Update run.py

run.py

@@ -48,8 +48,11 @@ class StreamingHandler(logging.Handler):
     
     def emit(self, record):
         msg = self.format(record)
-        for callback in self.callbacks:
-            callback(msg) # Do not append newline here, let the callback format if needed
+        # Check if the message is actually different or non-empty after stripping
+        # to avoid sending redundant empty strings, though `highlight_text` in app.py handles empty.
+        if msg.strip(): 
+            for callback in self.callbacks:
+                callback(msg + '\n') # Add newline to ensure distinct lines are processed by app.py's splitter
 
 
 class StreamingCapture(StringIO):
@@ -57,27 +60,19 @@ class StreamingCapture(StringIO):
     def __init__(self):
         super().__init__()
         self.callbacks = []
-        self._buffer = ""
     
     def add_callback(self, callback):
         self.callbacks.append(callback)
     
     def write(self, s):
-        super().write(s)
-        self._buffer += s
-        # Flush the buffer line by line or after significant chunks
-        while '\n' in self._buffer:
-            line, self._buffer = self._buffer.split('\n', 1)
-            if line.strip(): # Only send non-empty lines
-                for callback in self.callbacks:
-                    callback(line + '\n') # Add newline back for complete lines
-    
-    def flush(self):
-        # Ensure any remaining content in the buffer is sent
-        if self._buffer.strip():
+        # Pass the raw string 's' directly to callbacks immediately
+        if s: # Only send if there's actual content
             for callback in self.callbacks:
-                callback(self._buffer)
-            self._buffer = ""
+                callback(s)
+        super().write(s) # Still write to the underlying StringIO buffer
+
+
+    def flush(self):
         super().flush()
 
 
@@ -123,7 +118,7 @@ def create_agent(
         print("[DEBUG] Using default API endpoint:", api_endpoint)
         model_params["base_url"] = api_endpoint
         model_params["api_key"] = openai_api_key
-
+    # It's important that if an API key is missing for the chosen model, it fails here or upstream.
 
     model = LiteLLMModel(**model_params)
     print("[DEBUG] Model initialized")
@@ -147,23 +142,22 @@ def create_agent(
 
     search_tool = None
     if search_provider == "searxng":
-        print("[DEBUG] Using SearxNG-compatible DuckDuckGoSearchTool with base_url override")
+        print("[DEBUG] Using DuckDuckGoSearchTool (acting as a generic web search) for SearxNG context.")
         search_tool = DuckDuckGoSearchTool()
         if custom_search_url:
-            # Note: DuckDuckGoSearchTool itself does not inherently support a 'base_url' for SearxNG directly.
-            # This would typically require modifying the tool or using a custom tool implementation.
-            # For this example, we'll assume the DuckDuckGoSearchTool could be configured or subclassed
-            # to point to a SearxNG instance if its underlying search mechanism allowed.
-            # As a workaround, if SearxNG is chosen, ensure the tool selected is compatible.
-            # For simplicity, we keep DuckDuckGoSearchTool, but in a real scenario,
-            # you'd use a dedicated SearxNG tool.
-            search_tool.base_url = custom_search_url # This line might not have effect without custom logic in DuckDuckGoSearchTool
+            # Note: As mentioned before, DuckDuckGoSearchTool doesn't natively use a custom base_url
+            # for a completely different search engine like SearxNG. This line will likely have no effect.
+            # For true SearxNG integration, you'd need a custom tool or a modified DuckDuckGoSearchTool
+            # that knows how to query SearxNG instances.
+            print(f"[WARNING] DuckDuckGoSearchTool does not directly support 'custom_search_url' for SearxNG. Consider a dedicated SearxNG tool.")
+            # search_tool.base_url = custom_search_url # This line is often not effective for DDCSTool
     elif search_provider == "serper":
-        print("[DEBUG] Using DuckDuckGoSearchTool for Serper search functionality (requires SerpAPI key)")
-        search_tool = DuckDuckGoSearchTool() # DuckDuckGoSearchTool for general web search
-        # If Serper API key is explicitly provided, it should be passed to a Serper-specific tool
-        # For this setup, SerpAPI key is already in browser_config for ArchiveSearchTool,
-        # but a dedicated Serper search tool might be needed if not using SerpAPI with the browser directly.
+        print("[DEBUG] Using DuckDuckGoSearchTool (acting as a generic web search) for Serper context.")
+        search_tool = DuckDuckGoSearchTool() # You would need a separate SerperTool for direct Serper API calls.
+        if search_api_key:
+            print("[DEBUG] Serper API Key provided. Ensure your search tool (if custom) uses it.")
+            # If you had a dedicated SerperTool, you'd pass search_api_key to it.
+            # e.g., search_tool = SerperTool(api_key=search_api_key)
     else:
         print("[DEBUG] No specific search provider selected, or provider not directly supported. Defaulting to DuckDuckGoSearchTool.")
         search_tool = DuckDuckGoSearchTool()
@@ -184,7 +178,7 @@ def create_agent(
         model=model,
         tools=[tool for tool in WEB_TOOLS if tool is not None], # Filter out None if search_tool was not set
         max_steps=20,
-        verbosity_level=3,
+        verbosity_level=3, # Keep this high for detailed output
         planning_interval=4,
         name="search_agent",
         description="A team member that will search the internet to answer your question.",
@@ -199,7 +193,7 @@ Additionally, if after some searching you find out that you need more informatio
         model=model,
         tools=[visualizer, TextInspectorTool(model, text_limit)],
         max_steps=16,
-        verbosity_level=3,
+        verbosity_level=3, # Keep this high for detailed output
         additional_authorized_imports=AUTHORIZED_IMPORTS,
         planning_interval=4,
         managed_agents=[text_webbrowser_agent],
@@ -240,9 +234,15 @@ def run_agent_with_streaming(agent, question, stream_callback=None):
 
     try:
         # Configure logging to capture everything
+        # Set logging levels very low to capture all verbose output
         root_logger.setLevel(logging.DEBUG)
+        for handler in root_logger.handlers: # Remove existing handlers to avoid duplicate output
+            root_logger.removeHandler(handler)
         root_logger.addHandler(log_handler)
+
         smolagents_logger.setLevel(logging.DEBUG)
+        for handler in smolagents_logger.handlers: # Remove existing handlers
+            smolagents_logger.removeHandler(handler)
         smolagents_logger.addHandler(log_handler)
         
         # Redirect stdout/stderr
@@ -275,7 +275,7 @@ def run_agent_with_streaming(agent, question, stream_callback=None):
         sys.stdout = original_stdout
         sys.stderr = original_stderr
         
-        # Ensure any remaining buffered output is flushed
+        # Ensure any remaining buffered output is flushed (especially important for stdout/stderr)
         stdout_capture.flush()
         stderr_capture.flush()
 
@@ -302,12 +302,6 @@ def main():
     print("[DEBUG] CLI arguments parsed:", args)
 
     if args.gradio:
-        # This part should ideally be handled by app.py directly.
-        # However, if run.py is also executable to launch Gradio,
-        # it should call app.py's launch_interface.
-        # For this refactor, we are assuming app.py is the primary entry point for Gradio.
-        # If you intend run.py to be runnable with --gradio, you would import and call launch_interface() from app.py here.
-        # For now, this main function only supports CLI mode if --gradio is not passed.
         print("Please run `app.py` directly to launch the Gradio interface.")
         return
     else: