diff --git a/API/audio_api.py b/API/audio_api.py
index 774c844..89169df 100644
--- a/API/audio_api.py
+++ b/API/audio_api.py
@@ -47,10 +47,10 @@ from Core.config_loader import load_config
 LOG_DIR = os.path.join(PROJECT_ROOT, "logs")
 os.makedirs(LOG_DIR, exist_ok=True)
 
-# logging.basicConfig is idempotent per process: if marcus_voice configured
-# the root logger first, this call is a no-op and both modules share the same
-# RotatingFileHandler (stdlib FileHandlers hold an internal lock, so concurrent
-# writes to voice.log are safe). Rotation caps voice.log at 5 MB × 3 backups.
+# All voice-subsystem logs go ONLY to logs/voice.log, not stdout — the
+# terminal REPL needs a clean `Command:` prompt. Anything the operator
+# needs to see is print()-ed explicitly from the callback sites.
+# basicConfig is idempotent (no-op if marcus_voice installed handlers first).
 logging.basicConfig(
     level=logging.INFO,
     format="%(asctime)s [%(name)s] %(levelname)s: %(message)s",
@@ -59,7 +59,6 @@ logging.basicConfig(
             os.path.join(LOG_DIR, "voice.log"),
             maxBytes=5_000_000, backupCount=3, encoding="utf-8",
         ),
-        logging.StreamHandler(),
     ],
 )
 log = logging.getLogger("audio_api")
diff --git a/Brain/marcus_brain.py b/Brain/marcus_brain.py
index c5a6465..edb966f 100644
--- a/Brain/marcus_brain.py
+++ b/Brain/marcus_brain.py
@@ -172,7 +172,10 @@ def _init_voice():
             text = (text or "").strip()
             if not text:
                 return
-            print(f"  [Voice] {text}")
+            # One clean, distinctive line so the operator can see exactly
+            # what Whisper transcribed before the brain reacts. Everything
+            # else from the voice subsystem is file-only.
+            print(f'  [Sanad] heard: "{text}"')
             try:
                 result = process_command(text)
             except Exception as e:
@@ -182,10 +185,12 @@ def _init_voice():
                 sp = (result.get("speak") or "").strip()
                 if sp and _audio_api:
                     _audio_api.speak(sp)
+            # Redraw the Command: prompt that our print clobbered
+            print("Command: ", end="", flush=True)
 
         _voice_module = VoiceModule(_audio_api, on_command=_on_command)
         _voice_module.start()
-        print("  [Voice] Always listening (Whisper + G1 mic + TtsMaker)")
+        print("  [Voice] listening in background — say \"Sanad\" + your command")
     except Exception as e:
         print(f"  [Voice] Init failed: {e} — continuing without voice")
         _audio_api = None
@@ -477,13 +482,15 @@ def run_terminal():
 
     status = get_brain_status()
     print()
-    print("=" * 48)
-    print("         SANAD AI BRAIN — READY")
-    print("=" * 48)
+    print("=" * 54)
+    print("           SANAD AI BRAIN — READY")
+    print("=" * 54)
     for k, v in status.items():
-        print(f"  {k:<10}: {v}")
-    print("=" * 48)
-    print("  help | example | yolo | patrol | auto on/off | q")
+        print(f"  {k:<12}: {v}")
+    print("-" * 54)
+    print("  Type a command, or say \"Sanad, <command>\".")
+    print("  Shortcuts: help | example | yolo | patrol | auto on/off | q")
+    print("=" * 54)
     print()
 
     try:
diff --git a/Voice/marcus_voice.py b/Voice/marcus_voice.py
index a8a7d50..f712a37 100644
--- a/Voice/marcus_voice.py
+++ b/Voice/marcus_voice.py
@@ -40,9 +40,11 @@ from Core.config_loader import load_config
 LOG_DIR = os.path.join(PROJECT_ROOT, "logs")
 os.makedirs(LOG_DIR, exist_ok=True)
 
-# basicConfig is idempotent. Whichever of audio_api / marcus_voice imports
-# first installs the rotating handler; the other no-ops. Both loggers then
-# share the same file handle with stdlib's per-handler thread lock.
+# Voice runs as a background subsystem — its INFO/DEBUG logs go ONLY to
+# logs/voice.log so they don't drown out the interactive `Command:` prompt.
+# Anything the user needs to see (wake-word fired, command heard) is
+# print()-ed explicitly from the callbacks below.
+# basicConfig is idempotent; audio_api may have already called it.
 logging.basicConfig(
     level=logging.INFO,
     format="%(asctime)s [%(name)s] %(levelname)s: %(message)s",
@@ -51,7 +53,6 @@ logging.basicConfig(
             os.path.join(LOG_DIR, "voice.log"),
             maxBytes=5_000_000, backupCount=3, encoding="utf-8",
         ),
-        logging.StreamHandler(),
     ],
 )
 log = logging.getLogger("marcus_voice")
@@ -248,6 +249,11 @@ class VoiceModule:
 
         if self._check_wake_word(text):
             log.info("Wake word detected!")
+            # One clean line to the terminal so the operator knows voice
+            # actually heard them, even though all other voice logs are
+            # file-only. \n leads because we may be painting over a
+            # half-drawn `Command:` prompt.
+            print("\n  [Sanad] wake heard — recording command…")
             self._state = State.WAKE_HEARD
 
             # Acknowledge