Limit vision_enabled image formatting to OpenAI APIs and send vision to extract_questions query

src/khoj/processor/conversation/anthropic/anthropic_chat.py

@@ -6,7 +6,7 @@ from typing import Dict, Optional
 
 from langchain.schema import ChatMessage
 
-from khoj.database.models import Agent, KhojUser
+from khoj.database.models import Agent, ChatModelOptions, KhojUser
 from khoj.processor.conversation import prompts
 from khoj.processor.conversation.anthropic.utils import (
     anthropic_chat_completion_with_backoff,
@@ -188,6 +188,7 @@ def converse_anthropic(
         model_name=model,
         max_prompt_size=max_prompt_size,
         tokenizer_name=tokenizer_name,
+        model_type=ChatModelOptions.ModelType.ANTHROPIC,
     )
 
     if len(messages) > 1:

src/khoj/processor/conversation/offline/chat_model.py

@@ -7,7 +7,7 @@ from typing import Any, Iterator, List, Union
 from langchain.schema import ChatMessage
 from llama_cpp import Llama
 
-from khoj.database.models import Agent, KhojUser
+from khoj.database.models import Agent, ChatModelOptions, KhojUser
 from khoj.processor.conversation import prompts
 from khoj.processor.conversation.offline.utils import download_model
 from khoj.processor.conversation.utils import (
@@ -76,7 +76,11 @@ def extract_questions_offline(
     )
 
     messages = generate_chatml_messages_with_context(
-        example_questions, model_name=model, loaded_model=offline_chat_model, max_prompt_size=max_prompt_size
+        example_questions,
+        model_name=model,
+        loaded_model=offline_chat_model,
+        max_prompt_size=max_prompt_size,
+        model_type=ChatModelOptions.ModelType.OFFLINE,
     )
 
     state.chat_lock.acquire()
@@ -201,6 +205,7 @@ def converse_offline(
         loaded_model=offline_chat_model,
         max_prompt_size=max_prompt_size,
         tokenizer_name=tokenizer_name,
+        model_type=ChatModelOptions.ModelType.OFFLINE,
     )
 
     truncated_messages = "\n".join({f"{message.content[:70]}..." for message in messages})

src/khoj/processor/conversation/openai/gpt.py

@@ -5,13 +5,16 @@ from typing import Dict, Optional
 
 from langchain.schema import ChatMessage
 
-from khoj.database.models import Agent, KhojUser
+from khoj.database.models import Agent, ChatModelOptions, KhojUser
 from khoj.processor.conversation import prompts
 from khoj.processor.conversation.openai.utils import (
     chat_completion_with_backoff,
     completion_with_backoff,
 )
-from khoj.processor.conversation.utils import generate_chatml_messages_with_context
+from khoj.processor.conversation.utils import (
+    construct_structured_message,
+    generate_chatml_messages_with_context,
+)
 from khoj.utils.helpers import ConversationCommand, is_none_or_empty
 from khoj.utils.rawconfig import LocationData
 
@@ -24,9 +27,10 @@ def extract_questions(
     conversation_log={},
     api_key=None,
     api_base_url=None,
-    temperature=0.7,
     location_data: LocationData = None,
     user: KhojUser = None,
+    uploaded_image_url: Optional[str] = None,
+    vision_enabled: bool = False,
 ):
     """
     Infer search queries to retrieve relevant notes to answer user query
@@ -63,17 +67,17 @@ def extract_questions(
         location=location,
         username=username,
     )
+
+    prompt = construct_structured_message(
+        message=prompt,
+        image_url=uploaded_image_url,
+        model_type=ChatModelOptions.ModelType.OPENAI,
+        vision_enabled=vision_enabled,
+    )
+
     messages = [ChatMessage(content=prompt, role="user")]
 
-    # Get Response from GPT
+    response = send_message_to_model(messages, api_key, model, response_type="json_object", api_base_url=api_base_url)
-    response = completion_with_backoff(
-        messages=messages,
-        model=model,
-        temperature=temperature,
-        api_base_url=api_base_url,
-        model_kwargs={"response_format": {"type": "json_object"}},
-        openai_api_key=api_key,
-    )
 
     # Extract, Clean Message from GPT's Response
     try:
@@ -182,6 +186,7 @@ def converse(
         tokenizer_name=tokenizer_name,
         uploaded_image_url=image_url,
         vision_enabled=vision_available,
+        model_type=ChatModelOptions.ModelType.OPENAI,
     )
     truncated_messages = "\n".join({f"{message.content[:70]}..." for message in messages})
     logger.debug(f"Conversation Context for GPT: {truncated_messages}")

src/khoj/processor/conversation/utils.py

@@ -12,7 +12,7 @@ from llama_cpp.llama import Llama
 from transformers import AutoTokenizer
 
 from khoj.database.adapters import ConversationAdapters
-from khoj.database.models import ClientApplication, KhojUser
+from khoj.database.models import ChatModelOptions, ClientApplication, KhojUser
 from khoj.processor.conversation.offline.utils import download_model, infer_max_tokens
 from khoj.utils import state
 from khoj.utils.helpers import is_none_or_empty, merge_dicts
@@ -137,6 +137,13 @@ Khoj: "{inferred_queries if ("text-to-image" in intent_type) else chat_response}
     )
 
 
+# Format user and system messages to chatml format
+def construct_structured_message(message, image_url, model_type, vision_enabled):
+    if image_url and vision_enabled and model_type == ChatModelOptions.ModelType.OPENAI:
+        return [{"type": "text", "text": message}, {"type": "image_url", "image_url": {"url": image_url}}]
+    return message
+
+
 def generate_chatml_messages_with_context(
     user_message,
     system_message=None,
@@ -147,6 +154,7 @@ def generate_chatml_messages_with_context(
     tokenizer_name=None,
     uploaded_image_url=None,
     vision_enabled=False,
+    model_type="",
 ):
     """Generate messages for ChatGPT with context from previous conversation"""
     # Set max prompt size from user config or based on pre-configured for model and machine specs
@@ -156,12 +164,6 @@ def generate_chatml_messages_with_context(
         else:
             max_prompt_size = model_to_prompt_size.get(model_name, 2000)
 
-    # Format user and system messages to chatml format
-    def construct_structured_message(message, image_url):
-        if image_url and vision_enabled:
-            return [{"type": "text", "text": message}, {"type": "image_url", "image_url": {"url": image_url}}]
-        return message
-
     # Scale lookback turns proportional to max prompt size supported by model
     lookback_turns = max_prompt_size // 750
 
@@ -174,7 +176,9 @@ def generate_chatml_messages_with_context(
         message_content = chat["message"] + message_notes
 
         if chat.get("uploadedImageData") and vision_enabled:
-            message_content = construct_structured_message(message_content, chat.get("uploadedImageData"))
+            message_content = construct_structured_message(
+                message_content, chat.get("uploadedImageData"), model_type, vision_enabled
+            )
 
         reconstructed_message = ChatMessage(content=message_content, role=role)
 
@@ -186,7 +190,10 @@ def generate_chatml_messages_with_context(
     messages = []
     if not is_none_or_empty(user_message):
         messages.append(
-            ChatMessage(content=construct_structured_message(user_message, uploaded_image_url), role="user")
+            ChatMessage(
+                content=construct_structured_message(user_message, uploaded_image_url, model_type, vision_enabled),
+                role="user",
+            )
         )
     if len(chatml_messages) > 0:
         messages += chatml_messages

src/khoj/routers/api.py

@@ -331,6 +331,7 @@ async def extract_references_and_questions(
     conversation_commands: List[ConversationCommand] = [ConversationCommand.Default],
     location_data: LocationData = None,
     send_status_func: Optional[Callable] = None,
+    uploaded_image_url: Optional[str] = None,
 ):
     user = request.user.object if request.user.is_authenticated else None
 
@@ -370,6 +371,7 @@ async def extract_references_and_questions(
     with timer("Extracting search queries took", logger):
         # If we've reached here, either the user has enabled offline chat or the openai model is enabled.
         conversation_config = await ConversationAdapters.aget_default_conversation_config()
+        vision_enabled = conversation_config.vision_enabled
 
         if conversation_config.model_type == ChatModelOptions.ModelType.OFFLINE:
             using_offline_chat = True
@@ -403,6 +405,8 @@ async def extract_references_and_questions(
                 conversation_log=meta_log,
                 location_data=location_data,
                 user=user,
+                uploaded_image_url=uploaded_image_url,
+                vision_enabled=vision_enabled,
             )
         elif conversation_config.model_type == ChatModelOptions.ModelType.ANTHROPIC:
             api_key = conversation_config.openai_config.api_key

src/khoj/routers/api_chat.py

@@ -807,6 +807,7 @@ async def chat(
             conversation_commands,
             location,
             partial(send_event, ChatEvent.STATUS),
+            uploaded_image_url=uploaded_image_url,
         ):
             if isinstance(result, dict) and ChatEvent.STATUS in result:
                 yield result[ChatEvent.STATUS]

src/khoj/routers/helpers.py

@@ -330,7 +330,7 @@ async def aget_relevant_output_modes(
     chat_history = construct_chat_history(conversation_history)
 
     if uploaded_image_url:
-        query = f"[placeholder for image attached to this message]\n{query}"
+        query = f"<user uploaded content redacted> \n{query}"
 
     relevant_mode_prompt = prompts.pick_relevant_output_mode.format(
         query=query,
@@ -622,6 +622,7 @@ async def send_message_to_model_wrapper(
             tokenizer_name=tokenizer,
             max_prompt_size=max_tokens,
             vision_enabled=vision_available,
+            model_type=conversation_config.model_type,
         )
 
         return send_message_to_model_offline(
@@ -644,6 +645,7 @@ async def send_message_to_model_wrapper(
             tokenizer_name=tokenizer,
             vision_enabled=vision_available,
             uploaded_image_url=uploaded_image_url,
+            model_type=conversation_config.model_type,
         )
 
         openai_response = send_message_to_model(
@@ -664,6 +666,7 @@ async def send_message_to_model_wrapper(
             max_prompt_size=max_tokens,
             tokenizer_name=tokenizer,
             vision_enabled=vision_available,
+            model_type=conversation_config.model_type,
         )
 
         return anthropic_send_message_to_model(
@@ -700,6 +703,7 @@ def send_message_to_model_wrapper_sync(
             model_name=chat_model,
             loaded_model=loaded_model,
             vision_enabled=vision_available,
+            model_type=conversation_config.model_type,
         )
 
         return send_message_to_model_offline(
@@ -717,6 +721,7 @@ def send_message_to_model_wrapper_sync(
             system_message=system_message,
             model_name=chat_model,
             vision_enabled=vision_available,
+            model_type=conversation_config.model_type,
         )
 
         openai_response = send_message_to_model(
@@ -733,6 +738,7 @@ def send_message_to_model_wrapper_sync(
             model_name=chat_model,
             max_prompt_size=max_tokens,
             vision_enabled=vision_available,
+            model_type=conversation_config.model_type,
         )
 
         return anthropic_send_message_to_model(