Refactor wake phrase handling: introduce REQUIRE_WAKE_WORD configuration, update documentation, and adjust assistant behavior for improved user experience.

bolt_assistant/config.py

@@ -56,11 +56,18 @@ MIC_DEVICE = _device(os.environ.get("MIC_DEVICE", ""))
 SPEAKER_DEVICE = _device(os.environ.get("SPEAKER_DEVICE", ""))
 
 # ── wake phrases ────────────────────────────────────────────────────────────
+# With REQUIRE_WAKE_WORD off (the default) the mic itself is the activation:
+# every utterance the local detector passes on is treated as a request, and
+# whatever is said near an open mic gets answered. Set it to true to go back
+# to needing a phrase before each request.
+REQUIRE_WAKE_WORD = _flag("REQUIRE_WAKE_WORD", "false")
+
 # Heard through the server's speech-to-text, so these are matched against
 # *text*. A phrase counts only near the start of what was said ("hey bolt,
 # what's the time"), not buried in a sentence ("tighten that bolt"). Speech
 # recognisers often write "Bolt" as "bolts"/"volt"/"bold"; those spellings
-# only count after "hey"/"ok", where they can't be ordinary words.
+# only count after "hey"/"ok", where they can't be ordinary words. Only used
+# when REQUIRE_WAKE_WORD is on (a phrase said anyway is still stripped).
 
 WAKE_WORDS = [w.strip().lower() for w in os.environ.get(
     "WAKE_WORDS",