Update app.py

app.py

@@ -9,37 +9,39 @@ web_search = DuckDuckGoSearchTool()
 SYSTEM_PROMPT = """
 You are an AI research assistant that can search the web. Follow these steps:
 
-1. FIRST ANALYZE the user's question:
-   - If information is missing or ambiguous, ask ONE clarifying question
-   - If clear, proceed to search
-
-2. When searching:
-   - Generate multiple specific search queries wrapped in <search> tags
-   - Focus on factual keywords, one query per line
-   Example:
-   <search>
-   Pont des Arts exact length meters
-   History of Pont des Arts bridge
-   </search>
-
-3. After receiving results:
-   - Analyze information from multiple sources
-   - Cross-verify facts
-   - If needed, generate follow-up searches
-   - Provide final answer with:
-     - Clear structure
-     - Key facts with sources
-     - Concise explanations
-
-Never invent information. Cite sources for all facts. Use neutral, academic tone.
+<thinking>
+1. Analyze the user's question:
+   - Identify missing information or ambiguities
+   - Determine if clarification is needed
+   - Plan search strategy if required
+</thinking>
+
+If clarification is needed, ask ONE question. Otherwise:
+
+<search>
+[list of search queries]
+</search>
+
+After receiving results:
+<thinking>
+- Analyze information from multiple sources
+- Cross-verify facts
+- Determine if additional searches are needed
+</thinking>
+
+Final answer:
+- Clear structure with headings
+- Key facts with sources
+- Concise explanations
+- Academic tone
 """
 
 def process_searches(response):
     searches = re.findall(r'<search>(.*?)</search>', response, re.DOTALL)
     if searches:
         queries = [q.strip() for q in searches[0].split('\n') if q.strip()]
-        return queries, response  # Return both queries and original response
-    return None, response
+        return queries
+    return None
 
 def search_with_retry(query, max_retries=3, delay=2):
     for attempt in range(max_retries):
@@ -105,15 +107,13 @@ def respond(
                 token = chunk.choices[0].delta.content or ""
                 response += token
                 full_response += token
-                yield full_response  # Stream every token immediately
+                yield full_response  # Stream raw output with all tags
 
-            # Process searches without modifying original response
-            queries, original_response = process_searches(response)
+            queries = process_searches(response)
             
             if queries:
                 search_cycle = True
-                # Preserve original response with <search> tags
-                messages.append({"role": "assistant", "content": original_response})
+                messages.append({"role": "assistant", "content": response})
                 
                 search_results = []
                 for query in queries:
@@ -124,13 +124,10 @@ def respond(
                         search_results.append(f"⚠️ Search Error: {str(e)}\nQuery: {query}")
                         time.sleep(2)
                 
-                # Add search results to conversation
                 messages.append({
                     "role": "user",
                     "content": f"SEARCH RESULTS:\n{chr(10).join(search_results)}\nAnalyze these results..."
                 })
-                
-                # Show search status without breaking response flow
                 full_response += "\n🔍 Analyzing search results...\n"
                 yield full_response