Fixing structured outputs

app.py

@@ -911,20 +911,26 @@ async def chat_with_tool_visibility(
             if content:
                 # Clear "Thinking..." on first real content
                 if chat_response == "💭 _Thinking..._":
+                    print("[STREAM] Clearing 'Thinking...' placeholder")
                     chat_response = ""
                     progress(0.7, desc="📝 Generating response...")
 
                 # Handle both string (OpenAI) and list (Anthropic) content formats
+                content_to_add = ""
                 if isinstance(content, list):
                     # Anthropic returns list of content blocks - extract text
                     for block in content:
                         if hasattr(block, 'text'):
-                            chat_response += block.text
+                            content_to_add += block.text
                         elif isinstance(block, dict) and 'text' in block:
-                            chat_response += block['text']
+                            content_to_add += block['text']
                 else:
                     # OpenAI/HF return string directly
-                    chat_response += content
+                    content_to_add = content
+
+                if content_to_add:
+                    print(f"[STREAM] Adding LLM content: {content_to_add[:100]}...")
+                    chat_response += content_to_add
                 yield chat_response, tool_log
 
         # Tool finished
@@ -947,7 +953,10 @@ async def chat_with_tool_visibility(
             # Format output for chat display (with image rendering)
             formatted_output = format_tool_output_for_chat(tool_output)
             if formatted_output.strip():
+                print(f"[STREAM] Adding formatted tool output ({len(formatted_output)} chars): {formatted_output[:200]}...")
+                print(f"[STREAM] chat_response length before: {len(chat_response)}")
                 chat_response += formatted_output
+                print(f"[STREAM] chat_response length after: {len(chat_response)}")
 
             yield chat_response, tool_log
 
@@ -956,6 +965,10 @@ async def chat_with_tool_visibility(
     # yield chat_response, tool_log
     progress(0.9, desc="✨ Finalizing response...")
 
+    print(f"\n[FINAL] chat_response length before parsing: {len(chat_response)}")
+    print(f"[FINAL] chat_response preview (first 300): {chat_response[:300]}")
+    print(f"[FINAL] chat_response preview (last 300): {chat_response[-300:]}\n")
+
     try:
         from langgraph_agent.structured_output import parse_agent_response
         formatted_response = await parse_agent_response(

langgraph_agent/structured_output.py

@@ -52,11 +52,17 @@ def extract_urls_from_text(text: str) -> tuple[List[str], List[str]]:
     audio_pattern_files = r'https?://[^\s)}\]]+?\.(?:mp3|wav|ogg|m4a)(?:\?[^\s)}\]]*)?'
     audio_pattern_xenocanto = r'https?://xeno-canto\.org/\d+/download'
 
+    print(f"[EXTRACT_URLS] Searching text of length {len(text)}")
+
     # Extract all URLs
     raw_image_urls = re.findall(image_pattern, text, re.IGNORECASE)
     raw_audio_urls_files = re.findall(audio_pattern_files, text, re.IGNORECASE)
     audio_urls_xenocanto = list(set(re.findall(audio_pattern_xenocanto, text, re.IGNORECASE)))
 
+    print(f"[EXTRACT_URLS] Found {len(raw_image_urls)} raw image URLs")
+    print(f"[EXTRACT_URLS] Found {len(raw_audio_urls_files)} audio file URLs")
+    print(f"[EXTRACT_URLS] Found {len(audio_urls_xenocanto)} xeno-canto URLs")
+
     # Clean URLs (remove trailing quotes, commas, etc.)
     def clean_url(url: str) -> str:
         return url.rstrip('",;)')
@@ -112,6 +118,9 @@ async def parse_agent_response(
     """
     try:
         print("[STRUCTURED OUTPUT] Starting parsing...")
+        print(f"[STRUCTURED OUTPUT] Raw response length: {len(raw_response)} characters")
+        print(f"[STRUCTURED OUTPUT] First 500 chars: {raw_response[:500]}")
+        print(f"[STRUCTURED OUTPUT] Last 500 chars: {raw_response[-500:]}")
 
         # Extract URLs using regex (fast, no API call)
         image_urls, audio_urls = extract_urls_from_text(raw_response)