config.py

@@ -122,12 +122,8 @@ SESSION_DB_PATH = os.path.expanduser(
 
 # ─── RAG (YAML) ─────────────────────────────────────────────────────────────────
 
-RAG_PERSIST_DIR = os.path.expanduser(
-    os.getenv("RAG_PERSIST_DIR", default=_yaml_get("rag", "persist_dir", default="lancedb_data"))
-)
-RAG_MODEL_PATH = os.path.expanduser(
-    os.getenv("RAG_MODEL_PATH", default=_yaml_get("rag", "model_path", default=""))
-)
+RAG_PERSIST_DIR = os.getenv("RAG_PERSIST_DIR", default=_yaml_get("rag", "persist_dir", default="lancedb_data"))
+RAG_MODEL_PATH = os.getenv("RAG_MODEL_PATH", default=_yaml_get("rag", "model_path", default=""))
 
 
 # ─── Humanize Delay (YAML) ─────────────────────────────────────────────────────

lib/agent_loop.py

@@ -27,13 +27,7 @@ def execute_tool(tool_call, TOOL_HANDLERS):
 def run_agent_loop(session, llm_client, TOOLS, TOOL_HANDLERS, max_iterations, on_tool_calls=None):
     for step in range(max_iterations):
         print(f'[{_ts()}] Step {step + 1} — calling LLM...', flush=True)
-        
-        # Ambil konfigurasi disable_reasoning dari personality karakter
-        # Default ke False jika tidak didefinisikan
-        personality = getattr(session, 'personality', {})
-        disable_reasoning = personality.get('disable_reasoning', False)
-        
-        response = llm_client.chat(session.messages, tools=TOOLS, disable_reasoning=disable_reasoning)
+        response = llm_client.chat(session.messages, tools=TOOLS)
 
         if response.tool_calls:
             amsg = {

services/llm_client.py

@@ -19,7 +19,7 @@ class LLMClient:
         self.timeout = timeout
         self.cancel_requested = False
 
-    def chat(self, messages, tools=None, on_stream_chunk=None, disable_reasoning=False):
+    def chat(self, messages, tools=None, on_stream_chunk=None):
         url = f"{self.base_url}/chat/completions"
         payload = {
             "model": self.model,
@@ -30,9 +30,8 @@ class LLMClient:
             payload["tools"] = tools
             payload["tool_choice"] = "auto"
 
-        # Hanya kirim parameter reasoning jika diminta eksplisit
-        # Beberapa model/provider justru error jika parameter ini ada tapi tidak didukung
-        if disable_reasoning:
+        # Disable reasoning/thinking di level API bila didukung
+        # OpenRouter & beberapa provider support ini
         payload["reasoning"] = {"enabled": False}
 
         data = json.dumps(payload).encode('utf-8')
@@ -103,11 +102,11 @@ class LLMClient:
 
                     # Stream content (text response)
                     if 'content' in delta:
-                        chunk_text = delta['content'] or ""
+                        chunk_text = delta['content']
                         full_content += chunk_text
 
                         # Callback untuk streaming ke UI
-                        if on_stream_chunk and chunk_text:
+                        if on_stream_chunk:
                             on_stream_chunk(chunk_text)
 
             # Build final response