63a2916e6040645958e9ab4054/pikepdf-10.2.0-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:b6383219a1cd31400403a69737a4e2a0c5d2a2c4cb9f380bcf45e33e8de802ea", size = 3643423, upload-time = "2026-01-09T22:53:48.333Z" },
+    { url = "https://files.pythonhosted.org/packages/d4/1b/969dfb29dc9fd7b82fa7bc065df498e8a3e7ddb81e982140634ee539a8db/pikepdf-10.2.0-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:46d2f9ef5a84949bfc11152a323558f94cf85d9d97e9c510c061c7f803028f3f", size = 3854816, upload-time = "2026-01-09T22:53:50.158Z" },
+    { url = "https://files.pythonhosted.org/packages/f4/c5/e6f9e3407dd73ec570000a64747ff84e2f57b06b0477d1da6eaca5038162/pikepdf-10.2.0-cp313-cp313-macosx_14_0_arm64.whl", hash = "sha256:09ff28d1de7fc7711a7ef8dfc40396d9243b64ee24c37cd1ab2a9f9827895caa", size = 4737680, upload-time = "2026-01-09T22:53:54.905Z" },
+    { url = "https://files.pythonhosted.org/packages/eb/de/dffb785235ac2d930db86b215c1848d7258e625fa1949dd0633f8b72ab0a/pikepdf-10.2.0-cp313-cp313-macosx_15_0_x86_64.whl", hash = "sha256:62348b66e1401a4db0c64976b72dd74bb1a9eb3a33007a661500f4f8a64436bd", size = 5042150, upload-time = "2026-01-09T22:53:57.722Z" },
+    { url = "https://files.pythonhosted.org/packages/1a/f0/4d883f57304d98650ade30a8c73fe593582d9afd9a7dada1f5f3f4cce362/pikepdf-10.2.0-cp313-cp313-manylinux_2_26_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:6b9e91780cb9ea3c6a350ffbcf03d5d95c30084d238afbd1d4b927cdb9e3649d", size = 2445446, upload-time = "2026-01-09T22:53:59.993Z" },
+    { url = "https://files.pythonhosted.org/packages/62/65/ffe2555812a152d616accacea7c1c617c27a75590379ea7d9cc3a26bd92d/pikepdf-10.2.0-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:52360a49a22e9353ec9a08ff5713cec8aacaf3ef960c704bc0a89ca8f050bdad", size = 2696242, upload-time = "2026-01-09T22:54:01.687Z" },
+    { url = "https://files.pythonhosted.org/packages/ed/8c/2f937b0e2867cd48b523122e08753571fc9847978e239d7b5db9bd46879c/pikepdf-10.2.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:4ba4c1046939eb22c24c396deb37f8e0500caaa66b73114be55377d5554b4167", size = 3643730, upload-time = "2026-01-09T22:54:04.177Z" },
+    { url = "https://files.pythonhosted.org/packages/2e/17/f2919e4085c399e938bb945ea712dea70b3849e17cae6403f0cc1100e9ef/pikepdf-10.2.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:3dcd8957a08e0a47f7a138904dca8cf73962fa17a096a47cd8bc33eb83a4f0a7", size = 3856645, upload-time = "2026-01-09T22:54:07.887Z" },
+    { url = "https://files.pythonhosted.org/packages/1d/6e/846902abe8286d3b4ab70893e9ffbeec99aadd93ba1536cf471b222bb910/pikepdf-10.2.0-cp314-cp314-macosx_14_0_arm64.whl", hash = "sha256:194c9a81ecb49e425a5cd5162621270b5e42cf05709d87eac018bd6f9ce98f80", size = 4733930, upload-time = "2026-01-09T22:54:11.931Z" },
+    { url = "https://files.pythonhosted.org/packages/8b/6d/abdbb794d2a512d4e828ef2014cc47ca263ad3fbd1b65f25f791b9c0bb1e/pikepdf-10.2.0-cp314-cp314-macosx_15_0_x86_64.whl", hash = "sha256:5adcf87dbfff4e1cd0a850db487274f474c94a6bf6347f3842c53da8d0eaa8df", size = 5042477, upload-time = "2026-01-09T22:54:13.708Z" },
+    { url = "https://files.pythonhosted.org/packages/52/6c/6c42694fe1574a37aa2a40b4ba29a6713b4226436155ef7aa0bef649c117/pikepdf-10.2.0-cp314-cp314-manylinux_2_26_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:5de3cecbb35c4bc651e9326932974217be1d450d4a9840d77a592062eb507e27", size = 2448419, upload-time = "2026-01-09T22:54:15.6Z" },
+    { url = "https://files.pythonhosted.org/packages/45/f4/aca3286aa37ace581afc8e3e0644a0cc55b9f9ceb31f28219d12ca11536c/pikepdf-10.2.0-cp314-cp314-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:77868fd25182a45a4f3dec3c461aea8c696ef9565894c5cde4394bc8c32fb069", size = 2697600, upload-time = "2026-01-09T22:54:18.123Z" },
+    { url = "https://files.pythonhosted.org/packages/eb/a6/9135f9f0189634de61410573a0712d849e0157e3902e6b867339cc7dbf1b/pikepdf-10.2.0-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:a9a10e15e2f4d0bba36a2b4328342d00eff1a5a31399e1d1a93483c70d3c2b0e", size = 3647720, upload-time = "2026-01-09T22:54:20.158Z" },
+    { url = "https://files.pythonhosted.org/packages/83/60/f282077773a3321fad4cbfb16fe73ee3f8dd93b408df65c24779f12227c5/pikepdf-10.2.0-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:a8f80ecf00fb15a760f218432a1046e7797cd14eaa6ccb52c8814ae8852745d8", size = 3859133, upload-time = "2026-01-09T22:54:22.358Z" },
 ]

 [[package]]
@@ -3606,6 +3610,12 @@ wheels = [
    { url = "https://files.pythonhosted.org/packages/37/ed/89c2c620af0e1660354cd8aabf9f5b21f911597ce22acb37c805d6c86bc8/psycopg_pool-3.3.1-py3-none-any.whl", hash = "sha256:2af5b432941c4c9ad5c87b3fa410aec910ec8f7c122855897983a06c45f2e4b5", size = 40023, upload-time = "2026-05-01T23:31:53.136Z" },
 ]

+[[package]]
+name = "py-ubjson"
+version = "0.16.1"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/1d/c7/28220d37e041fe1df03e857fe48f768dcd30cd151480bf6f00da8713214a/py-ubjson-0.16.1.tar.gz", hash = "sha256:b9bfb8695a1c7e3632e800fb83c943bf67ed45ddd87cd0344851610c69a5a482", size = 50316, upload-time = "2020-04-18T15:05:57.698Z" }
+
 [[package]]
 name = "pyasn1"
 version = "0.6.4"
@@ -5165,52 +5175,6 @@ wheels = [
    { url = "https://files.pythonhosted.org/packages/29/47/54eacf96b5c835bbd6ca631aa2740e7705ed63d9e3a8afd2d2cc6d09cae5/transformers-5.13.1-py3-none-any.whl", hash = "sha256:53f0ea8aa397e29244c2377ba981bcaf0c87adcf44fbdd447ef6306522afcacd", size = 11503977, upload-time = "2026-07-11T09:15:46.801Z" },
 ]

-[[package]]
-name = "turbohtml"
-version = "1.10.0"
-source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/30/0b/94e323028f3aa652a5fb924681ed9c17537921716ac8b8267f03078c617f/turbohtml-1.10.0.tar.gz", hash = "sha256:a5f59f1f74b6b9060bc98751b1c7f68f5e74fd11e6252559eb838f948c8fc76d", size = 3497676, upload-time = "2026-09-24T21:58:12.43Z" }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/cb/13/a92f25cdfc64e333f3248fbf4ced0e65364be60279b90806670a8e78d0d9/turbohtml-1.10.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:e61c74a0bab25f00ac000f7a4aac00db9834938dbda3dc364307a1f11f6a60c0", size = 1435697, upload-time = "2026-09-24T21:58:12.27Z" },
-    { url = "https://files.pythonhosted.org/packages/52/7b/3577e58777d837317dd57c65140f3a4e52cc5a1fac99a30ba9dc5c0b26c3/turbohtml-1.10.0-cp311-cp311-manylinux_2_28_aarch64.whl", hash = "sha256:8ccb3c078e5300079e5b2a5cdd9bab6f866dad4818d4508679d71a8ad05f8a81", size = 1569438, upload-time = "2026-09-24T21:58:36.289Z" },
-    { url = "https://files.pythonhosted.org/packages/24/e1/b55930f6cb69ccb0ccb3877ff67c7ced8dc01cf2782a8c53264bbaaea532/turbohtml-1.10.0-cp311-cp311-manylinux_2_28_x86_64.whl", hash = "sha256:26502bcbea0926c9c9e468cfcf50381a5820a48e0c16b472271fe584abefd0f7", size = 1617559, upload-time = "2026-09-24T21:58:38.531Z" },
-    { url = "https://files.pythonhosted.org/packages/21/ec/a2aef28eb3dfceecc387b406aff28ee3c409ec6723b7c72d001b1a093708/turbohtml-1.10.0-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:8363bc65c1671a013b8d7a3b3e5e54fa72d6b65630c11a08eac6a83556c64602", size = 1583921, upload-time = "2026-09-24T21:58:40.004Z" },
-    { url = "https://files.pythonhosted.org/packages/7c/43/87fb508686714638ad8419dca027ca6e22977fc42dd66a98d2b4f2eefe90/turbohtml-1.10.0-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:84f40701d0a52a5b581f31e86223e2d40f311c9bc9d89264c730101ccd1f14a1", size = 1615781, upload-time = "2026-09-24T21:58:42.441Z" },
-    { url = "https://files.pythonhosted.org/packages/c0/a8/2156eeb17eff5ee5d5f3f90cc634593010cd8282fb813025d6489301102b/turbohtml-1.10.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:9addb768f79f93658030a128344b5458b242fa39cfa8611bc6e2cda3241e1deb", size = 1437483, upload-time = "2026-09-24T21:58:13.919Z" },
-    { url = "https://files.pythonhosted.org/packages/81/87/bbe8a0750c32e99bcefff6a6d98129599ebad68670013281bceeb748fc1a/turbohtml-1.10.0-cp312-cp312-manylinux_2_28_aarch64.whl", hash = "sha256:423f589598a09bad4d4230ab8667363fccd0ae81cdd0e7bf7b02ad39d5923761", size = 1571298, upload-time = "2026-09-24T21:58:44Z" },
-    { url = "https://files.pythonhosted.org/packages/07/97/8759f26fb77e66ca76408798a390c078f76cba30b3fef895b83716394eb0/turbohtml-1.10.0-cp312-cp312-manylinux_2_28_x86_64.whl", hash = "sha256:4030eb81946491bb6d1a039a8beb8cc9455af6449d77a25db98b471929edb58b", size = 1619769, upload-time = "2026-09-24T21:58:47.551Z" },
-    { url = "https://files.pythonhosted.org/packages/c8/9b/2dd565c20e7a962b26f2f1feb2a7513682551c4cbf7619480c3faa49999c/turbohtml-1.10.0-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:288b41d8e73466d0353f780acf30910ff39f82357d95752f9a85b0f09a71048f", size = 1586146, upload-time = "2026-09-24T21:58:51.198Z" },
-    { url = "https://files.pythonhosted.org/packages/39/25/174a1ba40cb738e8d06bc59d1884e95a955e16f02d9e81f020b55306c490/turbohtml-1.10.0-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:a13ea122aaff4b78e1f7042f104f133ee88b8f48bf0c4cbf9dafd8f73ac47722", size = 1622290, upload-time = "2026-09-24T21:58:50.444Z" },
-    { url = "https://files.pythonhosted.org/packages/fb/c2/d41f69814a0a217db1fd184da90ac3573094859b37aa8b73558fc2e1ac4a/turbohtml-1.10.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:b9b6171d883daf85bb6374aeae6f7c19c489e3fc61e0643a7f2f1da9822a0679", size = 1437198, upload-time = "2026-09-24T21:58:15.536Z" },
-    { url = "https://files.pythonhosted.org/packages/86/fe/99913911376f99ff34a18bef342db8c662eee9965205f0345f64f46cfdf0/turbohtml-1.10.0-cp313-cp313-manylinux_2_28_aarch64.whl", hash = "sha256:662e6b0650aaf30e090cca967dbc81fd4f1f3045be481a5e5e2715bf52578ef9", size = 1572058, upload-time = "2026-09-24T21:58:53.885Z" },
-    { url = "https://files.pythonhosted.org/packages/a2/e0/2d293264e67078ee6b08f1cb7c93dd71da945e243d65f69672e14f4497f8/turbohtml-1.10.0-cp313-cp313-manylinux_2_28_x86_64.whl", hash = "sha256:70380c063355c98b5ce120d8f201c7def00143e1bb6563ededc60f62807f0fe8", size = 1619317, upload-time = "2026-09-24T21:58:53.274Z" },
-    { url = "https://files.pythonhosted.org/packages/ff/49/20046efdcbb99c1c8cbe7445cee5a12f53141d500056973e1379b981bcb9/turbohtml-1.10.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:bb5a788781d79099676bba5706973a65b75df9ed51ee3d069ef648a3dd0d6804", size = 1584857, upload-time = "2026-09-24T21:58:57.189Z" },
-    { url = "https://files.pythonhosted.org/packages/96/49/281e656c86a056adc97c0d99273ed92bc9b17dfed090520515856d0e9b59/turbohtml-1.10.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:225613d470dfb580d5346c7446fdcbfa13ccb2f4da2dd8d19000103250e625ab", size = 1620185, upload-time = "2026-09-24T21:58:56.516Z" },
-    { url = "https://files.pythonhosted.org/packages/54/c2/93b519997efdc353753945f053317b493648071b5b1be65f36e09f37bf61/turbohtml-1.10.0-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:fab89d0ce289c7636cb825fa170c4b52b4083a5eda963576caff4cffad7bf8f9", size = 1437862, upload-time = "2026-09-24T21:58:17.11Z" },
-    { url = "https://files.pythonhosted.org/packages/6a/8e/ff1443ccdf61791103ff6b0ed669059769bd4e97e536f337fce72b9dda3a/turbohtml-1.10.0-cp314-cp314-manylinux_2_28_aarch64.whl", hash = "sha256:794f77b1dedcfb9eabfb4e17a1c0fafe047f54b7e2b6f3b920c45e9a36b5a885", size = 1572493, upload-time = "2026-09-24T21:58:59.198Z" },
-    { url = "https://files.pythonhosted.org/packages/7e/8d/4917f89a01410ecfe66c785163f4c6e6f16cf1ce19de7cc8b9193fbf0c7a/turbohtml-1.10.0-cp314-cp314-manylinux_2_28_x86_64.whl", hash = "sha256:e85b2d554e5a45dd81a43d8e47996b445d8ab4aaaa777c894026a2cbc184a93e", size = 1620012, upload-time = "2026-09-24T21:58:58.449Z" },
-    { url = "https://files.pythonhosted.org/packages/40/c2/5b07960f1429a7f66448d0631807113c0b6c1d7e54aeef043f3e06d3f552/turbohtml-1.10.0-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:3a0351e57de782a763a7a8c885fce8943e22b6facf72b0ecc95d5b44a681d3ab", size = 1587372, upload-time = "2026-09-24T21:59:01.314Z" },
-    { url = "https://files.pythonhosted.org/packages/e2/8f/2a17171aeb3004f1c35aa0fdec0091e9f7b80282a7dbc2a647de69b6b094/turbohtml-1.10.0-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:be0ef006fb1b6099efe6d6591f09ae57d5bccc5faf2e6163ee6e3880424ef7f2", size = 1621259, upload-time = "2026-09-24T21:59:00.594Z" },
-    { url = "https://files.pythonhosted.org/packages/0f/fb/bdc5a2fa1334f7878bbd72662445ca4f065af923229846786cca6fc317de/turbohtml-1.10.0-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:421f56edc437913379764ec58057d4f637e972e0b273ca808a091220c3cf7168", size = 1457281, upload-time = "2026-09-24T21:58:19.055Z" },
-    { url = "https://files.pythonhosted.org/packages/ca/47/e8a585f1f0f05d20a6616aa73174bf8c9b65f5bf7c21b384ff854c480459/turbohtml-1.10.0-cp314-cp314t-manylinux_2_28_aarch64.whl", hash = "sha256:89c7d933c6e06b5977110680cb0ad01f8c1e7ee579c0539215831943ff65b4fd", size = 1564047, upload-time = "2026-09-24T21:59:04.097Z" },
-    { url = "https://files.pythonhosted.org/packages/43/65/f881a937aef5020186f4a9795c13669e540471668019603c9a04157632b8/turbohtml-1.10.0-cp314-cp314t-manylinux_2_28_x86_64.whl", hash = "sha256:9b7164630c67d0ba241efe95651190b48de3b21d0d28f09c81405d6b6e4e3e3b", size = 1610056, upload-time = "2026-09-24T21:59:03.215Z" },
-    { url = "https://files.pythonhosted.org/packages/d8/62/bb2654d5f826b8fccd3309b0d350724cc8ed77e2e0703d9a016182f0aed4/turbohtml-1.10.0-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:92175e361a7e79b357d069a180452c5fdab3e83dbc8fe82af147ffd850efcb51", size = 1585075, upload-time = "2026-09-24T21:59:06.118Z" },
-    { url = "https://files.pythonhosted.org/packages/dc/7e/8e2315ef3d96e0f57cc0881e7160373e824d895890e7757b682818d7122e/turbohtml-1.10.0-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:a17ee91740811506c9d02b3c136d5676e6a328e243f026014bb813b484a5886c", size = 1625498, upload-time = "2026-09-24T21:59:05.501Z" },
-    { url = "https://files.pythonhosted.org/packages/e0/5b/fef20b454ceaea7594550fbdd95377b0f7062cf34b7be45f52f85266c89b/turbohtml-1.10.0-cp315-cp315-macosx_11_0_arm64.whl", hash = "sha256:8a89b5996149b43ff9b83a9db8609044ccd557c9bf109d2ee4b36076e1ba6102", size = 1437864, upload-time = "2026-09-24T21:58:29.029Z" },
-    { url = "https://files.pythonhosted.org/packages/71/20/229e187d0e47eac34958313bdb4746feedb5f33c17f90da00573c82d8ec6/turbohtml-1.10.0-cp315-cp315-manylinux_2_28_aarch64.whl", hash = "sha256:4f385ad6e14e50d59964b177e20348461eccda9decd59935f721cc3fccd5a238", size = 1574234, upload-time = "2026-09-24T21:59:07.897Z" },
-    { url = "https://files.pythonhosted.org/packages/dd/e3/1882db0fdc8760d172ef19c11d8d973e339f775f450ba20568efb6bb7e24/turbohtml-1.10.0-cp315-cp315-manylinux_2_28_x86_64.whl", hash = "sha256:01b9ef79d882e16402cefc19f78c0226f8b6f83c32588b8fe71733cd4bca83f1", size = 1618454, upload-time = "2026-09-24T21:59:07.158Z" },
-    { url = "https://files.pythonhosted.org/packages/f8/53/ecd6314b50bc3683219be261259ec2e0585a301d910fdf4afa6b52b7af66/turbohtml-1.10.0-cp315-cp315-musllinux_1_2_aarch64.whl", hash = "sha256:97bdde66856c1be748fd8cdb4c2a7840f578af7fcacd2e455bba390304b4cb0f", size = 1586289, upload-time = "2026-09-24T21:59:09.822Z" },
-    { url = "https://files.pythonhosted.org/packages/c7/f8/264e99df411e1d8b0034cd606fac1145800b42547d8d44b508fa6857fcd0/turbohtml-1.10.0-cp315-cp315-musllinux_1_2_x86_64.whl", hash = "sha256:ffe72bff195a479e4811fb704d0fdaa33717893ebcf246f337d91c96e49eb8e7", size = 1621608, upload-time = "2026-09-24T21:59:09.011Z" },
-    { url = "https://files.pythonhosted.org/packages/a1/4d/d4b5ab6fe192c56db95de48389a5fc6aa857295df2c282c01977bb97c0e5/turbohtml-1.10.0-cp315-cp315t-macosx_11_0_arm64.whl", hash = "sha256:702cc0c00754eca0531a7fdb804d98e36cdb79192bebd0a174016acc00e6dbe8", size = 1457298, upload-time = "2026-09-24T21:58:33.144Z" },
-    { url = "https://files.pythonhosted.org/packages/11/28/2c8eeea4ba04ab333a7c5911c52464952439785318d4781678281b50b176/turbohtml-1.10.0-cp315-cp315t-manylinux_2_28_aarch64.whl", hash = "sha256:5d45a01b10fd5eb4117e959f52349a621e7a154ba2d24ecbe25fb9c979abd9af", size = 1563687, upload-time = "2026-09-24T21:59:11.654Z" },
-    { url = "https://files.pythonhosted.org/packages/7a/a9/399f4525643b7c4205359f26908487de1cbf8b520ff8ff38646342f01cab/turbohtml-1.10.0-cp315-cp315t-manylinux_2_28_x86_64.whl", hash = "sha256:08064d33db70a04231ef70cdf88fc038e778f9c6ef6c08e543244b2f7c40a156", size = 1610603, upload-time = "2026-09-24T21:59:10.984Z" },
-    { url = "https://files.pythonhosted.org/packages/09/3d/91ac95593de11942e9021cd2d6a7219b6afc345118d4e8d71fae9b88b0b0/turbohtml-1.10.0-cp315-cp315t-musllinux_1_2_aarch64.whl", hash = "sha256:19528d9ac387ebd89c8083adf77c194d1fc553c01a777fc1158db437b1ee23b7", size = 1583422, upload-time = "2026-09-24T21:59:13.758Z" },
-    { url = "https://files.pythonhosted.org/packages/33/b5/47a7b9abecd16df67addadb4ea01cc0cdb608b480234756b2a793731c0fa/turbohtml-1.10.0-cp315-cp315t-musllinux_1_2_x86_64.whl", hash = "sha256:8806189da106986e3d3df695eca250dc935fcfafc077bd3ae7e111cd99b29498", size = 1624867, upload-time = "2026-09-24T21:59:13Z" },
-    { url = "https://files.pythonhosted.org/packages/ac/a6/6c0e7950d84306a470d523d3ec4cb6aee7942ed75056aa4fb3f0174d20f1/turbohtml-1.10.0-pp311-pypy311_pp73-macosx_11_0_arm64.whl", hash = "sha256:149e80b94b5a28b4799158306e7c8da2fca2cd55600374bb5c4d48403410d962", size = 1439722, upload-time = "2026-09-24T21:58:40.978Z" },
-    { url = "https://files.pythonhosted.org/packages/ce/26/21425ccc18a0085cbc0f3ab7b9f0372273934ae3614b405a2a666eca4e3c/turbohtml-1.10.0-pp311-pypy311_pp73-manylinux_2_28_aarch64.whl", hash = "sha256:ab81284db98aa9a75e125dc88baf9b08ccd6ae628951435b31dccd65270a4e51", size = 1611998, upload-time = "2026-09-24T21:59:17.911Z" },
-    { url = "https://files.pythonhosted.org/packages/0c/cf/ab65ae59b3c31add2cd4cddd6473720c5f6113fc9b3099fa80a119e9b9eb/turbohtml-1.10.0-pp311-pypy311_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:47b2ec25547e5461cd9bef807701443588790c39472d0400d60f74ad6d2ba12f", size = 1664674, upload-time = "2026-09-24T21:59:16.917Z" },
-]
-
 [[package]]
 name = "twisted"
 version = "26.4.0"
@@ -5259,6 +5223,18 @@ wheels = [
    { url = "https://files.pythonhosted.org/packages/80/87/b9fd69c92c6102a066e1b86a35243f53e70bd4c709f2a26d9f4fee4f4dc0/typer-0.26.8-py3-none-any.whl", hash = "sha256:3512ca79ac5c11113414b36e80281b872884477722440691c89d1112e321a49c", size = 122564, upload-time = "2026-06-26T09:22:44.72Z" },
 ]

+[[package]]
+name = "types-bleach"
+version = "6.4.0.20260607"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "types-html5lib" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/39/e7/07438d4d0bd58c69e79d0d8aa0320a2005b08d10cac209dd812b3609227c/types_bleach-6.4.0.20260607.tar.gz", hash = "sha256:62ff099b9d25fbcccb1402651c14a465a5f54a24138be987442d2eed38aca91c", size = 11686, upload-time = "2026-06-07T06:11:40.035Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/6c/93/44da16cfa99a3f47fcb35e6b7396c94c3530e2028d2b1027cc4ce6d95d8d/types_bleach-6.4.0.20260607-py3-none-any.whl", hash = "sha256:fc07e2fd0ef9bc0a22026e8abe589d76c5d06ea885d70357263a4ff8c3dc8516", size = 11985, upload-time = "2026-06-07T06:11:38.996Z" },
+]
+
 [[package]]
 name = "types-cffi"
 version = "1.17.0.20250915"
@@ -5302,6 +5278,18 @@ wheels = [
    { url = "https://files.pythonhosted.org/packages/04/5a/4fc8ea9c86e10465e820f920bc4983137c356b75b12b81574bb9b4d42605/types_dateparser-1.4.3.20260907-py3-none-any.whl", hash = "sha256:344c979fe7c0633f0a4f5098d7b26402a4969806ef7fc22f902d303f061593e1", size = 25729, upload-time = "2026-09-07T06:43:18.074Z" },
 ]

+[[package]]
+name = "types-html5lib"
+version = "1.1.11.20251117"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "types-webencodings" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/c8/f3/d9a1bbba7b42b5558a3f9fe017d967f5338cf8108d35991d9b15fdea3e0d/types_html5lib-1.1.11.20251117.tar.gz", hash = "sha256:1a6a3ac5394aa12bf547fae5d5eff91dceec46b6d07c4367d9b39a37f42f201a", size = 18100, upload-time = "2025-11-17T03:08:00.78Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/f0/ab/f5606db367c1f57f7400d3cb3bead6665ee2509621439af1b29c35ef6f9e/types_html5lib-1.1.11.20251117-py3-none-any.whl", hash = "sha256:2a3fc935de788a4d2659f4535002a421e05bea5e172b649d33232e99d4272d08", size = 24302, upload-time = "2025-11-17T03:07:59.996Z" },
+]
+
 [[package]]
 name = "types-markdown"
 version = "3.10.2.20260712"
@@ -5391,6 +5379,15 @@ wheels = [
    { url = "https://files.pythonhosted.org/packages/2b/7f/016dc5cc718ec6ccaa84fb73ed409ef1c261793fd5e637cdfaa18beb40a9/types_setuptools-80.10.0.20260124-py3-none-any.whl", hash = "sha256:efed7e044f01adb9c2806c7a8e1b6aa3656b8e382379b53d5f26ee3db24d4c01", size = 64333, upload-time = "2026-01-24T03:18:38.344Z" },
 ]

+[[package]]
+name = "types-webencodings"
+version = "0.5.0.20251108"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/66/d6/75e381959a2706644f02f7527d264de3216cf6ed333f98eff95954d78e07/types_webencodings-0.5.0.20251108.tar.gz", hash = "sha256:2378e2ceccced3d41bb5e21387586e7b5305e11519fc6b0659c629f23b2e5de4", size = 7470, upload-time = "2025-11-08T02:56:00.132Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/45/4e/8fcf33e193ce4af03c19d0e08483cf5f0838e883f800909c6bc61cb361be/types_webencodings-0.5.0.20251108-py3-none-any.whl", hash = "sha256:e21f81ff750795faffddaffd70a3d8bfff77d006f22c27e393eb7812586249d8", size = 8715, upload-time = "2025-11-08T02:55:59.456Z" },
+]
+
 [[package]]
 name = "typing-extensions"
 version = "4.15.0"
@@ -5705,6 +5702,15 @@ wheels = [
    { url = "https://files.pythonhosted.org/packages/f2/3e/45583b67c2ff08ad5a582d316fcb2f11d6cf0a50c7707ac09d212d25bc98/wcwidth-0.5.0-py3-none-any.whl", hash = "sha256:1efe1361b83b0ff7877b81ba57c8562c99cf812158b778988ce17ec061095695", size = 93772, upload-time = "2026-01-27T01:31:43.432Z" },
 ]

+[[package]]
+name = "webencodings"
+version = "0.5.1"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/0b/02/ae6ceac1baeda530866a85075641cec12989bd8d31af6d5ab4a3e8c92f47/webencodings-0.5.1.tar.gz", hash = "sha256:b36a1c245f2d304965eb4e0a82848379241dc04b865afcc4aab16748587e1923", size = 9721, upload-time = "2017-04-05T20:21:34.189Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/f4/24/2a3e3df732393fed8b3ebf2ec078f05546de641fe1b667ee316ec1dcf3b7/webencodings-0.5.1-py2.py3-none-any.whl", hash = "sha256:a0af1213f3c2226497a97e2b3aa01a7e4bee4f403f95be16fc9acd2947514a78", size = 11774, upload-time = "2017-04-05T20:21:32.581Z" },
+]
+
 [[package]]
 name = "whitenoise"
 version = "6.12.0"
Author	SHA1	Message	Date
stumpylogandClaude Opus 5	296ddff37e	Collapse near-duplicate outbound guard tests into parametrized cases Several tests in test_network.py and test_network_integration.py had grown into near-copies of each other, differing only in the URL or host they fed the guard while asserting the same thing. That made the suite noisier to read and to extend than the behaviour it covers warrants. Merge the invalid-host rejection tests into one parametrized test over all seven unusual host forms, and likewise for their allow_internal counterparts that assert nothing is rejected. Merge the resolve_public_addresses tests that only check the host reaches the resolver unchanged, and the ones that check a private answer blocks regardless of how the host was spelled, keeping the stricter resolver call assertion on both merged cases. Add localhost as a fourth case to the numeric-host-forms test in the transport integration tests, so the name and the non-canonical numeric spellings of loopback are covered by one test. No behaviour or assertion strength changes; this only reduces duplicated test bodies while keeping every original case addressable by its own parametrize id. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 13:49:35 -07:00
stumpylogandClaude Opus 5	de8edc31c5	Fold the outbound guard's public-address patch into a named fixture Six tests in test_network_integration.py and one in test_workflows.py each patched paperless.network.is_public_ip inline to make every resolved address count as public, so the guard's own address policy would not get in the way of whatever the test was actually checking. The line was repeated verbatim at every call site and understated what it did: read literally it makes every address public, while each docstring around it already explained the intent as loopback being treated as public. Add an every_address_is_public fixture to conftest.py, following the existing pattern for the other outbound test fixtures: a thin wrapper that imports a new allow_all_addresses helper from paperless_testing.outbound and calls it with mocker. Apply it at each call site with pytest.mark.usefixtures, dropping the mocker parameter from tests that no longer need it directly. The test that patches is_public_ip with a selective side effect to allow only one address is left untouched, since that selectivity is load-bearing for what it verifies. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 12:56:55 -07:00
stumpylogandClaude Opus 5	46c150eb67	Deduplicate the outbound connect guard's sync and async paths The two connect_tcp implementations each repeated the per-attempt budget arithmetic verbatim, and only the sync one delegated resolution and error mapping to a helper, so a drift between the copies could silently give one stack a different timeout policy from the other. In the IMAP client, _connect_pinned re-asserted a fact its only caller had already established, which reads as a runtime invariant check on a security-relevant path when it is only a narrowing aid. Move the budget arithmetic into one helper called from both loops, add an async twin of the resolve helper so the two loop bodies differ only by await, and pass the narrowed address tuple into _connect_pinned instead of asserting it. The PinnedIMAP4 docstring now spells out that no pinning and pinning that yielded nothing are different things, and the monotonic clock seam says why it exists. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 12:49:06 -07:00
stumpylogandClaude Opus 5	58777076c5	Cover webhooks to internal hosts when they are allowed Every webhook test that reaches a real socket ran with internal requests disallowed, and the rest intercept above the transport. Wiring the transport to always disallow internal addresses would therefore have passed the suite while breaking webhooks to internal hosts on every default install. A new test sends a webhook to localhost with internal requests allowed and checks that the payload arrives and that the guard resolved nothing. The Host header test's docstring is also corrected: the header now comes from the URL, not from a resolved hostname. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 12:49:06 -07:00
stumpylogandClaude Opus 5	d414297c05	Fix: Validate IP literals from the resolver's answer only Outbound host checks parsed IP literals themselves and skipped the resolver for them. That second parser is what let a host such as 8.8.8.8%2eexample pass as the public address 8.8.8.8, and even with zone ids limited to IPv6 it remains one more place where the checked host can be read differently from the connected one. The separate literal parsing is removed. Every host now goes to getaddrinfo, which answers numeric literals itself without a lookup, and only the addresses it returns are classified. Zone ids are still dropped from the resolver's answers, where a scoped IPv6 literal comes back as fe80::1%1. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 12:49:06 -07:00
stumpylogandClaude Opus 5	d53a930aed	Fix: Stop treating a dotted quad followed by "%" as an IP literal The literal check stripped everything after the first "%" from any host, so a name such as 8.8.8.8%2e169-254-169-254.sslip.io was accepted as the public address 8.8.8.8 without a DNS lookup. requests percent-decodes the host before connecting, so Remote OCR would then resolve 8.8.8.8.169-254-169-254.sslip.io and reach an internal address even with internal endpoints disallowed. A zone id is now stripped only when the part before "%" is an IPv6 address; any other host containing "%" is looked up as a name. URL validation with internal addresses disallowed also rejects a host that contains "%" at all, since the name checked there could otherwise differ from the one an HTTP client decodes and dials. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 12:49:05 -07:00
stumpylogandClaude Opus 5	78ca920771	Document outbound connection policy for internal-address settings The internal-address settings now describe that a hostname is blocked if any resolved address is non-public. Guarded requests connect directly without proxy variables, and webhook requests never follow redirects. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 12:49:05 -07:00
stumpylogandClaude Opus 5	4c5255a8dd	Remove the URL-rewriting pinned transport Every consumer now uses the guarded transports, so the request-rewriting transport and its helpers are no longer needed. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 12:49:05 -07:00
stumpylogandClaude Opus 5	283e49b1ee	Match get_mailbox annotations to what callers actually pass get_mailbox declared port and security as always-present types, but MailAccount.imap_port is nullable and imap_security is stored as a plain integer, so type checking flagged every call site as passing the wrong type. Widen the annotations to port: int \| None and security: int, matching the model fields; IntegerChoices members still compare equal to plain ints, so the existing branching in get_mailbox is unaffected. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 12:49:05 -07:00
stumpylogandClaude Opus 5	3130fc3a7c	Use shared outbound resolution for IMAP host pinning get_mailbox validates the IMAP host with resolve_public_addresses and keeps its existing error messages; the pinned client dials typed addresses. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 12:49:05 -07:00
stumpylogandClaude Opus 5	8f56c6167b	Use the guarded transport for workflow webhooks A blocked webhook now raises OutboundRequestBlockedError, which the task treats as an expected failure and does not retry. The webhook security tests run against a real local server instead of a patched resolver. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 12:48:44 -07:00
stumpylogandClaude Opus 5	eb644f3fac	Report outbound policy blocks from AI requests as 502 AIClient raises LLMBlockedError when a request was refused by the outbound connection policy, including when the openai SDK wraps the block in APIConnectionError. ai_suggestions answers 502 instead of a 500. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 12:48:44 -07:00
stumpylogandClaude Opus 5	6ad00ca55a	Use guarded transports for AI LLM and embedding clients Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 12:48:21 -07:00
stumpylogandClaude Opus 5	526adbad1a	Fix two weak assertions in the outbound guard integration tests The environment proxy test could not fail for any regression in the guard: httpx ignores environment proxy variables whenever a transport is passed explicitly, and the address the test pointed the proxy at was itself internal, so a blocked direct connection was indistinguishable from a blocked proxied one. It now builds the client through the production factory, which does not pass a transport, points the proxy variables at a second recording server, and asserts the real origin server receives the request while the proxy server sees no connection at all. Several tests asserting a blocked request never reached the target server checked only a connection counter that is incremented after accept() in the server thread, which does not rule out a guard that connects and then fails validation afterward. Each of those tests now also asserts the dial recorder saw no address dialled. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 12:48:21 -07:00
stumpylogandClaude Opus 5	d53453fb71	Add real-socket tests for the outbound connection guard A local HTTP server, a per-hostname resolver fake and dial spies exercise the guarded transports end to end: address fallback, blocking before any connection, the Host header, TLS server name, per-host connection pooling, numeric host spellings, environment proxies and redirects to blocked hosts. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 12:48:20 -07:00
stumpylogandClaude Opus 5	4eaf8032ad	Add guarded httpx transports and client factories The transports install the outbound guard on httpcore's connection pool after checking its exact layout, and accept no proxy, uds or retries options. httpx, httpcore and anyio become declared dependencies, pinned narrowly where private attributes are relied on. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 12:47:40 -07:00
stumpylogandClaude Opus 5	9e1f938d56	Guard outbound connections in the httpcore network backend With internal addresses disallowed, the backend resolves the origin host, rejects it if any address is non-public, and dials the validated literals in family-interleaved order under the caller's connect timeout. Resolver failures surface as connect errors; unix sockets are always refused. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 12:47:40 -07:00
stumpylogandClaude Opus 5	0da50ad348	Reject outbound URLs that HTTP clients may split differently urllib3 treats a backslash as the end of the URL authority, while urlparse and httpx do not. For a URL such as http://127.0.0.1\@evil.example/ the check resolved evil.example while urllib3 would connect to 127.0.0.1, so a redirect to such a URL could reach an internal host. When internal addresses are disallowed, validate_outbound_http_url now rejects any URL containing a backslash, an ASCII control character or whitespace before resolving it. URLs validated with internal addresses allowed are unaffected. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 12:47:40 -07:00
stumpylogandClaude Opus 5	5e971bc0ce	Resolve outbound hosts to validated public addresses resolve_public_addresses and its async twin return every resolved address in resolver order, de-duplicated and zone-stripped, and reject the whole name if any address is non-public. validate_outbound_http_url uses them and keeps its existing messages. validate_outbound_http_url now resolves the hostname as httpx and urllib3 encode it (IDNA 2008). It previously let getaddrinfo apply the stdlib IDNA 2003 codec, which encodes characters such as "ß" differently, so a URL could pass the check under one DNS name and be connected to under another. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 12:47:40 -07:00
stumpylogandClaude Opus 5	17a91c385d	Type outbound block errors and classify addresses with is_global is_public_ip now takes an ipaddress object and relies on is_global, keeping multicast and the NAT64 well-known prefix as explicit extra exclusions. Adds OutboundRequestBlockedError and HostResolutionError, both picklable so Celery keeps them intact on task failure. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-22 12:47:40 -07:00