Fix: key the AI suggestion cache by model and endpoint (#13449)

---------

Co-authored-by: shamoon <4887959+shamoon@users.noreply.github.com>

src/documents/tests/test_views.py

@@ -491,6 +491,42 @@ class TestAISuggestions(DirectoriesMixin, TestCase):
             "Titre IA",
         )
 
+    @patch("documents.views.get_ai_document_classification")
+    @override_settings(
+        AI_ENABLED=True,
+        LLM_BACKEND="mock_backend",
+        LLM_MODEL="model-a",
+        LLM_ENDPOINT="http://endpoint-a",
+    )
+    def test_ai_suggestions_cache_key_includes_model_and_endpoint(
+        self,
+        mock_get_ai_classification,
+    ) -> None:
+        """Cached suggestions are keyed by model and endpoint, so switching
+        either yields a cache miss instead of a stale hit."""
+        mock_get_ai_classification.return_value = {
+            "title": "Answer A",
+            "tags": [],
+            "correspondents": [],
+            "document_types": [],
+            "storage_paths": [],
+            "dates": [],
+        }
+
+        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_200_OK)
+
+        # Cached under a key that carries model + endpoint...
+        self.assertIsNotNone(
+            get_llm_suggestion_cache(
+                self.document.pk,
+                backend="mock_backend:model-a:http://endpoint-a",
+            ),
+        )
+
     @patch("documents.views.get_ai_document_classification")
     @override_settings(
         AI_ENABLED=True,

src/documents/views.py

@@ -1530,10 +1530,15 @@ class DocumentViewSet(
             return HttpResponseBadRequest("AI is required for this feature")
 
         output_language = _get_llm_output_language(ai_config=ai_config, request=request)
-        llm_cache_backend = (
-            f"{ai_config.llm_backend}:{output_language}"
-            if output_language
-            else ai_config.llm_backend
+        llm_cache_backend = ":".join(
+            part
+            for part in (
+                ai_config.llm_backend,
+                ai_config.llm_model,
+                ai_config.llm_endpoint,
+                output_language,
+            )
+            if part
         )
 
         cached_llm_suggestions = get_llm_suggestion_cache(