Ignore WS heartbeats and reject empty chat messages.

Closes #31. Prevent ping keepalives and blank/whitespace prompts from creating conversations or hitting the LLM.

llm_be/chat_backend/consumers.py

@@ -27,6 +27,7 @@ from .services.title_generator import title_generator
 from .services.moderation_classifier import moderation_classifier, ModerationLabel
 from .services.prompt_classifier.prompt_classifier import PromptClassifier, PromptType
 from .services.data_analysis_service import AsyncDataAnalysisService
+from .utils import has_usable_user_prompt, is_heartbeat_payload, normalize_user_message
 
 logger = logging.getLogger(__name__)
 
@@ -238,13 +239,29 @@ class ChatConsumerAgain(AsyncWebsocketConsumer):
         logger.debug(f"Bytes Data: {bytes_data}")
         if text_data:
             data = json.loads(text_data)
-            message = data.get("message", None)
+            # Keepalive frames must not create conversations or hit the LLM.
+            if is_heartbeat_payload(data):
+                return
+
+            message = normalize_user_message(data.get("message", None))
             conversation_id = data.get("conversation_id", None)
             email = data.get("email", None)
             file = data.get("file", None)
             file_type = data.get("fileType", "")
             model = data.get("modelName", "Turbo")
 
+            if not has_usable_user_prompt(message, file):
+                logger.info("Ignoring websocket payload with empty message")
+                await self.send_json_message(
+                    json.dumps(
+                        {
+                            "type": "error",
+                            "content": "Message text cannot be empty.",
+                        }
+                    )
+                )
+                return
+
             if not conversation_id:
                 # we need to create a new conversation
                 # we will generate a name for it too

llm_be/chat_backend/consumers_graph.py

@@ -22,6 +22,7 @@ from .services.title_generator import title_generator
 from .services.moderation_classifier import moderation_classifier, ModerationLabel
 from .services.prompt_classifier.prompt_classifier import PromptClassifier, PromptType
 from .services.data_analysis_service import AsyncDataAnalysisService
+from .utils import has_usable_user_prompt, is_heartbeat_payload, normalize_user_message
 
 logger = logging.getLogger(__name__)
 
@@ -288,13 +289,29 @@ class ChatConsumerGraph(AsyncWebsocketConsumer):
         print("Text Data: ", text_data)
         if text_data:
             data = json.loads(text_data)
+            # Keepalive frames must not create conversations or hit the LLM.
+            if is_heartbeat_payload(data):
+                return
+
             model = data.get("modelName", "Turbo")
-            message = data.get("message", None)
+            message = normalize_user_message(data.get("message", None))
             conversation_id = data.get("conversation_id", None)
             email = data.get("email", None)
             file = data.get("file", None)
             file_type = data.get("fileType", "")
-            
+
+            if not has_usable_user_prompt(message, file):
+                logger.info("Ignoring websocket payload with empty message")
+                await self.send_json_message(
+                    json.dumps(
+                        {
+                            "type": "error",
+                            "content": "Message text cannot be empty.",
+                        }
+                    )
+                )
+                return
+
             if not conversation_id:
                 title = await title_generator.generate_async(message)
                 conversation_id = await create_conversation(message, email, title)

llm_be/chat_backend/serializers.py

@@ -122,6 +122,11 @@ class PromptSerializer(serializers.ModelSerializer):
             "id",
         )
 
+    def validate_message(self, value: str) -> str:
+        if value is None or not str(value).strip():
+            raise serializers.ValidationError("Message text cannot be empty.")
+        return str(value).strip()
+
 
 class BasicUserSerializer(serializers.ModelSerializer):
     class Meta:

llm_be/chat_backend/tests/test_consumers.py

@@ -1,4 +1,5 @@
 from unittest import mock
+import json
 
 from asgiref.sync import sync_to_async
 from channels.testing import WebsocketCommunicator
@@ -412,3 +413,61 @@ class WebSocketRoutingTestCase(TransactionTestCase):
             [str(route.pattern) for route in websocket_urlpatterns],
             ["ws/chat_again/$", "ws/conditional_chat/$"],
         )
+
+
+class WebSocketReceiveGuardTestCase(TransactionTestCase):
+    """Heartbeats / empty prompts must not spawn conversations or LLM work."""
+
+    @parameterized.expand(
+        [("chat", "/ws/chat_again/"), ("conditional_chat", "/ws/conditional_chat/")]
+    )
+    async def test_ping_heartbeat_is_ignored(self, _name, path):
+        communicator = WebsocketCommunicator(application, path)
+        connected, _ = await communicator.connect()
+        self.assertTrue(connected)
+
+        with mock.patch(
+            "chat_backend.consumers.title_generator.generate_async",
+            new_callable=mock.AsyncMock,
+        ) as title_chat, mock.patch(
+            "chat_backend.consumers_graph.title_generator.generate_async",
+            new_callable=mock.AsyncMock,
+        ) as title_graph:
+            await communicator.send_json_to({"type": "ping", "email": "a@b.com"})
+            # No reply expected; give the event loop a tick.
+            self.assertTrue(await communicator.receive_nothing(timeout=0.2))
+            title_chat.assert_not_called()
+            title_graph.assert_not_called()
+
+        count = await sync_to_async(Conversation.objects.count)()
+        self.assertEqual(count, 0)
+        await communicator.disconnect()
+
+    @parameterized.expand(
+        [("chat", "/ws/chat_again/"), ("conditional_chat", "/ws/conditional_chat/")]
+    )
+    async def test_empty_message_is_rejected(self, _name, path):
+        communicator = WebsocketCommunicator(application, path)
+        connected, _ = await communicator.connect()
+        self.assertTrue(connected)
+
+        with mock.patch(
+            "chat_backend.consumers.title_generator.generate_async",
+            new_callable=mock.AsyncMock,
+        ) as title_chat, mock.patch(
+            "chat_backend.consumers_graph.title_generator.generate_async",
+            new_callable=mock.AsyncMock,
+        ) as title_graph:
+            await communicator.send_json_to(
+                {"message": "   ", "email": "a@b.com", "conversation_id": None}
+            )
+            response = await communicator.receive_from(timeout=1)
+            payload = json.loads(response)
+            self.assertEqual(payload["type"], "error")
+            self.assertIn("empty", payload["content"].lower())
+            title_chat.assert_not_called()
+            title_graph.assert_not_called()
+
+        count = await sync_to_async(Conversation.objects.count)()
+        self.assertEqual(count, 0)
+        await communicator.disconnect()

llm_be/chat_backend/tests/test_serializers.py

@@ -58,6 +58,22 @@ class PromptSerializerTestCase(TestCase):
         self.assertFalse(serializer.is_valid())
         self.assertIn("message", serializer.errors)
 
+    def test_message_rejects_blank_and_whitespace(self):
+        for payload in ("", "   ", "\n\t"):
+            serializer = PromptSerializer(
+                data={"message": payload, "user_created": True}
+            )
+            self.assertFalse(serializer.is_valid(), payload)
+            self.assertIn("message", serializer.errors)
+
+    def test_message_is_stripped(self):
+        serializer = PromptSerializer(
+            data={"message": "  hello  ", "user_created": True}
+        )
+
+        self.assertTrue(serializer.is_valid(), serializer.errors)
+        self.assertEqual(serializer.validated_data["message"], "hello")
+
     def test_user_created_is_required(self):
         serializer = PromptSerializer(data={"message": "hi"})
 

llm_be/chat_backend/tests/test_utils.py

@@ -11,7 +11,13 @@ from chat_backend.ollama_config import (
     ollama_llm_kwargs,
     ollama_model,
 )
-from chat_backend.utils import extract_token_usage, last_day_of_month
+from chat_backend.utils import (
+    extract_token_usage,
+    has_usable_user_prompt,
+    is_heartbeat_payload,
+    last_day_of_month,
+    normalize_user_message,
+)
 
 
 class ExtractTokenUsageTestCase(SimpleTestCase):
@@ -134,3 +140,23 @@ class OllamaConfigFallbackTestCase(SimpleTestCase):
         with override_settings(OLLAMA_MODEL="llama3.2"):
             del settings.OLLAMA_EMBED_MODEL
             self.assertEqual(ollama_embed_model(), "llama3.2")
+
+
+class UserPromptGuardTestCase(SimpleTestCase):
+    def test_heartbeat_payload_detected(self):
+        self.assertTrue(is_heartbeat_payload({"type": "ping", "email": "a@b.com"}))
+        self.assertFalse(is_heartbeat_payload({"message": "hi"}))
+        self.assertFalse(is_heartbeat_payload(None))
+
+    def test_normalize_user_message(self):
+        self.assertIsNone(normalize_user_message(None))
+        self.assertIsNone(normalize_user_message(""))
+        self.assertIsNone(normalize_user_message("   \n\t"))
+        self.assertEqual(normalize_user_message("  hello  "), "hello")
+
+    def test_has_usable_user_prompt(self):
+        self.assertFalse(has_usable_user_prompt(None))
+        self.assertFalse(has_usable_user_prompt("   "))
+        self.assertFalse(has_usable_user_prompt("", file="base64"))
+        self.assertTrue(has_usable_user_prompt("hi"))
+        self.assertTrue(has_usable_user_prompt("  hi  "))

llm_be/chat_backend/utils.py

@@ -1,6 +1,26 @@
 import datetime
 
 
+def is_heartbeat_payload(data) -> bool:
+    """True for app-level WS keepalive frames (see FE buildHeartbeatPayload)."""
+    return isinstance(data, dict) and data.get("type") == "ping"
+
+
+def normalize_user_message(message):
+    """Return stripped message text, or None if missing/blank."""
+    if message is None:
+        return None
+    if not isinstance(message, str):
+        message = str(message)
+    stripped = message.strip()
+    return stripped or None
+
+
+def has_usable_user_prompt(message, file=None) -> bool:
+    """Reject empty/whitespace chat text. ``file`` kept for call-site clarity."""
+    return normalize_user_message(message) is not None
+
+
 def last_day_of_month(any_day):
     # The day 28 exists in every month. 4 days later, it's always next month
     next_month = any_day.replace(day=28) + datetime.timedelta(days=4)

llm_be/chat_backend/views.py

@@ -469,6 +469,12 @@ class ConversationDetailView(APIView):
 
         # make sure that our model exists and it is running
         prompt = request.data.get("prompt")
+        if not isinstance(prompt, str) or not prompt.strip():
+            return Response(
+                {"detail": "Message text cannot be empty."},
+                status=status.HTTP_400_BAD_REQUEST,
+            )
+        prompt = prompt.strip()
 
         conversation_id = request.data.get("conversation_id")
         is_user = bool(request.data.get("is_user"))