Set Qwen2.5-7B-Instruct-4bit as default MLX model and add spoken text filtering

bot.py

@@ -159,8 +159,8 @@ def parse_args() -> argparse.Namespace:
     )
     parser.add_argument(
         "--mlx-model",
-        default="mlx-community/gemma-2-2b-it-4bit",
-        help="MLX model repo or path for local macOS execution (e.g. mlx-community/gemma-2-2b-it-4bit).",
+        default="mlx-community/Qwen2.5-7B-Instruct-4bit",
+        help="MLX model repo or path for local macOS execution (e.g. mlx-community/Qwen2.5-7B-Instruct-4bit).",
     )
     parser.add_argument(
         "--claude-model",
@@ -495,6 +495,8 @@ def build_llm(args: argparse.Namespace, vocabulary=None, brain=None, observer=No
         logger.info(f"LLM: macOS native model ({reason})")
         personality = read_personality(args.cwd) or "You are a helpful macOS voice assistant."
         system_prompt = personality + "\n\n" + VOICE_STYLE
+        if brain and (memory := brain.prompt_block()):
+            system_prompt += "\n\n" + memory
         return MacOSLLM(model=args.mlx_model, system_prompt=system_prompt, observer=observer)
 
     logger.info("LLM: Claude Code")