✨(backend) create ai endpoint

We created 2 new action endpoints on the document
to perform AI operations:
- POST /api/v1.0/documents/{uuid}/ai-transform
- POST /api/v1.0/documents/{uuid}/ai-translate

env.d/development/common.dist

@@ -39,3 +39,8 @@ LOGOUT_REDIRECT_URL=http://localhost:3000
 
 OIDC_REDIRECT_ALLOWED_HOSTS=["http://localhost:8083", "http://localhost:3000"]
 OIDC_AUTH_REQUEST_EXTRA_PARAMS={"acr_values": "eidas1"}
+
+# AI
+AI_BASE_URL=https://openaiendpoint.com
+AI_API_KEY=password
+AI_MODEL=llama

src/backend/core/api/serializers.py

@@ -9,7 +9,8 @@ from django.utils.translation import gettext_lazy as _
 import magic
 from rest_framework import exceptions, serializers
 
-from core import models
+from core import enums, models
+from core.services.ai_services import AI_ACTIONS
 
 
 class UserSerializer(serializers.ModelSerializer):
@@ -378,3 +379,33 @@ class VersionFilterSerializer(serializers.Serializer):
     page_size = serializers.IntegerField(
         required=False, min_value=1, max_value=50, default=20
     )
+
+
+class AITransformSerializer(serializers.Serializer):
+    """Serializer for AI transform requests."""
+
+    action = serializers.ChoiceField(choices=AI_ACTIONS, required=True)
+    text = serializers.CharField(required=True)
+
+    def validate_text(self, value):
+        """Ensure the text field is not empty."""
+
+        if len(value.strip()) == 0:
+            raise serializers.ValidationError("Text field cannot be empty.")
+        return value
+
+
+class AITranslateSerializer(serializers.Serializer):
+    """Serializer for AI translate requests."""
+
+    language = serializers.ChoiceField(
+        choices=tuple(enums.ALL_LANGUAGES.items()), required=True
+    )
+    text = serializers.CharField(required=True)
+
+    def validate_text(self, value):
+        """Ensure the text field is not empty."""
+
+        if len(value.strip()) == 0:
+            raise serializers.ValidationError("Text field cannot be empty.")
+        return value

src/backend/core/api/utils.py

@@ -1,8 +1,14 @@
 """Util to generate S3 authorization headers for object storage access control"""
 
+import time
+from abc import ABC, abstractmethod
+
+from django.conf import settings
+from django.core.cache import cache
 from django.core.files.storage import default_storage
 
 import botocore
+from rest_framework.throttling import BaseThrottle
 
 
 def generate_s3_authorization_headers(key):
@@ -31,3 +37,93 @@ def generate_s3_authorization_headers(key):
     auth.add_auth(request)
 
     return request
+
+
+class AIBaseRateThrottle(BaseThrottle, ABC):
+    """Base throttle class for AI-related rate limiting with backoff."""
+
+    def __init__(self, rates):
+        """Initialize instance attributes with configurable rates."""
+        super().__init__()
+        self.rates = rates
+        self.cache_key = None
+        self.recent_requests_minute = 0
+        self.recent_requests_hour = 0
+        self.recent_requests_day = 0
+
+    @abstractmethod
+    def get_cache_key(self, request, view):
+        """Abstract method to generate cache key for throttling."""
+
+    def allow_request(self, request, view):
+        """Check if the request is allowed based on rate limits."""
+        self.cache_key = self.get_cache_key(request, view)
+        if not self.cache_key:
+            return True  # Allow if no cache key is generated
+
+        now = time.time()
+        history = cache.get(self.cache_key, [])
+        # Keep requests within the last 24 hours
+        history = [req for req in history if req > now - 86400]
+
+        # Calculate recent requests
+        self.recent_requests_minute = len([req for req in history if req > now - 60])
+        self.recent_requests_hour = len([req for req in history if req > now - 3600])
+        self.recent_requests_day = len(history)
+
+        # Check rate limits
+        if self.recent_requests_minute >= self.rates["minute"]:
+            return False
+        if self.recent_requests_hour >= self.rates["hour"]:
+            return False
+        if self.recent_requests_day >= self.rates["day"]:
+            return False
+
+        # Log the request
+        history.append(now)
+        cache.set(self.cache_key, history, timeout=86400)
+        return True
+
+    def wait(self):
+        """Implement a backoff strategy by increasing wait time based on limits hit."""
+        if self.recent_requests_day >= self.rates["day"]:
+            return 86400
+        if self.recent_requests_hour >= self.rates["hour"]:
+            return 3600
+        if self.recent_requests_minute >= self.rates["minute"]:
+            return 60
+        return None
+
+
+class AIDocumentRateThrottle(AIBaseRateThrottle):
+    """Throttle for limiting AI requests per document with backoff."""
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(settings.AI_DOCUMENT_RATE_THROTTLE_RATES)
+
+    def get_cache_key(self, request, view):
+        """Include document ID in the cache key."""
+        document_id = view.kwargs["pk"]
+        return f"document_{document_id}_throttle_ai"
+
+
+class AIUserRateThrottle(AIBaseRateThrottle):
+    """Throttle that limits requests per user or IP with backoff and rate limits."""
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(settings.AI_USER_RATE_THROTTLE_RATES)
+
+    def get_cache_key(self, request, view=None):
+        """Generate a cache key based on the user ID or IP for anonymous users."""
+        if request.user.is_authenticated:
+            return f"user_{request.user.id!s}_throttle_ai"
+        return f"anonymous_{self.get_ident(request)}_throttle_ai"
+
+    def get_ident(self, request):
+        """Return the request IP address."""
+        x_forwarded_for = request.META.get("HTTP_X_FORWARDED_FOR")
+        return (
+            x_forwarded_for.split(",")[0]
+            if x_forwarded_for
+            else request.META.get("REMOTE_ADDR")
+        )

src/backend/core/api/viewsets.py

@@ -21,6 +21,7 @@ from rest_framework import (
     decorators,
     exceptions,
     filters,
+    metadata,
     mixins,
     pagination,
     status,
@@ -30,7 +31,8 @@ from rest_framework import (
     response as drf_response,
 )
 
-from core import models
+from core import enums, models
+from core.services.ai_services import AIService
 
 from . import permissions, serializers, utils
 
@@ -302,6 +304,23 @@ class ResourceAccessViewsetMixin:
         serializer.save()
 
 
+class DocumentMetadata(metadata.SimpleMetadata):
+    """Custom metadata class to add information"""
+
+    def determine_metadata(self, request, view):
+        """Add language choices only for the list endpoint."""
+        simple_metadata = super().determine_metadata(request, view)
+
+        if request.path.endswith("/documents/"):
+            simple_metadata["actions"]["POST"]["language"] = {
+                "choices": [
+                    {"value": code, "display_name": name}
+                    for code, name in enums.ALL_LANGUAGES.items()
+                ]
+            }
+        return simple_metadata
+
+
 class DocumentViewSet(
     ResourceViewsetMixin,
     mixins.CreateModelMixin,
@@ -319,6 +338,7 @@ class DocumentViewSet(
     resource_field_name = "document"
     queryset = models.Document.objects.all()
     ordering = ["-updated_at"]
+    metadata_class = DocumentMetadata
 
     def list(self, request, *args, **kwargs):
         """Restrict resources returned by the list endpoint"""
@@ -455,10 +475,7 @@ class DocumentViewSet(
         serializer = serializers.LinkDocumentSerializer(
             document, data=request.data, partial=True
         )
-        if not serializer.is_valid():
+        serializer.is_valid(raise_exception=True)
-            return drf_response.Response(
-                serializer.errors, status=status.HTTP_400_BAD_REQUEST
-            )
 
         serializer.save()
         return drf_response.Response(serializer.data, status=status.HTTP_200_OK)
@@ -471,10 +488,7 @@ class DocumentViewSet(
 
         # Validate metadata in payload
         serializer = serializers.FileUploadSerializer(data=request.data)
-        if not serializer.is_valid():
+        serializer.is_valid(raise_exception=True)
-            return drf_response.Response(
-                serializer.errors, status=status.HTTP_400_BAD_REQUEST
-            )
 
         # Generate a generic yet unique filename to store the image in object storage
         file_id = uuid.uuid4()
@@ -482,13 +496,13 @@ class DocumentViewSet(
         key = f"{document.key_base}/{ATTACHMENTS_FOLDER:s}/{file_id!s}.{extension:s}"
 
         # Prepare metadata for storage
-        metadata = {"Metadata": {"owner": str(request.user.id)}}
+        extra_args = {"Metadata": {"owner": str(request.user.id)}}
         if serializer.validated_data["is_unsafe"]:
-            metadata["Metadata"]["is_unsafe"] = "true"
+            extra_args["Metadata"]["is_unsafe"] = "true"
 
         file = serializer.validated_data["file"]
         default_storage.connection.meta.client.upload_fileobj(
-            file, default_storage.bucket_name, key, ExtraArgs=metadata
+            file, default_storage.bucket_name, key, ExtraArgs=extra_args
         )
 
         return drf_response.Response(
@@ -537,6 +551,63 @@ class DocumentViewSet(
         request = utils.generate_s3_authorization_headers(f"{pk:s}/{attachment_key:s}")
         return drf_response.Response("authorized", headers=request.headers, status=200)
 
+    @decorators.action(
+        detail=True,
+        methods=["post"],
+        name="Apply a transformation action on a piece of text with AI",
+        url_path="ai-transform",
+        throttle_classes=[utils.AIDocumentRateThrottle, utils.AIUserRateThrottle],
+    )
+    def ai_transform(self, request, *args, **kwargs):
+        """
+        POST /api/v1.0/documents/<resource_id>/ai-transform
+        with expected data:
+        - text: str
+        - action: str [prompt, correct, rephrase, summarize]
+        Return JSON response with the processed text.
+        """
+        # Check permissions first
+        self.get_object()
+
+        serializer = serializers.AITransformSerializer(data=request.data)
+        serializer.is_valid(raise_exception=True)
+
+        text = serializer.validated_data["text"]
+        action = serializer.validated_data["action"]
+
+        response = AIService().transform(text, action)
+
+        return drf_response.Response(response, status=status.HTTP_200_OK)
+
+    @decorators.action(
+        detail=True,
+        methods=["post"],
+        name="Translate a piece of text with AI",
+        serializer_class=serializers.AITranslateSerializer,
+        url_path="ai-translate",
+        throttle_classes=[utils.AIDocumentRateThrottle, utils.AIUserRateThrottle],
+    )
+    def ai_translate(self, request, *args, **kwargs):
+        """
+        POST /api/v1.0/documents/<resource_id>/ai-translate
+        with expected data:
+        - text: str
+        - language: str [settings.LANGUAGES]
+        Return JSON response with the translated text.
+        """
+        # Check permissions first
+        self.get_object()
+
+        serializer = self.get_serializer(data=request.data)
+        serializer.is_valid(raise_exception=True)
+
+        text = serializer.validated_data["text"]
+        language = serializer.validated_data["language"]
+
+        response = AIService().translate(text, language)
+
+        return drf_response.Response(response, status=status.HTTP_200_OK)
+
 
 class DocumentAccessViewSet(
     ResourceAccessViewsetMixin,

src/backend/core/enums.py

@@ -2,15 +2,11 @@
 Core application enums declaration
 """
 
-from django.conf import global_settings, settings
+from django.conf import global_settings
 from django.utils.translation import gettext_lazy as _
 
-# Django sets `LANGUAGES` by default with all supported languages. We can use it for
+# In Django's code base, `LANGUAGES` is set by default with all supported languages.
-# the choice of languages which should not be limited to the few languages active in
+# We can use it for the choice of languages which should not be limited to the few languages
-# the app.
+# active in the app.
 # pylint: disable=no-member
-ALL_LANGUAGES = getattr(
+ALL_LANGUAGES = {language: _(name) for language, name in global_settings.LANGUAGES}
-    settings,
-    "ALL_LANGUAGES",
-    [(language, _(name)) for language, name in global_settings.LANGUAGES],
-)

src/backend/core/models.py

@@ -508,6 +508,8 @@ class Document(BaseModel):
         can_get = bool(roles)
 
         return {
+            "ai_transform": is_owner_or_admin or is_editor,
+            "ai_translate": is_owner_or_admin or is_editor,
             "attachment_upload": is_owner_or_admin or is_editor,
             "destroy": RoleChoices.OWNER in roles,
             "link_configuration": is_owner_or_admin,

src/backend/core/services/ai_services.py

@@ -0,0 +1,89 @@
+"""AI services."""
+
+import json
+import re
+
+from django.conf import settings
+from django.core.exceptions import ImproperlyConfigured
+
+from openai import OpenAI
+
+from core import enums
+
+AI_ACTIONS = {
+    "prompt": (
+        "Answer the prompt in markdown format. Return JSON: "
+        '{"answer": "Your markdown answer"}. '
+        "Do not provide any other information."
+    ),
+    "correct": (
+        "Correct grammar and spelling of the markdown text, "
+        "preserving language and markdown formatting. "
+        'Return JSON: {"answer": "your corrected markdown text"}. '
+        "Do not provide any other information."
+    ),
+    "rephrase": (
+        "Rephrase the given markdown text, "
+        "preserving language and markdown formatting. "
+        'Return JSON: {"answer": "your rephrased markdown text"}. '
+        "Do not provide any other information."
+    ),
+    "summarize": (
+        "Summarize the markdown text, preserving language and markdown formatting. "
+        'Return JSON: {"answer": "your markdown summary"}. '
+        "Do not provide any other information."
+    ),
+}
+
+AI_TRANSLATE = (
+    "Translate the markdown text to {language:s}, preserving markdown formatting. "
+    'Return JSON: {{"answer": "your translated markdown text in {language:s}"}}. '
+    "Do not provide any other information."
+)
+
+
+class AIService:
+    """Service class for AI-related operations."""
+
+    def __init__(self):
+        """Ensure that the AI configuration is set properly."""
+        if (
+            settings.AI_BASE_URL is None
+            or settings.AI_API_KEY is None
+            or settings.AI_MODEL is None
+        ):
+            raise ImproperlyConfigured("AI configuration not set")
+        self.client = OpenAI(base_url=settings.AI_BASE_URL, api_key=settings.AI_API_KEY)
+
+    def call_ai_api(self, system_content, text):
+        """Helper method to call the OpenAI API and process the response."""
+        response = self.client.chat.completions.create(
+            model=settings.AI_MODEL,
+            response_format={"type": "json_object"},
+            messages=[
+                {"role": "system", "content": system_content},
+                {"role": "user", "content": json.dumps({"markdown_input": text})},
+            ],
+        )
+
+        content = response.choices[0].message.content
+        sanitized_content = re.sub(r"(?<!\\)\n", "\\\\n", content)
+        sanitized_content = re.sub(r"(?<!\\)\t", "\\\\t", sanitized_content)
+
+        json_response = json.loads(sanitized_content)
+
+        if "answer" not in json_response:
+            raise RuntimeError("AI response does not contain an answer")
+
+        return json_response
+
+    def transform(self, text, action):
+        """Transform text based on specified action."""
+        system_content = AI_ACTIONS[action]
+        return self.call_ai_api(system_content, text)
+
+    def translate(self, text, language):
+        """Translate text to a specified language."""
+        language_display = enums.ALL_LANGUAGES.get(language, language)
+        system_content = AI_TRANSLATE.format(language=language_display)
+        return self.call_ai_api(system_content, text)

src/backend/core/tests/documents/test_api_documents_ai_transform.py

@@ -0,0 +1,346 @@
+"""
+Test AI transform API endpoint for users in impress's core app.
+"""
+
+from unittest.mock import MagicMock, patch
+
+from django.core.cache import cache
+from django.test import override_settings
+
+import pytest
+from rest_framework.test import APIClient
+
+from core import factories
+from core.tests.conftest import TEAM, USER, VIA
+
+pytestmark = pytest.mark.django_db
+
+
+@pytest.fixture(autouse=True)
+def clear_cache():
+    """Fixture to clear the cache before each test."""
+    cache.clear()
+
+
+@pytest.fixture
+def ai_settings():
+    """Fixture to set AI settings."""
+    with override_settings(
+        AI_BASE_URL="http://example.com", AI_API_KEY="test-key", AI_MODEL="llama"
+    ):
+        yield
+
+
+@pytest.mark.parametrize(
+    "reach, role",
+    [
+        ("restricted", "reader"),
+        ("restricted", "editor"),
+        ("authenticated", "reader"),
+        ("authenticated", "editor"),
+        ("public", "reader"),
+    ],
+)
+def test_api_documents_ai_transform_anonymous_forbidden(reach, role):
+    """
+    Anonymous users should not be able to request AI transform if the link reach
+    and role don't allow it.
+    """
+    document = factories.DocumentFactory(link_reach=reach, link_role=role)
+
+    url = f"/api/v1.0/documents/{document.id!s}/ai-transform/"
+    response = APIClient().post(url, {"text": "hello", "action": "prompt"})
+
+    assert response.status_code == 401
+    assert response.json() == {
+        "detail": "Authentication credentials were not provided."
+    }
+
+
+@pytest.mark.usefixtures("ai_settings")
+@patch("openai.resources.chat.completions.Completions.create")
+def test_api_documents_ai_transform_anonymous_success(mock_create):
+    """
+    Anonymous users should be able to request AI transform to a document
+    if the link reach and role permit it.
+    """
+    document = factories.DocumentFactory(link_reach="public", link_role="editor")
+
+    answer = '{"answer": "Salut"}'
+    mock_create.return_value = MagicMock(
+        choices=[MagicMock(message=MagicMock(content=answer))]
+    )
+
+    url = f"/api/v1.0/documents/{document.id!s}/ai-transform/"
+    response = APIClient().post(url, {"text": "Hello", "action": "summarize"})
+
+    assert response.status_code == 200
+    assert response.json() == {"answer": "Salut"}
+    mock_create.assert_called_once_with(
+        model="llama",
+        response_format={"type": "json_object"},
+        messages=[
+            {
+                "role": "system",
+                "content": (
+                    "Summarize the markdown text, preserving language and markdown formatting. "
+                    'Return JSON: {"answer": "your markdown summary"}. Do not provide any other '
+                    "information."
+                ),
+            },
+            {"role": "user", "content": '{"markdown_input": "Hello"}'},
+        ],
+    )
+
+
+@pytest.mark.parametrize(
+    "reach, role",
+    [
+        ("restricted", "reader"),
+        ("restricted", "editor"),
+        ("authenticated", "reader"),
+        ("public", "reader"),
+    ],
+)
+def test_api_documents_ai_transform_authenticated_forbidden(reach, role):
+    """
+    Users who are not related to a document can't request AI transform if the
+    link reach and role don't allow it.
+    """
+    user = factories.UserFactory()
+
+    client = APIClient()
+    client.force_login(user)
+
+    document = factories.DocumentFactory(link_reach=reach, link_role=role)
+
+    url = f"/api/v1.0/documents/{document.id!s}/ai-transform/"
+    response = client.post(url, {"text": "Hello", "action": "prompt"})
+
+    assert response.status_code == 403
+    assert response.json() == {
+        "detail": "You do not have permission to perform this action."
+    }
+
+
+@pytest.mark.parametrize(
+    "reach, role",
+    [
+        ("authenticated", "editor"),
+        ("public", "editor"),
+    ],
+)
+@pytest.mark.usefixtures("ai_settings")
+@patch("openai.resources.chat.completions.Completions.create")
+def test_api_documents_ai_transform_authenticated_success(mock_create, reach, role):
+    """
+    Autenticated who are not related to a document should be able to request AI transform
+    if the link reach and role permit it.
+    """
+    user = factories.UserFactory()
+
+    client = APIClient()
+    client.force_login(user)
+
+    document = factories.DocumentFactory(link_reach=reach, link_role=role)
+
+    answer = '{"answer": "Salut"}'
+    mock_create.return_value = MagicMock(
+        choices=[MagicMock(message=MagicMock(content=answer))]
+    )
+
+    url = f"/api/v1.0/documents/{document.id!s}/ai-transform/"
+    response = client.post(url, {"text": "Hello", "action": "prompt"})
+
+    assert response.status_code == 200
+    assert response.json() == {"answer": "Salut"}
+    mock_create.assert_called_once_with(
+        model="llama",
+        response_format={"type": "json_object"},
+        messages=[
+            {
+                "role": "system",
+                "content": (
+                    'Answer the prompt in markdown format. Return JSON: {"answer": '
+                    '"Your markdown answer"}. Do not provide any other information.'
+                ),
+            },
+            {"role": "user", "content": '{"markdown_input": "Hello"}'},
+        ],
+    )
+
+
+@pytest.mark.parametrize("via", VIA)
+def test_api_documents_ai_transform_reader(via, mock_user_teams):
+    """
+    Users who are simple readers on a document should not be allowed to request AI transform.
+    """
+    user = factories.UserFactory()
+
+    client = APIClient()
+    client.force_login(user)
+
+    document = factories.DocumentFactory(link_role="reader")
+    if via == USER:
+        factories.UserDocumentAccessFactory(document=document, user=user, role="reader")
+    elif via == TEAM:
+        mock_user_teams.return_value = ["lasuite", "unknown"]
+        factories.TeamDocumentAccessFactory(
+            document=document, team="lasuite", role="reader"
+        )
+
+    url = f"/api/v1.0/documents/{document.id!s}/ai-transform/"
+    response = client.post(url, {"text": "Hello", "action": "prompt"})
+
+    assert response.status_code == 403
+    assert response.json() == {
+        "detail": "You do not have permission to perform this action."
+    }
+
+
+@pytest.mark.parametrize("role", ["editor", "administrator", "owner"])
+@pytest.mark.parametrize("via", VIA)
+@pytest.mark.usefixtures("ai_settings")
+@patch("openai.resources.chat.completions.Completions.create")
+def test_api_documents_ai_transform_success(mock_create, via, role, mock_user_teams):
+    """
+    Editors, administrators and owners of a document should be able to request AI transform.
+    """
+    user = factories.UserFactory()
+
+    client = APIClient()
+    client.force_login(user)
+
+    document = factories.DocumentFactory()
+    if via == USER:
+        factories.UserDocumentAccessFactory(document=document, user=user, role=role)
+    elif via == TEAM:
+        mock_user_teams.return_value = ["lasuite", "unknown"]
+        factories.TeamDocumentAccessFactory(
+            document=document, team="lasuite", role=role
+        )
+
+    answer = '{"answer": "Salut"}'
+    mock_create.return_value = MagicMock(
+        choices=[MagicMock(message=MagicMock(content=answer))]
+    )
+
+    url = f"/api/v1.0/documents/{document.id!s}/ai-transform/"
+    response = client.post(url, {"text": "Hello", "action": "prompt"})
+
+    assert response.status_code == 200
+    assert response.json() == {"answer": "Salut"}
+    mock_create.assert_called_once_with(
+        model="llama",
+        response_format={"type": "json_object"},
+        messages=[
+            {
+                "role": "system",
+                "content": (
+                    'Answer the prompt in markdown format. Return JSON: {"answer": '
+                    '"Your markdown answer"}. Do not provide any other information.'
+                ),
+            },
+            {"role": "user", "content": '{"markdown_input": "Hello"}'},
+        ],
+    )
+
+
+def test_api_documents_ai_transform_empty_text():
+    """The text should not be empty when requesting AI transform."""
+    user = factories.UserFactory()
+
+    client = APIClient()
+    client.force_login(user)
+
+    document = factories.DocumentFactory(link_reach="public", link_role="editor")
+
+    url = f"/api/v1.0/documents/{document.id!s}/ai-transform/"
+    response = client.post(url, {"text": " ", "action": "prompt"})
+
+    assert response.status_code == 400
+    assert response.json() == {"text": ["This field may not be blank."]}
+
+
+def test_api_documents_ai_transform_invalid_action():
+    """The action should valid when requesting AI transform."""
+    user = factories.UserFactory()
+
+    client = APIClient()
+    client.force_login(user)
+
+    document = factories.DocumentFactory(link_reach="public", link_role="editor")
+
+    url = f"/api/v1.0/documents/{document.id!s}/ai-transform/"
+    response = client.post(url, {"text": "Hello", "action": "invalid"})
+
+    assert response.status_code == 400
+    assert response.json() == {"action": ['"invalid" is not a valid choice.']}
+
+
+@override_settings(AI_DOCUMENT_RATE_THROTTLE_RATES={"minute": 3, "hour": 6, "day": 10})
+@pytest.mark.usefixtures("ai_settings")
+@patch("openai.resources.chat.completions.Completions.create")
+def test_api_documents_ai_transform_throttling_document(mock_create):
+    """
+    Throttling per document should be triggered on the AI transform endpoint.
+    For full throttle class test see: `test_api_utils_ai_document_rate_throttles`
+    """
+    client = APIClient()
+    document = factories.DocumentFactory(link_reach="public", link_role="editor")
+
+    answer = '{"answer": "Salut"}'
+    mock_create.return_value = MagicMock(
+        choices=[MagicMock(message=MagicMock(content=answer))]
+    )
+
+    url = f"/api/v1.0/documents/{document.id!s}/ai-transform/"
+    for _ in range(3):
+        user = factories.UserFactory()
+        client.force_login(user)
+        response = client.post(url, {"text": "Hello", "action": "summarize"})
+        assert response.status_code == 200
+        assert response.json() == {"answer": "Salut"}
+
+    user = factories.UserFactory()
+    client.force_login(user)
+    response = client.post(url, {"text": "Hello", "action": "summarize"})
+
+    assert response.status_code == 429
+    assert response.json() == {
+        "detail": "Request was throttled. Expected available in 60 seconds."
+    }
+
+
+@override_settings(AI_USER_RATE_THROTTLE_RATES={"minute": 3, "hour": 6, "day": 10})
+@pytest.mark.usefixtures("ai_settings")
+@patch("openai.resources.chat.completions.Completions.create")
+def test_api_documents_ai_transform_throttling_user(mock_create):
+    """
+    Throttling per user should be triggered on the AI transform endpoint.
+    For full throttle class test see: `test_api_utils_ai_user_rate_throttles`
+    """
+    user = factories.UserFactory()
+    client = APIClient()
+    client.force_login(user)
+
+    answer = '{"answer": "Salut"}'
+    mock_create.return_value = MagicMock(
+        choices=[MagicMock(message=MagicMock(content=answer))]
+    )
+
+    for _ in range(3):
+        document = factories.DocumentFactory(link_reach="public", link_role="editor")
+        url = f"/api/v1.0/documents/{document.id!s}/ai-transform/"
+        response = client.post(url, {"text": "Hello", "action": "summarize"})
+        assert response.status_code == 200
+        assert response.json() == {"answer": "Salut"}
+
+    document = factories.DocumentFactory(link_reach="public", link_role="editor")
+    url = f"/api/v1.0/documents/{document.id!s}/ai-transform/"
+    response = client.post(url, {"text": "Hello", "action": "summarize"})
+
+    assert response.status_code == 429
+    assert response.json() == {
+        "detail": "Request was throttled. Expected available in 60 seconds."
+    }

src/backend/core/tests/documents/test_api_documents_ai_translate.py

@@ -0,0 +1,370 @@
+"""
+Test AI translate API endpoint for users in impress's core app.
+"""
+
+from unittest.mock import MagicMock, patch
+
+from django.core.cache import cache
+from django.test import override_settings
+
+import pytest
+from rest_framework.test import APIClient
+
+from core import factories
+from core.tests.conftest import TEAM, USER, VIA
+
+pytestmark = pytest.mark.django_db
+
+
+@pytest.fixture(autouse=True)
+def clear_cache():
+    """Fixture to clear the cache before each test."""
+    cache.clear()
+
+
+@pytest.fixture
+def ai_settings():
+    """Fixture to set AI settings."""
+    with override_settings(
+        AI_BASE_URL="http://example.com", AI_API_KEY="test-key", AI_MODEL="llama"
+    ):
+        yield
+
+
+def test_api_documents_ai_translate_viewset_options_metadata():
+    """The documents endpoint should give us the list of available languages."""
+    user = factories.UserFactory()
+
+    client = APIClient()
+    client.force_login(user)
+
+    factories.DocumentFactory(link_reach="public", link_role="editor")
+
+    response = APIClient().options("/api/v1.0/documents/")
+
+    assert response.status_code == 200
+    metadata = response.json()
+    assert metadata["name"] == "Document List"
+    assert metadata["actions"]["POST"]["language"]["choices"][0] == {
+        "value": "af",
+        "display_name": "Afrikaans",
+    }
+
+
+@pytest.mark.parametrize(
+    "reach, role",
+    [
+        ("restricted", "reader"),
+        ("restricted", "editor"),
+        ("authenticated", "reader"),
+        ("authenticated", "editor"),
+        ("public", "reader"),
+    ],
+)
+def test_api_documents_ai_translate_anonymous_forbidden(reach, role):
+    """
+    Anonymous users should not be able to request AI translate if the link reach
+    and role don't allow it.
+    """
+    document = factories.DocumentFactory(link_reach=reach, link_role=role)
+
+    url = f"/api/v1.0/documents/{document.id!s}/ai-translate/"
+    response = APIClient().post(url, {"text": "hello", "language": "es"})
+
+    assert response.status_code == 401
+    assert response.json() == {
+        "detail": "Authentication credentials were not provided."
+    }
+
+
+@pytest.mark.usefixtures("ai_settings")
+@patch("openai.resources.chat.completions.Completions.create")
+def test_api_documents_ai_translate_anonymous_success(mock_create):
+    """
+    Anonymous users should be able to request AI translate to a document
+    if the link reach and role permit it.
+    """
+    document = factories.DocumentFactory(link_reach="public", link_role="editor")
+
+    answer = '{"answer": "Salut"}'
+    mock_create.return_value = MagicMock(
+        choices=[MagicMock(message=MagicMock(content=answer))]
+    )
+
+    url = f"/api/v1.0/documents/{document.id!s}/ai-translate/"
+    response = APIClient().post(url, {"text": "Hello", "language": "es"})
+
+    assert response.status_code == 200
+    assert response.json() == {"answer": "Salut"}
+    mock_create.assert_called_once_with(
+        model="llama",
+        response_format={"type": "json_object"},
+        messages=[
+            {
+                "role": "system",
+                "content": (
+                    "Translate the markdown text to Spanish, preserving markdown formatting. "
+                    'Return JSON: {"answer": "your translated markdown text in Spanish"}. '
+                    "Do not provide any other information."
+                ),
+            },
+            {"role": "user", "content": '{"markdown_input": "Hello"}'},
+        ],
+    )
+
+
+@pytest.mark.parametrize(
+    "reach, role",
+    [
+        ("restricted", "reader"),
+        ("restricted", "editor"),
+        ("authenticated", "reader"),
+        ("public", "reader"),
+    ],
+)
+def test_api_documents_ai_translate_authenticated_forbidden(reach, role):
+    """
+    Users who are not related to a document can't request AI translate if the
+    link reach and role don't allow it.
+    """
+    user = factories.UserFactory()
+
+    client = APIClient()
+    client.force_login(user)
+
+    document = factories.DocumentFactory(link_reach=reach, link_role=role)
+
+    url = f"/api/v1.0/documents/{document.id!s}/ai-translate/"
+    response = client.post(url, {"text": "Hello", "language": "es"})
+
+    assert response.status_code == 403
+    assert response.json() == {
+        "detail": "You do not have permission to perform this action."
+    }
+
+
+@pytest.mark.parametrize(
+    "reach, role",
+    [
+        ("authenticated", "editor"),
+        ("public", "editor"),
+    ],
+)
+@pytest.mark.usefixtures("ai_settings")
+@patch("openai.resources.chat.completions.Completions.create")
+def test_api_documents_ai_translate_authenticated_success(mock_create, reach, role):
+    """
+    Autenticated who are not related to a document should be able to request AI translate
+    if the link reach and role permit it.
+    """
+    user = factories.UserFactory()
+
+    client = APIClient()
+    client.force_login(user)
+
+    document = factories.DocumentFactory(link_reach=reach, link_role=role)
+
+    answer = '{"answer": "Salut"}'
+    mock_create.return_value = MagicMock(
+        choices=[MagicMock(message=MagicMock(content=answer))]
+    )
+
+    url = f"/api/v1.0/documents/{document.id!s}/ai-translate/"
+    response = client.post(url, {"text": "Hello", "language": "es-co"})
+
+    assert response.status_code == 200
+    assert response.json() == {"answer": "Salut"}
+    mock_create.assert_called_once_with(
+        model="llama",
+        response_format={"type": "json_object"},
+        messages=[
+            {
+                "role": "system",
+                "content": (
+                    "Translate the markdown text to Colombian Spanish, "
+                    "preserving markdown formatting. Return JSON: "
+                    '{"answer": "your translated markdown text in Colombian Spanish"}. '
+                    "Do not provide any other information."
+                ),
+            },
+            {"role": "user", "content": '{"markdown_input": "Hello"}'},
+        ],
+    )
+
+
+@pytest.mark.parametrize("via", VIA)
+def test_api_documents_ai_translate_reader(via, mock_user_teams):
+    """
+    Users who are simple readers on a document should not be allowed to request AI translate.
+    """
+    user = factories.UserFactory()
+
+    client = APIClient()
+    client.force_login(user)
+
+    document = factories.DocumentFactory(link_role="reader")
+    if via == USER:
+        factories.UserDocumentAccessFactory(document=document, user=user, role="reader")
+    elif via == TEAM:
+        mock_user_teams.return_value = ["lasuite", "unknown"]
+        factories.TeamDocumentAccessFactory(
+            document=document, team="lasuite", role="reader"
+        )
+
+    url = f"/api/v1.0/documents/{document.id!s}/ai-translate/"
+    response = client.post(url, {"text": "Hello", "language": "es"})
+
+    assert response.status_code == 403
+    assert response.json() == {
+        "detail": "You do not have permission to perform this action."
+    }
+
+
+@pytest.mark.parametrize("role", ["editor", "administrator", "owner"])
+@pytest.mark.parametrize("via", VIA)
+@pytest.mark.usefixtures("ai_settings")
+@patch("openai.resources.chat.completions.Completions.create")
+def test_api_documents_ai_translate_success(mock_create, via, role, mock_user_teams):
+    """
+    Editors, administrators and owners of a document should be able to request AI translate.
+    """
+    user = factories.UserFactory()
+
+    client = APIClient()
+    client.force_login(user)
+
+    document = factories.DocumentFactory()
+    if via == USER:
+        factories.UserDocumentAccessFactory(document=document, user=user, role=role)
+    elif via == TEAM:
+        mock_user_teams.return_value = ["lasuite", "unknown"]
+        factories.TeamDocumentAccessFactory(
+            document=document, team="lasuite", role=role
+        )
+
+    answer = '{"answer": "Salut"}'
+    mock_create.return_value = MagicMock(
+        choices=[MagicMock(message=MagicMock(content=answer))]
+    )
+
+    url = f"/api/v1.0/documents/{document.id!s}/ai-translate/"
+    response = client.post(url, {"text": "Hello", "language": "es-co"})
+
+    assert response.status_code == 200
+    assert response.json() == {"answer": "Salut"}
+    mock_create.assert_called_once_with(
+        model="llama",
+        response_format={"type": "json_object"},
+        messages=[
+            {
+                "role": "system",
+                "content": (
+                    "Translate the markdown text to Colombian Spanish, "
+                    "preserving markdown formatting. Return JSON: "
+                    '{"answer": "your translated markdown text in Colombian Spanish"}. '
+                    "Do not provide any other information."
+                ),
+            },
+            {"role": "user", "content": '{"markdown_input": "Hello"}'},
+        ],
+    )
+
+
+def test_api_documents_ai_translate_empty_text():
+    """The text should not be empty when requesting AI translate."""
+    user = factories.UserFactory()
+
+    client = APIClient()
+    client.force_login(user)
+
+    document = factories.DocumentFactory(link_reach="public", link_role="editor")
+
+    url = f"/api/v1.0/documents/{document.id!s}/ai-translate/"
+    response = client.post(url, {"text": " ", "language": "es"})
+
+    assert response.status_code == 400
+    assert response.json() == {"text": ["This field may not be blank."]}
+
+
+def test_api_documents_ai_translate_invalid_action():
+    """The action should valid when requesting AI translate."""
+    user = factories.UserFactory()
+
+    client = APIClient()
+    client.force_login(user)
+
+    document = factories.DocumentFactory(link_reach="public", link_role="editor")
+
+    url = f"/api/v1.0/documents/{document.id!s}/ai-translate/"
+    response = client.post(url, {"text": "Hello", "language": "invalid"})
+
+    assert response.status_code == 400
+    assert response.json() == {"language": ['"invalid" is not a valid choice.']}
+
+
+@override_settings(AI_DOCUMENT_RATE_THROTTLE_RATES={"minute": 3, "hour": 6, "day": 10})
+@pytest.mark.usefixtures("ai_settings")
+@patch("openai.resources.chat.completions.Completions.create")
+def test_api_documents_ai_translate_throttling_document(mock_create):
+    """
+    Throttling per document should be triggered on the AI translate endpoint.
+    For full throttle class test see: `test_api_utils_ai_document_rate_throttles`
+    """
+    client = APIClient()
+    document = factories.DocumentFactory(link_reach="public", link_role="editor")
+
+    answer = '{"answer": "Salut"}'
+    mock_create.return_value = MagicMock(
+        choices=[MagicMock(message=MagicMock(content=answer))]
+    )
+
+    url = f"/api/v1.0/documents/{document.id!s}/ai-translate/"
+    for _ in range(3):
+        user = factories.UserFactory()
+        client.force_login(user)
+        response = client.post(url, {"text": "Hello", "language": "es"})
+        assert response.status_code == 200
+        assert response.json() == {"answer": "Salut"}
+
+    user = factories.UserFactory()
+    client.force_login(user)
+    response = client.post(url, {"text": "Hello", "language": "es"})
+
+    assert response.status_code == 429
+    assert response.json() == {
+        "detail": "Request was throttled. Expected available in 60 seconds."
+    }
+
+
+@override_settings(AI_USER_RATE_THROTTLE_RATES={"minute": 3, "hour": 6, "day": 10})
+@pytest.mark.usefixtures("ai_settings")
+@patch("openai.resources.chat.completions.Completions.create")
+def test_api_documents_ai_translate_throttling_user(mock_create):
+    """
+    Throttling per user should be triggered on the AI translate endpoint.
+    For full throttle class test see: `test_api_utils_ai_user_rate_throttles`
+    """
+    user = factories.UserFactory()
+    client = APIClient()
+    client.force_login(user)
+
+    answer = '{"answer": "Salut"}'
+    mock_create.return_value = MagicMock(
+        choices=[MagicMock(message=MagicMock(content=answer))]
+    )
+
+    for _ in range(3):
+        document = factories.DocumentFactory(link_reach="public", link_role="editor")
+        url = f"/api/v1.0/documents/{document.id!s}/ai-translate/"
+        response = client.post(url, {"text": "Hello", "language": "es"})
+        assert response.status_code == 200
+        assert response.json() == {"answer": "Salut"}
+
+    document = factories.DocumentFactory(link_reach="public", link_role="editor")
+    url = f"/api/v1.0/documents/{document.id!s}/ai-translate/"
+    response = client.post(url, {"text": "Hello", "language": "es"})
+
+    assert response.status_code == 429
+    assert response.json() == {
+        "detail": "Request was throttled. Expected available in 60 seconds."
+    }

src/backend/core/tests/documents/test_api_documents_retrieve.py

@@ -21,6 +21,8 @@ def test_api_documents_retrieve_anonymous_public():
     assert response.json() == {
         "id": str(document.id),
         "abilities": {
+            "ai_transform": document.link_role == "editor",
+            "ai_translate": document.link_role == "editor",
             "attachment_upload": document.link_role == "editor",
             "destroy": False,
             "link_configuration": False,
@@ -75,6 +77,8 @@ def test_api_documents_retrieve_authenticated_unrelated_public_or_authenticated(
     assert response.json() == {
         "id": str(document.id),
         "abilities": {
+            "ai_transform": document.link_role == "editor",
+            "ai_translate": document.link_role == "editor",
             "attachment_upload": document.link_role == "editor",
             "link_configuration": False,
             "destroy": False,

src/backend/core/tests/test_api_utils_ai_document_rate_throttles.py

@@ -0,0 +1,127 @@
+"""
+Test throttling on documents for the AI endpoint.
+"""
+
+from unittest.mock import patch
+
+from django.core.cache import cache
+from django.test import override_settings
+
+import pytest
+from rest_framework.response import Response
+from rest_framework.test import APIRequestFactory
+from rest_framework.views import APIView
+
+from core.api.utils import AIDocumentRateThrottle
+
+
+class DocumentAPIView(APIView):
+    """A simple view to test the throttle"""
+
+    throttle_classes = [AIDocumentRateThrottle]
+
+    def get(self, request, *args, **kwargs):
+        """Minimal get method for testing purposes."""
+        return Response({"message": "Success"})
+
+
+@pytest.fixture(autouse=True)
+def clear_cache():
+    """Fixture to clear the cache before each test."""
+    cache.clear()
+
+
+@override_settings(AI_DOCUMENT_RATE_THROTTLE_RATES={"minute": 3, "hour": 6, "day": 10})
+@patch("time.time")
+def test_api_utils_ai_document_rate_throttle_minute_limit(mock_time):
+    """Test that minute limit is enforced."""
+    api_rf = APIRequestFactory()
+    mock_time.return_value = 1000000
+
+    # Simulate requests to the document API
+    for _i in range(3):  # 3 first requests should be allowed
+        request = api_rf.get("/documents/1/")
+        response = DocumentAPIView.as_view()(request, pk=1)
+        assert response.status_code == 200
+
+    # Simulate passage of time
+    mock_time.return_value += 59
+
+    # 4th request should be throttled
+    request = api_rf.get("/documents/1/")
+    response = DocumentAPIView.as_view()(request, pk=1)
+    assert response.status_code == 429
+
+    # After the 60s backoff wait time has passed, we can make a request again
+    mock_time.return_value += 1
+
+    request = api_rf.get("/documents/1/")
+    response = DocumentAPIView.as_view()(request, pk=1)
+    assert response.status_code == 200
+
+
+@override_settings(
+    AI_DOCUMENT_RATE_THROTTLE_RATES={"minute": 100000, "hour": 6, "day": 10}
+)
+@patch("time.time")
+def test_ai_document_rate_throttle_hour_limit(mock_time):
+    """Test that the hour limit is enforced without hitting the minute limit."""
+    api_rf = APIRequestFactory()
+    mock_time.return_value = 1000000
+
+    # Make requests to the document API, one per 21 seconds to avoid hitting the minute limit
+    for _i in range(6):
+        request = api_rf.get("/documents/1/")
+        response = DocumentAPIView.as_view()(request, pk=1)
+        assert response.status_code == 200
+
+        # Simulate passage of time
+        mock_time.return_value += 21
+
+    # Simulate passage of time
+    mock_time.return_value += 3600 - 6 * 21 - 1
+
+    # 7th request should be throttled
+    request = api_rf.get("/documents/1/")
+    response = DocumentAPIView.as_view()(request, pk=1)
+    assert response.status_code == 429
+
+    # After the 1h backoff wait time has passed, we can make a request again
+    mock_time.return_value += 1
+
+    request = api_rf.get("/documents/1/")
+    response = DocumentAPIView.as_view()(request, pk=1)
+    assert response.status_code == 200
+
+
+@override_settings(AI_DOCUMENT_RATE_THROTTLE_RATES={"minute": 3, "hour": 6, "day": 10})
+@patch("time.time")
+def test_api_utils_ai_document_rate_throttle_day_limit(mock_time):
+    """Test that day limit is enforced."""
+    api_rf = APIRequestFactory()
+    mock_time.return_value = 1000000
+
+    # Make requests to the document API, one per 10 minutes to avoid hitting
+    # the minute and hour limits
+    for _i in range(10):  # 10 requests should be allowed
+        request = api_rf.get("/documents/1/")
+        response = DocumentAPIView.as_view()(request, pk=1)
+        assert response.status_code == 200
+
+        # Simulate passage of time
+        mock_time.return_value += 60 * 10
+
+    # Simulate passage of time
+    mock_time.return_value += 24 * 3600 - 10 * 60 * 10 - 1
+
+    # 11th request should be throttled
+    request = api_rf.get("/documents/1/")
+    response = DocumentAPIView.as_view()(request, pk=1)
+    assert response.status_code == 429
+
+    # After the 24h backoff wait time has passed we can make a request again
+    mock_time.return_value += 1
+
+    request = api_rf.get("/documents/1/")
+    response = DocumentAPIView.as_view()(request, pk=1)
+    assert response.status_code == 200

src/backend/core/tests/test_api_utils_ai_user_rate_throttles.py

@@ -0,0 +1,146 @@
+"""
+Test throttling on users for the AI endpoint.
+"""
+
+from unittest.mock import patch
+from uuid import uuid4
+
+from django.core.cache import cache
+from django.test import override_settings
+
+import pytest
+from rest_framework.response import Response
+from rest_framework.test import APIRequestFactory
+from rest_framework.views import APIView
+
+from core.api.utils import AIUserRateThrottle
+from core.factories import UserFactory
+
+pytestmark = pytest.mark.django_db
+
+
+class DocumentAPIView(APIView):
+    """A simple view to test the throttle"""
+
+    throttle_classes = [AIUserRateThrottle]
+
+    def get(self, request, *args, **kwargs):
+        """Minimal get method for testing purposes."""
+        return Response({"message": "Success"})
+
+
+@pytest.fixture(autouse=True)
+def clear_cache():
+    """Fixture to clear the cache before each test."""
+    cache.clear()
+
+
+@override_settings(AI_USER_RATE_THROTTLE_RATES={"minute": 3, "hour": 6, "day": 10})
+@patch("time.time")
+def test_api_utils_ai_user_rate_throttle_minute_limit(mock_time):
+    """Test that minute limit is enforced."""
+    user = UserFactory()
+    api_rf = APIRequestFactory()
+    mock_time.return_value = 1000000
+
+    # Simulate requests to the document API
+    for _i in range(3):  # 3 first requests should be allowed
+        document_id = str(uuid4())
+        request = api_rf.get(f"/documents/{document_id:s}/")
+        request.user = user
+        response = DocumentAPIView.as_view()(request, pk=document_id)
+        assert response.status_code == 200
+
+    # Simulate passage of time
+    mock_time.return_value += 59
+
+    # 4th request should be throttled
+    document_id = str(uuid4())
+    request = api_rf.get(f"/documents/{document_id:s}/")
+    request.user = user
+    response = DocumentAPIView.as_view()(request, pk=document_id)
+    assert response.status_code == 429
+
+    # After the 60s backoff wait time has passed, we can make a request again
+    mock_time.return_value += 1
+
+    document_id = str(uuid4())
+    request = api_rf.get(f"/documents/{document_id:s}/")
+    request.user = user
+    response = DocumentAPIView.as_view()(request, pk=document_id)
+    assert response.status_code == 200
+
+
+@override_settings(AI_USER_RATE_THROTTLE_RATES={"minute": 100000, "hour": 6, "day": 10})
+@patch("time.time")
+def test_ai_user_rate_throttle_hour_limit(mock_time):
+    """Test that the hour limit is enforced without hitting the minute limit."""
+    user = UserFactory()
+    api_rf = APIRequestFactory()
+    mock_time.return_value = 1000000
+
+    # Make requests to the document API, one per 21 seconds to avoid hitting the minute limit
+    for _i in range(6):
+        document_id = str(uuid4())
+        request = api_rf.get(f"/documents/{document_id:s}/")
+        request.user = user
+        response = DocumentAPIView.as_view()(request, pk=document_id)
+        assert response.status_code == 200
+
+        # Simulate passage of time
+        mock_time.return_value += 21
+
+    # Simulate passage of time
+    mock_time.return_value += 3600 - 6 * 21 - 1
+
+    # 7th request should be throttled
+    request = api_rf.get(f"/documents/{document_id:s}/")
+    request.user = user
+    response = DocumentAPIView.as_view()(request, pk=document_id)
+    assert response.status_code == 429
+
+    # After the 1h backoff wait time has passed, we can make a request again
+    mock_time.return_value += 1
+
+    request = api_rf.get(f"/documents/{document_id:s}/")
+    request.user = user
+    response = DocumentAPIView.as_view()(request, pk=document_id)
+    assert response.status_code == 200
+
+
+@override_settings(AI_USER_RATE_THROTTLE_RATES={"minute": 3, "hour": 6, "day": 10})
+@patch("time.time")
+def test_api_utils_ai_user_rate_throttle_day_limit(mock_time):
+    """Test that day limit is enforced."""
+    user = UserFactory()
+    api_rf = APIRequestFactory()
+    mock_time.return_value = 1000000
+
+    # Make requests to the document API, one per 10 minutes to avoid hitting
+    # the minute and hour limits
+    for _i in range(10):  # 10 requests should be allowed
+        document_id = str(uuid4())
+        request = api_rf.get(f"/documents/{document_id:s}/")
+        request.user = user
+        response = DocumentAPIView.as_view()(request, pk=document_id)
+        assert response.status_code == 200
+
+        # Simulate passage of time
+        mock_time.return_value += 60 * 10
+
+    # Simulate passage of time
+    mock_time.return_value += 24 * 3600 - 10 * 60 * 10 - 1
+
+    # 11th request should be throttled
+    request = api_rf.get(f"/documents/{document_id:s}/")
+    request.user = user
+    response = DocumentAPIView.as_view()(request, pk=document_id)
+    assert response.status_code == 429
+
+    # After the 24h backoff wait time has passed we can make a request again
+    mock_time.return_value += 1
+
+    request = api_rf.get(f"/documents/{document_id:s}/")
+    request.user = user
+    response = DocumentAPIView.as_view()(request, pk=document_id)
+    assert response.status_code == 200

src/backend/core/tests/test_models_documents.py

@@ -83,6 +83,8 @@ def test_models_documents_get_abilities_forbidden(is_authenticated, reach, role)
     user = factories.UserFactory() if is_authenticated else AnonymousUser()
     abilities = document.get_abilities(user)
     assert abilities == {
+        "ai_transform": False,
+        "ai_translate": False,
         "attachment_upload": False,
         "link_configuration": False,
         "destroy": False,
@@ -113,6 +115,8 @@ def test_models_documents_get_abilities_reader(is_authenticated, reach):
     user = factories.UserFactory() if is_authenticated else AnonymousUser()
     abilities = document.get_abilities(user)
     assert abilities == {
+        "ai_transform": False,
+        "ai_translate": False,
         "attachment_upload": False,
         "destroy": False,
         "link_configuration": False,
@@ -143,6 +147,8 @@ def test_models_documents_get_abilities_editor(is_authenticated, reach):
     user = factories.UserFactory() if is_authenticated else AnonymousUser()
     abilities = document.get_abilities(user)
     assert abilities == {
+        "ai_transform": True,
+        "ai_translate": True,
         "attachment_upload": True,
         "destroy": False,
         "link_configuration": False,
@@ -162,6 +168,8 @@ def test_models_documents_get_abilities_owner():
     access = factories.UserDocumentAccessFactory(role="owner", user=user)
     abilities = access.document.get_abilities(access.user)
     assert abilities == {
+        "ai_transform": True,
+        "ai_translate": True,
         "attachment_upload": True,
         "destroy": True,
         "link_configuration": True,
@@ -180,6 +188,8 @@ def test_models_documents_get_abilities_administrator():
     access = factories.UserDocumentAccessFactory(role="administrator")
     abilities = access.document.get_abilities(access.user)
     assert abilities == {
+        "ai_transform": True,
+        "ai_translate": True,
         "attachment_upload": True,
         "destroy": False,
         "link_configuration": True,
@@ -201,6 +211,8 @@ def test_models_documents_get_abilities_editor_user(django_assert_num_queries):
         abilities = access.document.get_abilities(access.user)
 
     assert abilities == {
+        "ai_transform": True,
+        "ai_translate": True,
         "attachment_upload": True,
         "destroy": False,
         "link_configuration": False,
@@ -224,6 +236,8 @@ def test_models_documents_get_abilities_reader_user(django_assert_num_queries):
         abilities = access.document.get_abilities(access.user)
 
     assert abilities == {
+        "ai_transform": False,
+        "ai_translate": False,
         "attachment_upload": False,
         "destroy": False,
         "link_configuration": False,
@@ -248,6 +262,8 @@ def test_models_documents_get_abilities_preset_role(django_assert_num_queries):
         abilities = access.document.get_abilities(access.user)
 
     assert abilities == {
+        "ai_transform": False,
+        "ai_translate": False,
         "attachment_upload": False,
         "destroy": False,
         "link_configuration": False,

src/backend/core/tests/test_services_ai_services.py

@@ -0,0 +1,104 @@
+"""
+Test ai API endpoints in the impress core app.
+"""
+
+import json
+from unittest.mock import MagicMock, patch
+
+from django.core.exceptions import ImproperlyConfigured
+from django.test.utils import override_settings
+
+import pytest
+from openai import OpenAIError
+
+from core.services.ai_services import AIService
+
+pytestmark = pytest.mark.django_db
+
+
+@pytest.mark.parametrize(
+    "setting_name, setting_value",
+    [
+        ("AI_BASE_URL", None),
+        ("AI_API_KEY", None),
+        ("AI_MODEL", None),
+    ],
+)
+def test_api_ai_setting_missing(setting_name, setting_value):
+    """Setting should be set"""
+
+    with override_settings(**{setting_name: setting_value}):
+        with pytest.raises(
+            ImproperlyConfigured,
+            match="AI configuration not set",
+        ):
+            AIService()
+
+
+@override_settings(
+    AI_BASE_URL="http://example.com", AI_API_KEY="test-key", AI_MODEL="test-model"
+)
+@patch("openai.resources.chat.completions.Completions.create")
+def test_api_ai__client_error(mock_create):
+    """Fail when the client raises an error"""
+
+    mock_create.side_effect = OpenAIError("Mocked client error")
+
+    with pytest.raises(
+        OpenAIError,
+        match="Mocked client error",
+    ):
+        AIService().transform("hello", "prompt")
+
+
+@override_settings(
+    AI_BASE_URL="http://example.com", AI_API_KEY="test-key", AI_MODEL="test-model"
+)
+@patch("openai.resources.chat.completions.Completions.create")
+def test_api_ai__client_invalid_response(mock_create):
+    """Fail when the client response is invalid"""
+
+    answer = {"no_answer": "This is an invalid response"}
+    mock_create.return_value = MagicMock(
+        choices=[MagicMock(message=MagicMock(content=json.dumps(answer)))]
+    )
+
+    with pytest.raises(
+        RuntimeError,
+        match="AI response does not contain an answer",
+    ):
+        AIService().transform("hello", "prompt")
+
+
+@override_settings(
+    AI_BASE_URL="http://example.com", AI_API_KEY="test-key", AI_MODEL="test-model"
+)
+@patch("openai.resources.chat.completions.Completions.create")
+def test_api_ai__success(mock_create):
+    """The AI request should work as expect when called with valid arguments."""
+
+    answer = '{"answer": "Salut"}'
+    mock_create.return_value = MagicMock(
+        choices=[MagicMock(message=MagicMock(content=answer))]
+    )
+
+    response = AIService().transform("hello", "prompt")
+
+    assert response == {"answer": "Salut"}
+
+
+@override_settings(
+    AI_BASE_URL="http://example.com", AI_API_KEY="test-key", AI_MODEL="test-model"
+)
+@patch("openai.resources.chat.completions.Completions.create")
+def test_api_ai__success_sanitize(mock_create):
+    """The AI response should be sanitized"""
+
+    answer = '{"answer": "Salut\\n \tle \nmonde"}'
+    mock_create.return_value = MagicMock(
+        choices=[MagicMock(message=MagicMock(content=answer))]
+    )
+
+    response = AIService().transform("hello", "prompt")
+
+    assert response == {"answer": "Salut\n \tle \nmonde"}

src/backend/impress/settings.py

@@ -454,6 +454,20 @@ class Base(Configuration):
     ALLOW_LOGOUT_GET_METHOD = values.BooleanValue(
         default=True, environ_name="ALLOW_LOGOUT_GET_METHOD", environ_prefix=None
     )
+    AI_API_KEY = values.Value(None, environ_name="AI_API_KEY", environ_prefix=None)
+    AI_BASE_URL = values.Value(None, environ_name="AI_BASE_URL", environ_prefix=None)
+    AI_MODEL = values.Value(None, environ_name="AI_MODEL", environ_prefix=None)
+
+    AI_DOCUMENT_RATE_THROTTLE_RATES = {
+        "minute": 5,
+        "hour": 100,
+        "day": 500,
+    }
+    AI_USER_RATE_THROTTLE_RATES = {
+        "minute": 3,
+        "hour": 50,
+        "day": 200,
+    }
 
     USER_OIDC_FIELDS_TO_FULLNAME = values.ListValue(
         default=["first_name", "last_name"],

src/backend/pyproject.toml

@@ -46,6 +46,7 @@ dependencies = [
     "jsonschema==4.23.0",
     "markdown==3.7",
     "nested-multipart-parser==1.5.0",
+    "openai==1.44.1",
     "psycopg[binary]==3.2.3",
     "PyJWT==2.9.0",
     "pypandoc==1.14",