khoj 1.21.6.dev14__py3-none-any.whl → 1.21.7.dev4__py3-none-any.whl

khoj/routers/api_chat.py

@@ -1,4 +1,5 @@
 import asyncio
+import base64
 import json
 import logging
 import time
@@ -46,11 +47,13 @@ from khoj.routers.helpers import (
     update_telemetry_state,
     validate_conversation_config,
 )
+from khoj.routers.storage import upload_image_to_bucket
 from khoj.utils import state
 from khoj.utils.helpers import (
     AsyncIteratorWrapper,
     ConversationCommand,
     command_descriptions,
+    convert_image_to_webp,
     get_device,
     is_none_or_empty,
 )
@@ -517,21 +520,26 @@ async def set_conversation_title(
     )
 
 
-@api_chat.get("")
+class ChatRequestBody(BaseModel):
+    q: str
+    n: Optional[int] = 7
+    d: Optional[float] = None
+    stream: Optional[bool] = False
+    title: Optional[str] = None
+    conversation_id: Optional[int] = None
+    city: Optional[str] = None
+    region: Optional[str] = None
+    country: Optional[str] = None
+    timezone: Optional[str] = None
+    image: Optional[str] = None
+
+
+@api_chat.post("")
 @requires(["authenticated"])
 async def chat(
     request: Request,
     common: CommonQueryParams,
-    q: str,
-    n: int = 7,
-    d: float = None,
-    stream: Optional[bool] = False,
-    title: Optional[str] = None,
-    conversation_id: Optional[int] = None,
-    city: Optional[str] = None,
-    region: Optional[str] = None,
-    country: Optional[str] = None,
-    timezone: Optional[str] = None,
+    body: ChatRequestBody,
     rate_limiter_per_minute=Depends(
         ApiUserRateLimiter(requests=60, subscribed_requests=60, window=60, slug="chat_minute")
     ),
@@ -539,7 +547,20 @@ async def chat(
         ApiUserRateLimiter(requests=600, subscribed_requests=600, window=60 * 60 * 24, slug="chat_day")
     ),
 ):
-    async def event_generator(q: str):
+    # Access the parameters from the body
+    q = body.q
+    n = body.n
+    d = body.d
+    stream = body.stream
+    title = body.title
+    conversation_id = body.conversation_id
+    city = body.city
+    region = body.region
+    country = body.country
+    timezone = body.timezone
+    image = body.image
+
+    async def event_generator(q: str, image: str):
         start_time = time.perf_counter()
         ttft = None
         chat_metadata: dict = {}
@@ -550,6 +571,17 @@ async def chat(
         q = unquote(q)
         nonlocal conversation_id
 
+        uploaded_image_url = None
+        if image:
+            decoded_string = unquote(image)
+            base64_data = decoded_string.split(",", 1)[1]
+            image_bytes = base64.b64decode(base64_data)
+            webp_image_bytes = convert_image_to_webp(image_bytes)
+            try:
+                uploaded_image_url = upload_image_to_bucket(webp_image_bytes, request.user.object.id)
+            except:
+                uploaded_image_url = None
+
         async def send_event(event_type: ChatEvent, data: str | dict):
             nonlocal connection_alive, ttft
             if not connection_alive or await request.is_disconnected():
@@ -637,7 +669,7 @@ async def chat(
 
         if conversation_commands == [ConversationCommand.Default] or is_automated_task:
             conversation_commands = await aget_relevant_information_sources(
-                q, meta_log, is_automated_task, subscribed=subscribed
+                q, meta_log, is_automated_task, subscribed=subscribed, uploaded_image_url=uploaded_image_url
             )
             conversation_commands_str = ", ".join([cmd.value for cmd in conversation_commands])
             async for result in send_event(
@@ -645,7 +677,7 @@ async def chat(
             ):
                 yield result
 
-            mode = await aget_relevant_output_modes(q, meta_log, is_automated_task)
+            mode = await aget_relevant_output_modes(q, meta_log, is_automated_task, uploaded_image_url)
             async for result in send_event(ChatEvent.STATUS, f"**Decided Response Mode:** {mode.value}"):
                 yield result
             if mode not in conversation_commands:
@@ -693,7 +725,9 @@ async def chat(
                     ):
                         yield result
 
-                    response = await extract_relevant_summary(q, contextual_data, subscribed=subscribed)
+                    response = await extract_relevant_summary(
+                        q, contextual_data, subscribed=subscribed, uploaded_image_url=uploaded_image_url
+                    )
                     response_log = str(response)
                     async for result in send_llm_response(response_log):
                         yield result
@@ -711,6 +745,7 @@ async def chat(
                 intent_type="summarize",
                 client_application=request.user.client_app,
                 conversation_id=conversation_id,
+                uploaded_image_url=uploaded_image_url,
             )
             return
 
@@ -753,6 +788,7 @@ async def chat(
                 conversation_id=conversation_id,
                 inferred_queries=[query_to_run],
                 automation_id=automation.id,
+                uploaded_image_url=uploaded_image_url,
             )
             async for result in send_llm_response(llm_response):
                 yield result
@@ -807,6 +843,7 @@ async def chat(
                     subscribed,
                     partial(send_event, ChatEvent.STATUS),
                     custom_filters,
+                    uploaded_image_url=uploaded_image_url,
                 ):
                     if isinstance(result, dict) and ChatEvent.STATUS in result:
                         yield result[ChatEvent.STATUS]
@@ -823,7 +860,13 @@ async def chat(
         if ConversationCommand.Webpage in conversation_commands:
             try:
                 async for result in read_webpages(
-                    defiltered_query, meta_log, location, user, subscribed, partial(send_event, ChatEvent.STATUS)
+                    defiltered_query,
+                    meta_log,
+                    location,
+                    user,
+                    subscribed,
+                    partial(send_event, ChatEvent.STATUS),
+                    uploaded_image_url=uploaded_image_url,
                 ):
                     if isinstance(result, dict) and ChatEvent.STATUS in result:
                         yield result[ChatEvent.STATUS]
@@ -869,6 +912,7 @@ async def chat(
                 online_results=online_results,
                 subscribed=subscribed,
                 send_status_func=partial(send_event, ChatEvent.STATUS),
+                uploaded_image_url=uploaded_image_url,
             ):
                 if isinstance(result, dict) and ChatEvent.STATUS in result:
                     yield result[ChatEvent.STATUS]
@@ -898,6 +942,7 @@ async def chat(
                 conversation_id=conversation_id,
                 compiled_references=compiled_references,
                 online_results=online_results,
+                uploaded_image_url=uploaded_image_url,
             )
             content_obj = {
                 "intentType": intent_type,
@@ -924,6 +969,7 @@ async def chat(
             conversation_id,
             location,
             user_name,
+            uploaded_image_url,
         )
 
         # Send Response
@@ -949,9 +995,9 @@ async def chat(
 
     ## Stream Text Response
     if stream:
-        return StreamingResponse(event_generator(q), media_type="text/plain")
+        return StreamingResponse(event_generator(q, image=image), media_type="text/plain")
     ## Non-Streaming Text Response
     else:
-        response_iterator = event_generator(q)
+        response_iterator = event_generator(q, image=image)
         response_data = await read_chat_stream(response_iterator)
         return Response(content=json.dumps(response_data), media_type="application/json", status_code=200)

khoj/routers/helpers.py

@@ -97,6 +97,7 @@ from khoj.utils.helpers import (
     LRU,
     ConversationCommand,
     ImageIntentType,
+    convert_image_to_webp,
     is_none_or_empty,
     is_valid_url,
     log_telemetry,
@@ -252,7 +253,9 @@ async def acreate_title_from_query(query: str) -> str:
     return response.strip()
 
 
-async def aget_relevant_information_sources(query: str, conversation_history: dict, is_task: bool, subscribed: bool):
+async def aget_relevant_information_sources(
+    query: str, conversation_history: dict, is_task: bool, subscribed: bool, uploaded_image_url: str = None
+):
     """
     Given a query, determine which of the available tools the agent should use in order to answer appropriately.
     """
@@ -266,6 +269,9 @@ async def aget_relevant_information_sources(query: str, conversation_history: di
 
     chat_history = construct_chat_history(conversation_history)
 
+    if uploaded_image_url:
+        query = f"[placeholder for image attached to this message]\n{query}"
+
     relevant_tools_prompt = prompts.pick_relevant_information_collection_tools.format(
         query=query,
         tools=tool_options_str,
@@ -274,7 +280,9 @@ async def aget_relevant_information_sources(query: str, conversation_history: di
 
     with timer("Chat actor: Infer information sources to refer", logger):
         response = await send_message_to_model_wrapper(
-            relevant_tools_prompt, response_type="json_object", subscribed=subscribed
+            relevant_tools_prompt,
+            response_type="json_object",
+            subscribed=subscribed,
         )
 
     try:
@@ -302,7 +310,9 @@ async def aget_relevant_information_sources(query: str, conversation_history: di
         return [ConversationCommand.Default]
 
 
-async def aget_relevant_output_modes(query: str, conversation_history: dict, is_task: bool = False):
+async def aget_relevant_output_modes(
+    query: str, conversation_history: dict, is_task: bool = False, uploaded_image_url: str = None
+):
     """
     Given a query, determine which of the available tools the agent should use in order to answer appropriately.
     """
@@ -319,6 +329,9 @@ async def aget_relevant_output_modes(query: str, conversation_history: dict, is_
 
     chat_history = construct_chat_history(conversation_history)
 
+    if uploaded_image_url:
+        query = f"[placeholder for image attached to this message]\n{query}"
+
     relevant_mode_prompt = prompts.pick_relevant_output_mode.format(
         query=query,
         modes=mode_options_str,
@@ -347,7 +360,7 @@ async def aget_relevant_output_modes(query: str, conversation_history: dict, is_
 
 
 async def infer_webpage_urls(
-    q: str, conversation_history: dict, location_data: LocationData, user: KhojUser
+    q: str, conversation_history: dict, location_data: LocationData, user: KhojUser, uploaded_image_url: str = None
 ) -> List[str]:
     """
     Infer webpage links from the given query
@@ -366,7 +379,9 @@ async def infer_webpage_urls(
     )
 
     with timer("Chat actor: Infer webpage urls to read", logger):
-        response = await send_message_to_model_wrapper(online_queries_prompt, response_type="json_object")
+        response = await send_message_to_model_wrapper(
+            online_queries_prompt, uploaded_image_url=uploaded_image_url, response_type="json_object"
+        )
 
     # Validate that the response is a non-empty, JSON-serializable list of URLs
     try:
@@ -381,7 +396,7 @@ async def infer_webpage_urls(
 
 
 async def generate_online_subqueries(
-    q: str, conversation_history: dict, location_data: LocationData, user: KhojUser
+    q: str, conversation_history: dict, location_data: LocationData, user: KhojUser, uploaded_image_url: str = None
 ) -> List[str]:
     """
     Generate subqueries from the given query
@@ -400,7 +415,9 @@ async def generate_online_subqueries(
     )
 
     with timer("Chat actor: Generate online search subqueries", logger):
-        response = await send_message_to_model_wrapper(online_queries_prompt, response_type="json_object")
+        response = await send_message_to_model_wrapper(
+            online_queries_prompt, uploaded_image_url=uploaded_image_url, response_type="json_object"
+        )
 
     # Validate that the response is a non-empty, JSON-serializable list
     try:
@@ -419,7 +436,7 @@ async def generate_online_subqueries(
         return [q]
 
 
-async def schedule_query(q: str, conversation_history: dict) -> Tuple[str, ...]:
+async def schedule_query(q: str, conversation_history: dict, uploaded_image_url: str = None) -> Tuple[str, ...]:
     """
     Schedule the date, time to run the query. Assume the server timezone is UTC.
     """
@@ -430,7 +447,9 @@ async def schedule_query(q: str, conversation_history: dict) -> Tuple[str, ...]:
         chat_history=chat_history,
     )
 
-    raw_response = await send_message_to_model_wrapper(crontime_prompt, response_type="json_object")
+    raw_response = await send_message_to_model_wrapper(
+        crontime_prompt, uploaded_image_url=uploaded_image_url, response_type="json_object"
+    )
 
     # Validate that the response is a non-empty, JSON-serializable list
     try:
@@ -468,7 +487,9 @@ async def extract_relevant_info(q: str, corpus: str, subscribed: bool) -> Union[
     return response.strip()
 
 
-async def extract_relevant_summary(q: str, corpus: str, subscribed: bool = False) -> Union[str, None]:
+async def extract_relevant_summary(
+    q: str, corpus: str, subscribed: bool = False, uploaded_image_url: str = None
+) -> Union[str, None]:
     """
     Extract relevant information for a given query from the target corpus
     """
@@ -489,6 +510,7 @@ async def extract_relevant_summary(q: str, corpus: str, subscribed: bool = False
             prompts.system_prompt_extract_relevant_summary,
             chat_model_option=chat_model,
             subscribed=subscribed,
+            uploaded_image_url=uploaded_image_url,
         )
     return response.strip()
 
@@ -501,6 +523,7 @@ async def generate_better_image_prompt(
     online_results: Optional[dict] = None,
     model_type: Optional[str] = None,
     subscribed: bool = False,
+    uploaded_image_url: Optional[str] = None,
 ) -> str:
     """
     Generate a better image prompt from the given query
@@ -549,7 +572,7 @@ async def generate_better_image_prompt(
 
     with timer("Chat actor: Generate contextual image prompt", logger):
         response = await send_message_to_model_wrapper(
-            image_prompt, chat_model_option=chat_model, subscribed=subscribed
+            image_prompt, chat_model_option=chat_model, subscribed=subscribed, uploaded_image_url=uploaded_image_url
         )
         response = response.strip()
         if response.startswith(('"', "'")) and response.endswith(('"', "'")):
@@ -564,11 +587,19 @@ async def send_message_to_model_wrapper(
     response_type: str = "text",
     chat_model_option: ChatModelOptions = None,
     subscribed: bool = False,
+    uploaded_image_url: str = None,
 ):
     conversation_config: ChatModelOptions = (
         chat_model_option or await ConversationAdapters.aget_default_conversation_config()
     )
 
+    vision_available = conversation_config.vision_enabled
+    if not vision_available and uploaded_image_url:
+        vision_enabled_config = ConversationAdapters.get_vision_enabled_config()
+        if vision_enabled_config:
+            conversation_config = vision_enabled_config
+            vision_available = True
+
     chat_model = conversation_config.chat_model
     max_tokens = (
         conversation_config.subscribed_max_prompt_size
@@ -576,6 +607,7 @@ async def send_message_to_model_wrapper(
         else conversation_config.max_prompt_size
     )
     tokenizer = conversation_config.tokenizer
+    vision_available = conversation_config.vision_enabled
 
     if conversation_config.model_type == "offline":
         if state.offline_chat_processor_config is None or state.offline_chat_processor_config.loaded_model is None:
@@ -589,6 +621,7 @@ async def send_message_to_model_wrapper(
             loaded_model=loaded_model,
             tokenizer_name=tokenizer,
             max_prompt_size=max_tokens,
+            vision_enabled=vision_available,
         )
 
         return send_message_to_model_offline(
@@ -609,6 +642,8 @@ async def send_message_to_model_wrapper(
             model_name=chat_model,
             max_prompt_size=max_tokens,
             tokenizer_name=tokenizer,
+            vision_enabled=vision_available,
+            uploaded_image_url=uploaded_image_url,
         )
 
         openai_response = send_message_to_model(
@@ -628,6 +663,7 @@ async def send_message_to_model_wrapper(
             model_name=chat_model,
             max_prompt_size=max_tokens,
             tokenizer_name=tokenizer,
+            vision_enabled=vision_available,
         )
 
         return anthropic_send_message_to_model(
@@ -651,6 +687,7 @@ def send_message_to_model_wrapper_sync(
 
     chat_model = conversation_config.chat_model
     max_tokens = conversation_config.max_prompt_size
+    vision_available = conversation_config.vision_enabled
 
     if conversation_config.model_type == "offline":
         if state.offline_chat_processor_config is None or state.offline_chat_processor_config.loaded_model is None:
@@ -658,7 +695,11 @@ def send_message_to_model_wrapper_sync(
 
         loaded_model = state.offline_chat_processor_config.loaded_model
         truncated_messages = generate_chatml_messages_with_context(
-            user_message=message, system_message=system_message, model_name=chat_model, loaded_model=loaded_model
+            user_message=message,
+            system_message=system_message,
+            model_name=chat_model,
+            loaded_model=loaded_model,
+            vision_enabled=vision_available,
         )
 
         return send_message_to_model_offline(
@@ -672,7 +713,10 @@ def send_message_to_model_wrapper_sync(
     elif conversation_config.model_type == "openai":
         api_key = conversation_config.openai_config.api_key
         truncated_messages = generate_chatml_messages_with_context(
-            user_message=message, system_message=system_message, model_name=chat_model
+            user_message=message,
+            system_message=system_message,
+            model_name=chat_model,
+            vision_enabled=vision_available,
         )
 
         openai_response = send_message_to_model(
@@ -688,6 +732,7 @@ def send_message_to_model_wrapper_sync(
             system_message=system_message,
             model_name=chat_model,
             max_prompt_size=max_tokens,
+            vision_enabled=vision_available,
         )
 
         return anthropic_send_message_to_model(
@@ -712,6 +757,7 @@ def generate_chat_response(
     conversation_id: int = None,
     location_data: LocationData = None,
     user_name: Optional[str] = None,
+    uploaded_image_url: Optional[str] = None,
 ) -> Tuple[Union[ThreadedGenerator, Iterator[str]], Dict[str, str]]:
     # Initialize Variables
     chat_response = None
@@ -719,7 +765,6 @@ def generate_chat_response(
 
     metadata = {}
     agent = AgentAdapters.get_conversation_agent_by_id(conversation.agent.id) if conversation.agent else None
-
     try:
         partial_completion = partial(
             save_to_conversation_log,
@@ -731,9 +776,17 @@ def generate_chat_response(
             inferred_queries=inferred_queries,
             client_application=client_application,
             conversation_id=conversation_id,
+            uploaded_image_url=uploaded_image_url,
         )
 
         conversation_config = ConversationAdapters.get_valid_conversation_config(user, conversation)
+        vision_available = conversation_config.vision_enabled
+        if not vision_available and uploaded_image_url:
+            vision_enabled_config = ConversationAdapters.get_vision_enabled_config()
+            if vision_enabled_config:
+                conversation_config = vision_enabled_config
+                vision_available = True
+
         if conversation_config.model_type == "offline":
             loaded_model = state.offline_chat_processor_config.loaded_model
             chat_response = converse_offline(
@@ -759,6 +812,7 @@ def generate_chat_response(
             chat_response = converse(
                 compiled_references,
                 q,
+                image_url=uploaded_image_url,
                 online_results=online_results,
                 conversation_log=meta_log,
                 model=chat_model,
@@ -771,6 +825,7 @@ def generate_chat_response(
                 location_data=location_data,
                 user_name=user_name,
                 agent=agent,
+                vision_available=vision_available,
             )
 
         elif conversation_config.model_type == "anthropic":
@@ -809,6 +864,7 @@ async def text_to_image(
     online_results: Dict[str, Any],
     subscribed: bool = False,
     send_status_func: Optional[Callable] = None,
+    uploaded_image_url: Optional[str] = None,
 ):
     status_code = 200
     image = None
@@ -845,6 +901,7 @@ async def text_to_image(
         online_results=online_results,
         model_type=text_to_image_config.model_type,
         subscribed=subscribed,
+        uploaded_image_url=uploaded_image_url,
     )
 
     if send_status_func:
@@ -908,13 +965,7 @@ async def text_to_image(
 
     with timer("Convert image to webp", logger):
         # Convert png to webp for faster loading
-        image_io = io.BytesIO(decoded_image)
-        png_image = Image.open(image_io)
-        webp_image_io = io.BytesIO()
-        png_image.save(webp_image_io, "WEBP")
-        webp_image_bytes = webp_image_io.getvalue()
-        webp_image_io.close()
-        image_io.close()
+        webp_image_bytes = convert_image_to_webp(decoded_image)
 
     with timer("Upload image to S3", logger):
         image_url = upload_image(webp_image_bytes, user.uuid)
@@ -1095,6 +1146,7 @@ def should_notify(original_query: str, executed_query: str, ai_response: str) ->
 
     with timer("Chat actor: Decide to notify user of automation response", logger):
         try:
+            # TODO Replace with async call so we don't have to maintain a sync version
             response = send_message_to_model_wrapper_sync(to_notify_or_not)
             should_notify_result = "no" not in response.lower()
             logger.info(f'Decided to {"not " if not should_notify_result else ""}notify user of automation response.')

khoj/routers/storage.py

@@ -33,3 +33,31 @@ def upload_image(image: bytes, user_id: uuid.UUID):
     except Exception as e:
         logger.error(f"Failed to upload image to S3: {e}")
         return None
+
+
+AWS_USER_UPLOADED_IMAGES_BUCKET_NAME = os.getenv("AWS_USER_UPLOADED_IMAGES_BUCKET_NAME")
+
+
+def upload_image_to_bucket(image: bytes, user_id: uuid.UUID):
+    """Upload the image to the S3 bucket"""
+    if not aws_enabled:
+        logger.info("AWS is not enabled. Skipping image upload")
+        return None
+
+    image_key = f"{user_id}/{uuid.uuid4()}.webp"
+    if not AWS_USER_UPLOADED_IMAGES_BUCKET_NAME:
+        logger.error("AWS_USER_UPLOADED_IMAGES_BUCKET_NAME is not set")
+        return None
+
+    try:
+        s3_client.put_object(
+            Bucket=AWS_USER_UPLOADED_IMAGES_BUCKET_NAME,
+            Key=image_key,
+            Body=image,
+            ACL="public-read",
+            ContentType="image/webp",
+        )
+        return f"https://{AWS_USER_UPLOADED_IMAGES_BUCKET_NAME}/{image_key}"
+    except Exception as e:
+        logger.error(f"Failed to upload image to S3: {e}")
+        return None

khoj/utils/helpers.py

@@ -1,6 +1,7 @@
 from __future__ import annotations  # to avoid quoting type hints
 
 import datetime
+import io
 import logging
 import os
 import platform
@@ -22,6 +23,7 @@ import requests
 import torch
 from asgiref.sync import sync_to_async
 from magika import Magika
+from PIL import Image
 
 from khoj.utils import constants
 
@@ -416,3 +418,16 @@ def is_internet_connected():
         return response.status_code == 200
     except:
         return False
+
+
+def convert_image_to_webp(image_bytes):
+    """Convert image bytes to webp format for faster loading"""
+    image_io = io.BytesIO(image_bytes)
+    with Image.open(image_io) as original_image:
+        webp_image_io = io.BytesIO()
+        original_image.save(webp_image_io, "WEBP")
+
+        # Encode the WebP image back to base64
+        webp_image_bytes = webp_image_io.getvalue()
+        webp_image_io.close()
+        return webp_image_bytes

{khoj-1.21.6.dev14.dist-info → khoj-1.21.7.dev4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: khoj
-Version: 1.21.6.dev14
+Version: 1.21.7.dev4
 Summary: Your Second Brain
 Project-URL: Homepage, https://khoj.dev
 Project-URL: Documentation, https://docs.khoj.dev