khoj 2.0.0b13.dev23__py3-none-any.whl → 2.0.0b14__py3-none-any.whl

khoj/processor/conversation/openai/utils.py

@@ -78,10 +78,8 @@ def _extract_text_for_instructions(content: Union[str, List, Dict, None]) -> str
 @retry(
     retry=(
         retry_if_exception_type(openai._exceptions.APITimeoutError)
-        | retry_if_exception_type(openai._exceptions.APIError)
-        | retry_if_exception_type(openai._exceptions.APIConnectionError)
         | retry_if_exception_type(openai._exceptions.RateLimitError)
-        | retry_if_exception_type(openai._exceptions.APIStatusError)
+        | retry_if_exception_type(openai._exceptions.InternalServerError)
         | retry_if_exception_type(ValueError)
     ),
     wait=wait_random_exponential(min=1, max=10),
@@ -128,7 +126,7 @@ def completion_with_backoff(
         if model_name.startswith("grok-4"):
             # Grok-4 models do not support reasoning_effort parameter
             model_kwargs.pop("reasoning_effort", None)
-    elif model_name.startswith("deepseek-reasoner"):
+    elif model_name.startswith("deepseek-reasoner") or model_name.startswith("deepseek-chat"):
         stream_processor = in_stream_thought_processor
         # Two successive messages cannot be from the same role. Should merge any back-to-back messages from the same role.
         # The first message should always be a user message (except system message).
@@ -147,6 +145,8 @@ def completion_with_backoff(
         # See https://qwenlm.github.io/blog/qwen3/#advanced-usages
         if not deepthought:
             add_qwen_no_think_tag(formatted_messages)
+    elif is_groq_api(api_base_url):
+        model_kwargs["service_tier"] = "auto"
 
     read_timeout = 300 if is_local_api(api_base_url) else 60
     if os.getenv("KHOJ_LLM_SEED"):
@@ -172,8 +172,16 @@ def completion_with_backoff(
                     chunk.type == "chunk"
                     and chunk.chunk.choices
                     and hasattr(chunk.chunk.choices[0].delta, "reasoning_content")
+                    and chunk.chunk.choices[0].delta.reasoning_content
                 ):
                     thoughts += chunk.chunk.choices[0].delta.reasoning_content
+                elif (
+                    chunk.type == "chunk"
+                    and chunk.chunk.choices
+                    and hasattr(chunk.chunk.choices[0].delta, "reasoning")
+                    and chunk.chunk.choices[0].delta.reasoning
+                ):
+                    thoughts += chunk.chunk.choices[0].delta.reasoning
                 elif chunk.type == "chunk" and chunk.chunk.choices and chunk.chunk.choices[0].delta.tool_calls:
                     tool_ids += [tool_call.id for tool_call in chunk.chunk.choices[0].delta.tool_calls]
                 elif chunk.type == "tool_calls.function.arguments.done":
@@ -196,7 +204,6 @@ def completion_with_backoff(
         chunk = client.beta.chat.completions.parse(
             messages=formatted_messages,  # type: ignore
             model=model_name,
-            temperature=temperature,
             timeout=httpx.Timeout(30, read=read_timeout),
             **model_kwargs,
         )
@@ -221,6 +228,10 @@ def completion_with_backoff(
             # Json dump tool calls into aggregated response
             aggregated_response = json.dumps([tool_call.__dict__ for tool_call in tool_calls])
 
+    # Align chunk definition with non-streaming mode for post stream completion usage
+    if hasattr(chunk, "chunk"):
+        chunk = chunk.chunk
+
     # Calculate cost of chat
     input_tokens = chunk.usage.prompt_tokens if hasattr(chunk, "usage") and chunk.usage else 0
     output_tokens = chunk.usage.completion_tokens if hasattr(chunk, "usage") and chunk.usage else 0
@@ -249,10 +260,8 @@ def completion_with_backoff(
 @retry(
     retry=(
         retry_if_exception_type(openai._exceptions.APITimeoutError)
-        | retry_if_exception_type(openai._exceptions.APIError)
-        | retry_if_exception_type(openai._exceptions.APIConnectionError)
         | retry_if_exception_type(openai._exceptions.RateLimitError)
-        | retry_if_exception_type(openai._exceptions.APIStatusError)
+        | retry_if_exception_type(openai._exceptions.InternalServerError)
         | retry_if_exception_type(ValueError)
     ),
     wait=wait_exponential(multiplier=1, min=4, max=10),
@@ -313,8 +322,12 @@ async def chat_completion_with_backoff(
         # Grok-4 models do not support reasoning_effort parameter
         if not model_name.startswith("grok-4"):
             model_kwargs["reasoning_effort"] = reasoning_effort
-    elif model_name.startswith("deepseek-reasoner") or "deepseek-r1" in model_name:
-        # Official Deepseek reasoner model and some inference APIs like vLLM return structured thinking output.
+    elif (
+        model_name.startswith("deepseek-chat")
+        or model_name.startswith("deepseek-reasoner")
+        or "deepseek-r1" in model_name
+    ):
+        # Official Deepseek models and some inference APIs like vLLM return structured thinking output.
         # Others like DeepInfra return it in response stream.
         # Using the instream thought processor handles both cases, structured thoughts and in response thoughts.
         stream_processor = ain_stream_thought_processor
@@ -339,6 +352,8 @@ async def chat_completion_with_backoff(
         # See https://qwenlm.github.io/blog/qwen3/#advanced-usages
         if not deepthought:
             add_qwen_no_think_tag(formatted_messages)
+    elif is_groq_api(api_base_url):
+        model_kwargs["service_tier"] = "auto"
 
     read_timeout = 300 if is_local_api(api_base_url) else 60
     if os.getenv("KHOJ_LLM_SEED"):
@@ -415,10 +430,8 @@ async def chat_completion_with_backoff(
 @retry(
     retry=(
         retry_if_exception_type(openai._exceptions.APITimeoutError)
-        | retry_if_exception_type(openai._exceptions.APIError)
-        | retry_if_exception_type(openai._exceptions.APIConnectionError)
         | retry_if_exception_type(openai._exceptions.RateLimitError)
-        | retry_if_exception_type(openai._exceptions.APIStatusError)
+        | retry_if_exception_type(openai._exceptions.InternalServerError)
         | retry_if_exception_type(ValueError)
     ),
     wait=wait_random_exponential(min=1, max=10),
@@ -460,6 +473,7 @@ def responses_completion_with_backoff(
         temperature = 1
         reasoning_effort = "medium" if deepthought else "low"
         model_kwargs["reasoning"] = {"effort": reasoning_effort, "summary": "auto"}
+        model_kwargs["include"] = ["reasoning.encrypted_content"]
         # Remove unsupported params for reasoning models
         model_kwargs.pop("top_p", None)
         model_kwargs.pop("stop", None)
@@ -474,7 +488,6 @@ def responses_completion_with_backoff(
         temperature=temperature,
         timeout=httpx.Timeout(30, read=read_timeout),  # type: ignore
         store=False,
-        include=["reasoning.encrypted_content"],
         **model_kwargs,
     )
     if not model_response or not isinstance(model_response, OpenAIResponse) or not model_response.output:
@@ -529,10 +542,8 @@ def responses_completion_with_backoff(
 @retry(
     retry=(
         retry_if_exception_type(openai._exceptions.APITimeoutError)
-        | retry_if_exception_type(openai._exceptions.APIError)
-        | retry_if_exception_type(openai._exceptions.APIConnectionError)
         | retry_if_exception_type(openai._exceptions.RateLimitError)
-        | retry_if_exception_type(openai._exceptions.APIStatusError)
+        | retry_if_exception_type(openai._exceptions.InternalServerError)
         | retry_if_exception_type(ValueError)
     ),
     wait=wait_exponential(multiplier=1, min=4, max=10),
@@ -764,7 +775,7 @@ def format_message_for_api(raw_messages: List[ChatMessage], api_base_url: str) -
                         {
                             "type": "function_call_output",
                             "call_id": tool_call_id,
-                            "output": part.get("content"),
+                            "output": part.get("content") or "No output",
                         }
                     )
                 else:
@@ -773,7 +784,7 @@ def format_message_for_api(raw_messages: List[ChatMessage], api_base_url: str) -
                             "role": "tool",
                             "tool_call_id": tool_call_id,
                             "name": part.get("name"),
-                            "content": part.get("content"),
+                            "content": part.get("content") or "No output",
                         }
                     )
             continue
@@ -810,8 +821,9 @@ def format_message_for_api(raw_messages: List[ChatMessage], api_base_url: str) -
                 # OpenAI models use the Responses API which uses slightly different content types
                 if part["type"] == "text":
                     part["type"] = "output_text" if message.role == "assistant" else "input_text"
-                if part["type"] == "image":
+                if part["type"] == "image_url":
                     part["type"] = "output_image" if message.role == "assistant" else "input_image"
+                    part["image_url"] = part["image_url"]["url"]
             # If no valid content parts left, remove the message
             if is_none_or_empty(message.content):
                 messages.remove(message)
@@ -836,8 +848,10 @@ def is_openai_reasoning_model(model_name: str, api_base_url: str = None) -> bool
     """
     Check if the model is an OpenAI reasoning model
     """
-    return is_openai_api(api_base_url) and (
-        model_name.lower().startswith("o") or model_name.lower().startswith("gpt-5")
+    return (
+        is_openai_api(api_base_url)
+        and (model_name.lower().startswith("o") or model_name.lower().startswith("gpt-5"))
+        or model_name.lower().startswith("gpt-oss")
     )
 
 
@@ -861,6 +875,13 @@ def is_twitter_reasoning_model(model_name: str, api_base_url: str = None) -> boo
     )
 
 
+def is_groq_api(api_base_url: str = None) -> bool:
+    """
+    Check if the model is served over the Groq API
+    """
+    return api_base_url is not None and api_base_url.startswith("https://api.groq.com")
+
+
 def is_qwen_style_reasoning_model(model_name: str, api_base_url: str = None) -> bool:
     """
     Check if the model is a Qwen style reasoning model
@@ -934,6 +955,9 @@ async def astream_thought_processor(
             if not chunk_data.get("object") or chunk_data.get("object") != "chat.completion.chunk":
                 logger.warning(f"Skipping invalid chunk with object field: {chunk_data.get('object', 'missing')}")
                 continue
+            # Handle unsupported service tiers like "on_demand" by Groq
+            if chunk.service_tier and chunk.service_tier == "on_demand":
+                chunk_data["service_tier"] = "auto"
 
             tchunk = ChatCompletionWithThoughtsChunk.model_validate(chunk_data)
 
@@ -945,6 +969,14 @@ async def astream_thought_processor(
             ):
                 tchunk.choices[0].delta.thought = chunk.choices[0].delta.reasoning_content
 
+            # Handlle openai reasoning style response with thoughts. Used by gpt-oss.
+            if (
+                len(tchunk.choices) > 0
+                and hasattr(tchunk.choices[0].delta, "reasoning")
+                and tchunk.choices[0].delta.reasoning
+            ):
+                tchunk.choices[0].delta.thought = chunk.choices[0].delta.reasoning
+
             # Handlle llama.cpp server style response with thoughts.
             elif len(tchunk.choices) > 0 and tchunk.choices[0].delta.model_extra.get("reasoning_content"):
                 tchunk.choices[0].delta.thought = tchunk.choices[0].delta.model_extra.get("reasoning_content")
@@ -1075,6 +1107,10 @@ async def ain_stream_thought_processor(
             yield chunk
             continue
 
+        if chunk.choices[0].delta.content is None:
+            # If delta content is None, we can't process it, just yield the chunk
+            continue
+
         buf += chunk.choices[0].delta.content
 
         if mode == "detect_start":

khoj/processor/conversation/prompts.py

@@ -4,21 +4,27 @@ from langchain_core.prompts import PromptTemplate
 ## --
 personality = PromptTemplate.from_template(
     """
-You are Khoj, a smart, inquisitive and helpful personal assistant.
+You are Khoj, a smart, curious, empathetic and helpful personal assistant.
 Use your general knowledge and past conversation with the user as context to inform your responses.
-You were created by Khoj Inc. with the following capabilities:
 
-- You *CAN REMEMBER ALL NOTES and PERSONAL INFORMATION FOREVER* that the user ever shares with you.
-- Users can share files and other information with you using the Khoj Desktop, Obsidian or Emacs app. They can also drag and drop their files into the chat window.
-- You *CAN* generate images, look-up real-time information from the internet, set reminders and answer questions based on the user's notes.
+You were created by Khoj Inc. More information about you, the company or Khoj apps can be found at https://khoj.dev.
+
+Today is {day_of_week}, {current_date} in UTC.
+
+# Capabilities
+- Users can share files and other information with you using the Khoj Web, Desktop, Obsidian or Emacs app. They can also drag and drop their files into the chat window.
+- You can look up information from the user's notes and documents synced via the Khoj apps.
+- You can generate images, look-up real-time information from the internet, analyze data and answer questions based on the user's notes.
+
+# Style
+- Your responses should be helpful, conversational and tuned to the user's communication style.
 - Make sure to use the specific LaTeX math mode delimiters for your response. LaTex math mode specific delimiters as following
     - inline math mode : \\( and \\)
     - display math mode: insert linebreak after opening $$, \\[ and before closing $$, \\]
-- Sometimes the user will share personal information that needs to be remembered, like an account ID or a residential address. These can be acknowledged with a simple "Got it" or "Okay".
-- Provide inline references to quotes from the user's notes or any web pages you refer to in your responses in markdown format. For example, "The farmer had ten sheep. [1](https://example.com)". *ALWAYS CITE YOUR SOURCES AND PROVIDE REFERENCES*. Add them inline to directly support your claim.
-
-Note: More information about you, the company or Khoj apps can be found at https://khoj.dev.
-Today is {day_of_week}, {current_date} in UTC.
+- Provide inline citations to documents and websites referenced. Add them inline in markdown format to directly support your claim.
+  For example: "The weather today is sunny [1](https://weather.com)."
+- Mention generated assets like images by reference, e.g ![chart](/visualization/image.png). Do not manually output raw, b64 encoded bytes in your response.
+- Do not respond with raw programs or scripts in your final response unless you know the user is a programmer or has explicitly requested code.
 """.strip()
 )
 
@@ -26,18 +32,23 @@ custom_personality = PromptTemplate.from_template(
     """
 You are {name}, a personal agent on Khoj.
 Use your general knowledge and past conversation with the user as context to inform your responses.
-You were created by Khoj Inc. with the following capabilities:
 
-- You *CAN REMEMBER ALL NOTES and PERSONAL INFORMATION FOREVER* that the user ever shares with you.
-- Users can share files and other information with you using the Khoj Desktop, Obsidian or Emacs app. They can also drag and drop their files into the chat window.
+You were created on the Khoj platform. More information about you, the company or Khoj apps can be found at https://khoj.dev.
+
+Today is {day_of_week}, {current_date} in UTC.
+
+# Base Capabilities
+- Users can share files and other information with you using the Khoj Web, Desktop, Obsidian or Emacs app. They can also drag and drop their files into the chat window.
+
+# Style
 - Make sure to use the specific LaTeX math mode delimiters for your response. LaTex math mode specific delimiters as following
     - inline math mode : `\\(` and `\\)`
     - display math mode: insert linebreak after opening `$$`, `\\[` and before closing `$$`, `\\]`
-- Sometimes the user will share personal information that needs to be remembered, like an account ID or a residential address. These can be acknowledged with a simple "Got it" or "Okay".
+- Provide inline citations to documents and websites referenced. Add them inline in markdown format to directly support your claim.
+  For example: "The weather today is sunny [1](https://weather.com)."
+- Mention generated assets like images by reference, e.g ![chart](/visualization/image.png). Do not manually output raw, b64 encoded bytes in your response.
 
-Today is {day_of_week}, {current_date} in UTC.
-
-Instructions:\n{bio}
+# Instructions:\n{bio}
 """.strip()
 )
 
@@ -641,16 +652,17 @@ Here's some additional context about you:
 
 plan_function_execution = PromptTemplate.from_template(
     """
-You are Khoj, a smart, creative and meticulous researcher. Use the provided tool AIs to accomplish the task assigned to you.
+You are Khoj, a smart, creative and meticulous researcher.
 Create a multi-step plan and intelligently iterate on the plan to complete the task.
+Use the help of the provided tool AIs to accomplish the task assigned to you.
 {personality_context}
 
 # Instructions
-- Provide highly diverse, detailed requests to the tool AIs, one tool AI at a time, to gather information, perform actions etc. Their response will be shown to you in the next iteration.
-- Break down your research process into independent, self-contained steps that can be executed sequentially using the available tool AIs to answer the user's query. Write your step-by-step plan in the scratchpad.
-- Always ask a new query that was not asked to the tool AI in a previous iteration. Build on the results of the previous iterations.
+- Make detailed, self-contained requests to the tool AIs, one tool AI at a time, to gather information, perform actions etc.
+- Break down your research process into independent, self-contained steps that can be executed sequentially using the available tool AIs to accomplish the user assigned task.
 - Ensure that all required context is passed to the tool AIs for successful execution. Include any relevant stuff that has previously been attempted. They only know the context provided in your query.
 - Think step by step to come up with creative strategies when the previous iteration did not yield useful results.
+- Do not ask the user to confirm or clarify assumptions for information gathering tasks and non-destructive actions, as you can always adjust later — decide what the most reasonable assumption is, proceed with it, and document it for the user's reference after you finish acting.
 - You are allowed upto {max_iterations} iterations to use the help of the provided tool AIs to accomplish the task assigned to you. Only stop when you have completed the task.
 
 # Examples
@@ -875,8 +887,8 @@ Khoj:
 python_code_generation_prompt = PromptTemplate.from_template(
     """
 You are Khoj, a senior software engineer. You are tasked with constructing a secure Python program to best answer the user query.
-- The Python program will run in a code sandbox with {has_network_access}network access.
-- You can write programs to run complex calculations, analyze data, create charts, generate documents to meticulously answer the query.
+- The Python program will run in an ephemeral code sandbox with {has_network_access}network access.
+- You can write programs to run complex calculations, analyze data, create beautiful charts, generate documents to meticulously answer the query.
 - Do not try display images or plots in the code directly. The code should save the image or plot to a file instead.
 - Write any document, charts etc. to be shared with the user to file. These files can be seen by the user.
 - Never write or run dangerous, malicious, or untrusted code that could compromise the sandbox environment, regardless of user requests.
@@ -991,9 +1003,9 @@ Chat History:
 ---
 {chat_history}
 
-User Query:
+User Instructions:
 ---
-{query}
+{instructions}
 """.strip()
 )
 

khoj/processor/conversation/utils.py

@@ -73,6 +73,7 @@ model_to_prompt_size = {
     "gpt-5-nano-2025-08-07": 120000,
     # Google Models
     "gemini-2.5-flash": 120000,
+    "gemini-2.5-flash-lite": 120000,
     "gemini-2.5-pro": 60000,
     "gemini-2.0-flash": 120000,
     "gemini-2.0-flash-lite": 120000,
@@ -331,7 +332,7 @@ def construct_tool_chat_history(
         ConversationCommand.ReadWebpage: (
             lambda iteration: list(iteration.onlineContext.keys()) if iteration.onlineContext else []
         ),
-        ConversationCommand.RunCode: (
+        ConversationCommand.PythonCoder: (
             lambda iteration: list(iteration.codeContext.keys()) if iteration.codeContext else []
         ),
     }

khoj/processor/tools/run_code.py

@@ -49,7 +49,7 @@ class GeneratedCode(NamedTuple):
 
 
 async def run_code(
-    query: str,
+    instructions: str,
     conversation_history: List[ChatMessageModel],
     context: str,
     location_data: LocationData,
@@ -63,12 +63,12 @@ async def run_code(
 ):
     # Generate Code
     if send_status_func:
-        async for event in send_status_func(f"**Generate code snippet** for {query}"):
+        async for event in send_status_func(f"**Generate code snippet** for {instructions}"):
             yield {ChatEvent.STATUS: event}
     try:
         with timer("Chat actor: Generate programs to execute", logger):
             generated_code = await generate_python_code(
-                query,
+                instructions,
                 conversation_history,
                 context,
                 location_data,
@@ -79,7 +79,7 @@ async def run_code(
                 query_files,
             )
     except Exception as e:
-        raise ValueError(f"Failed to generate code for {query} with error: {e}")
+        raise ValueError(f"Failed to generate code for {instructions} with error: {e}")
 
     # Prepare Input Data
     input_data = []
@@ -101,21 +101,21 @@ async def run_code(
             code = result.pop("code")
             cleaned_result = truncate_code_context({"cleaned": {"results": result}})["cleaned"]["results"]
             logger.info(f"Executed Code\n----\n{code}\n----\nResult\n----\n{cleaned_result}\n----")
-            yield {query: {"code": code, "results": result}}
+            yield {instructions: {"code": code, "results": result}}
     except asyncio.TimeoutError as e:
         # Call the sandbox_url/stop GET API endpoint to stop the code sandbox
-        error = f"Failed to run code for {query} with Timeout error: {e}"
+        error = f"Failed to run code for {instructions} with Timeout error: {e}"
         try:
             await aiohttp.ClientSession().get(f"{sandbox_url}/stop", timeout=5)
         except Exception as e:
             error += f"\n\nFailed to stop code sandbox with error: {e}"
         raise ValueError(error)
     except Exception as e:
-        raise ValueError(f"Failed to run code for {query} with error: {e}")
+        raise ValueError(f"Failed to run code for {instructions} with error: {e}")
 
 
 async def generate_python_code(
-    q: str,
+    instructions: str,
     chat_history: List[ChatMessageModel],
     context: str,
     location_data: LocationData,
@@ -142,7 +142,7 @@ async def generate_python_code(
     network_access_context = "**NO** " if not is_e2b_code_sandbox_enabled() else ""
 
     code_generation_prompt = prompts.python_code_generation_prompt.format(
-        query=q,
+        instructions=instructions,
         chat_history=chat_history_str,
         context=context,
         has_network_access=network_access_context,
@@ -252,8 +252,12 @@ async def execute_e2b(code: str, input_files: list[dict]) -> dict[str, Any]:
 
         # Identify new files created during execution
         new_files = set(E2bFile(f.name, f.path) for f in await sandbox.files.list("~")) - original_files
+
         # Read newly created files in parallel
-        download_tasks = [sandbox.files.read(f.path, request_timeout=30) for f in new_files]
+        def read_format(f):
+            return "bytes" if Path(f.name).suffix in image_file_ext else "text"
+
+        download_tasks = [sandbox.files.read(f.path, format=read_format(f), request_timeout=30) for f in new_files]
         downloaded_files = await asyncio.gather(*download_tasks)
         for f, content in zip(new_files, downloaded_files):
             if isinstance(content, bytes):
@@ -261,23 +265,12 @@ async def execute_e2b(code: str, input_files: list[dict]) -> dict[str, Any]:
                 b64_data = base64.b64encode(content).decode("utf-8")
             elif Path(f.name).suffix in image_file_ext:
                 # Ignore image files as they are extracted from execution results below for inline display
-                continue
+                b64_data = base64.b64encode(content).decode("utf-8")
             else:
                 # Text files - encode utf-8 string as base64
                 b64_data = content
             output_files.append({"filename": f.name, "b64_data": b64_data})
 
-        # Collect output files from execution results
-        # Repect ordering of output result types to disregard text output associated with images
-        output_result_types = ["png", "jpeg", "svg", "text", "markdown", "json"]
-        for idx, result in enumerate(execution.results):
-            if getattr(result, "chart", None):
-                continue
-            for result_type in output_result_types:
-                if b64_data := getattr(result, result_type, None):
-                    output_files.append({"filename": f"{idx}.{result_type}", "b64_data": b64_data})
-                    break
-
         # collect logs
         success = not execution.error and not execution.logs.stderr
         stdout = "\n".join(execution.logs.stdout)

khoj/routers/api_chat.py

@@ -786,6 +786,9 @@ async def event_generator(
                     if interrupt_query == ChatEvent.END_EVENT.value:
                         cancellation_event.set()
                         logger.debug(f"Chat cancelled by user {user} via interrupt queue.")
+                    elif interrupt_query == ChatEvent.INTERRUPT.value:
+                        cancellation_event.set()
+                        logger.debug("Chat interrupted.")
                     else:
                         # Pass the interrupt query to child tasks
                         logger.info(f"Continuing chat with the new instruction: {interrupt_query}")
@@ -995,7 +998,7 @@ async def event_generator(
             )
         except ValueError as e:
             logger.error(f"Error getting data sources and output format: {e}. Falling back to default.")
-            conversation_commands = [ConversationCommand.General]
+            chosen_io = {"sources": [ConversationCommand.General], "output": ConversationCommand.Text}
 
         conversation_commands = chosen_io.get("sources") + [chosen_io.get("output")]
 
@@ -1523,6 +1526,8 @@ async def chat_ws(
                         ack_type = "interrupt_acknowledged"
                         await websocket.send_text(json.dumps({"type": ack_type}))
                 else:
+                    ack_type = "interrupt_acknowledged"
+                    await websocket.send_text(json.dumps({"type": ack_type}))
                     logger.info(f"No ongoing task to interrupt for user {websocket.scope['user'].object.id}")
                 continue
 
@@ -1556,7 +1561,7 @@ async def chat_ws(
     except WebSocketDisconnect:
         logger.info(f"WebSocket disconnected for user {websocket.scope['user'].object.id}")
         if current_task and not current_task.done():
-            current_task.cancel()
+            interrupt_queue.put_nowait(ChatEvent.INTERRUPT.value)
     except Exception as e:
         logger.error(f"Error in websocket chat: {e}", exc_info=True)
         if current_task and not current_task.done():
@@ -1701,8 +1706,8 @@ async def process_chat_request(
         logger.debug(f"Chat request cancelled for user {websocket.scope['user'].object.id}")
         raise
     except Exception as e:
-        logger.error(f"Error processing chat request: {e}", exc_info=True)
         await websocket.send_text(json.dumps({"error": "Internal server error"}))
+        logger.error(f"Error processing chat request: {e}", exc_info=True)
         raise
 
 

khoj/routers/api_content.py

@@ -570,7 +570,7 @@ async def indexer(
                     file_data.content.decode(file_data.encoding) if file_data.encoding else file_data.content
                 )
             else:
-                logger.warning(f"Skipped indexing unsupported file type sent by {client} client: {file_data.name}")
+                logger.debug(f"Skipped indexing unsupported file type sent by {client} client: {file_data.name}")
 
         indexer_input = IndexerInput(
             org=index_files["org"],