Update response_modes.py

response_modes.py

@@ -425,6 +425,7 @@ def chat_stream(message, history, current_model, logged_in_username, user_chats,
 
 You MUST follow the above instruction naturally in EVERY response."""
 
+    # ====================== CODING MODE — 3-AGENT CREWAI ======================
     if mode == "💻 Coding Mode":
         # Gather persistent project files
         coding_files_str = ""
@@ -443,24 +444,32 @@ You MUST follow the above instruction naturally in EVERY response."""
         history.append({"role": "assistant", "content": "🚀 Starting **3-Agent Coding Crew** (Cerebras Reviewer active)..."})
         yield history
 
-        search_tool = DuckDuckGoSearchRun()
-
-        # Cerebras Reviewer (gpt-oss-120b with high reasoning)
+        # ================== CUSTOM SEARCH TOOL (uses your existing DDGS) ==================
+        @tool
+        def web_search_tool(query: str) -> str:
+            """Search the web (StackOverflow, Reddit, GitHub, docs) for coding solutions."""
+            try:
+                results = web_search(query, max_results=15)  # reuse your existing function
+                return "\n\n".join(results[:12]) if results else "No relevant results found."
+            except Exception as e:
+                return f"Search failed: {str(e)[:200]}"
+
+        # Cerebras Reviewer
         cerebras_llm = ChatOpenAI(
             base_url="https://api.cerebras.ai/v1",
             api_key=os.environ.get("CEREBRAS_API_KEY"),
             model="gpt-oss-120b",
             temperature=0.3,
             max_tokens=16384,
-            reasoning_effort="high"   # ← exactly what you wanted
+            reasoning_effort="high"
         )
 
-        # === 3 AGENTS ===
+        # === AGENTS ===
         researcher = Agent(
             role="Senior Coding Researcher",
             goal="Find the best, most up-to-date solutions from StackOverflow, Reddit, GitHub, and official docs",
-            backstory="You are world-class at quickly finding high-quality coding examples and solutions.",
-            tools=[search_tool],
+            backstory="You are world-class at quickly finding high-quality coding examples.",
+            tools=[web_search_tool],
             llm="groq/qwen/qwen3-32b",
             verbose=True
         )
@@ -468,58 +477,57 @@ You MUST follow the above instruction naturally in EVERY response."""
         coder = Agent(
             role="Senior Full-Stack Engineer",
             goal="Write clean, efficient, production-ready, well-commented code",
-            backstory="You are an expert developer who produces beautiful, correct code.",
-            tools=[search_tool],
+            backstory="You produce beautiful, correct code.",
+            tools=[web_search_tool],
             llm=current_model,
             verbose=True
         )
 
         reviewer = Agent(
             role="Principal Code Reviewer & Architect",
-            goal="Critically review every line for bugs, security issues, performance, best practices, and maintainability",
-            backstory="You are extremely strict, highly analytical, and have exceptional reasoning ability.",
-            tools=[search_tool],
-            llm=cerebras_llm,          # ← Cerebras gpt-oss-120b
+            goal="Critically review every line for bugs, security, performance and best practices",
+            backstory="You are extremely strict and have exceptional reasoning ability.",
+            tools=[web_search_tool],
+            llm=cerebras_llm,
             verbose=True
         )
 
         # === TASKS ===
         task_research = Task(
-            description=f"Research the best current solutions for: {full_task}",
+            description=f"Research the best current solutions: {full_task}",
             agent=researcher,
-            expected_output="Detailed research summary with useful code patterns and links"
+            expected_output="Detailed research summary with code patterns"
         )
 
         task_code = Task(
-            description=f"Using the research and persistent project files, implement the request perfectly: {full_task}",
+            description=f"Using research and project files, implement perfectly: {full_task}",
             agent=coder,
-            expected_output="Complete, clean, ready-to-use code with comments"
+            expected_output="Complete, clean, ready-to-use code"
         )
 
         task_review = Task(
-            description="Perform a deep, thorough review of the code. Point out every issue and provide the final polished version.",
+            description="Deep review the code. Point out every issue and give the final polished version.",
             agent=reviewer,
             expected_output="Detailed review + Final recommended code"
         )
 
-        # === RUN THE CREW ===
-        crew = Crew(
-            agents=[researcher, coder, reviewer],
-            tasks=[task_research, task_code, task_review],
-            process=Process.sequential,
-            verbose=2
-        )
-
+        # === RUN CREW WITH FULL ERROR PROTECTION ===
         try:
+            crew = Crew(
+                agents=[researcher, coder, reviewer],
+                tasks=[task_research, task_code, task_review],
+                process=Process.sequential,
+                verbose=2
+            )
             result = crew.kickoff()
             final_output = f"✅ **3-Agent Coding Crew Complete**\n\n{result}"
         except Exception as e:
-            final_output = f"❌ CrewAI Error: {str(e)[:400]}"
+            final_output = f"❌ CrewAI Error: {str(e)[:600]}\n\nTry again or check your CEREBRAS_API_KEY."
 
         history[-1]["content"] = final_output
         yield history
 
-        # Auto-save chat
+        # Auto-save
         if logged_in_username and user_chats:
             for chat in user_chats:
                 if chat["title"] == current_chat_title: