feat: aggiunge OllamaToolAgent con tool calling

- Nuovo modulo ollama_agent.py con parsing delle risposte
- Supporta chiamate tool per provider senza bind_tools
- Tool built-in: execute, read_file, write_file, ls, search, system_info
- Il modello risponde con [TOOL_CALL] che noi parsiamo ed eseguiamo
- Integrato in main.py per uso automatico con Ollama

main.py

@@ -414,19 +414,52 @@ def create_langcode_agent(system_prompt: Optional[str] = None, use_simple_agent:
     """
     Create and return an agent instance with configurable model.
     
-    For Ollama and other providers without full tool-calling support,
-    uses a simple chat agent instead of deepagents.
+    For Ollama uses custom OllamaToolAgent with tool calling support.
+    For other providers with full tool support uses deepagents.
     """
-    # Detect if we need simple agent (for Ollama without tool support)
-    if not use_simple_agent and CONFIG_AVAILABLE:
+    # Try to get config info for Ollama detection
+    is_ollama = False
+    if CONFIG_AVAILABLE:
         active_name = config_manager.get_active_configuration()
         if active_name:
             config = config_manager.get_configuration(active_name)
             if config and config.get("provider") == "ollama":
-                # Ollama doesn't fully support bind_tools in deepagents
-                # Always use simple agent for Ollama
-                print("⚠️  Using simple chat agent (Ollama detected)")
-                use_simple_agent = True
+                is_ollama = True
+    
+    # Use OllamaToolAgent for Ollama with tool calling
+    if is_ollama and not use_simple_agent:
+        try:
+            from ollama_agent import create_ollama_tool_agent
+            model = load_model_from_config()
+            custom_tools = get_custom_tools()
+            
+            # Enhance with memory and planning tools if available
+            if ENHANCEMENTS_AVAILABLE:
+                try:
+                    memory_tools = create_simple_memory_tools(simple_memory)
+                    custom_tools.extend(memory_tools)
+                except Exception:
+                    pass
+                try:
+                    planning_tools = enhanced_planner.create_planning_tools()
+                    custom_tools.extend(planning_tools)
+                except Exception:
+                    pass
+            
+            print("🔧 Using OllamaToolAgent with tool calling")
+            return create_ollama_tool_agent(
+                model=model,
+                tools=custom_tools,
+                system_prompt=system_prompt,
+                include_builtin_tools=True
+            )
+        except Exception as e:
+            print(f"⚠️  Could not load OllamaToolAgent: {e}")
+            print("⚠️  Falling back to simple agent")
+            use_simple_agent = True
+    elif is_ollama:
+        print("⚠️  Using simple chat agent (Ollama detected)")
+        use_simple_agent = True
     
     if use_simple_agent:
         return create_simple_agent(system_prompt)
@@ -1099,9 +1132,17 @@ def main():
             # Otherwise, treat it as a chat message
             try:
                 result = agent.invoke({"messages": [HumanMessage(content=user_input)]})
-                # Extract the AI's response
-                ai_message = result["messages"][-1]
-                print(f"\n🤖 LangCode: {ai_message.content}")
+                
+                # Extract the AI's response (works for both simple and tool agent)
+                if "response" in result:
+                    # OllamaToolAgent returns a separate response field
+                    response_text = result["response"]
+                else:
+                    # Simple agent returns the last message
+                    ai_message = result["messages"][-1]
+                    response_text = ai_message.content if hasattr(ai_message, 'content') else str(ai_message)
+                
+                print(f"\n🤖 LangCode: {response_text}")
             except Exception as e:
                 error_msg = str(e)
                 if not error_msg: