Bolt Assistant: wake-phrase voice assistant for Bolt

bolt_assistant/config.py

@@ -0,0 +1,110 @@
+"""Settings for Bolt Assistant, from a `.env` next to the project (or the
+real environment, which wins). Everything else reads config from here."""
+
+from __future__ import annotations
+
+import os
+import platform
+from pathlib import Path
+
+HERE = Path(__file__).resolve().parent.parent  # project root
+
+
+def _load_env() -> None:
+    env_path = HERE / ".env"
+    if not env_path.exists():
+        return
+    for line in env_path.read_text(encoding="utf-8").splitlines():
+        line = line.strip()
+        if not line or line.startswith("#") or "=" not in line:
+            continue
+        key, _, value = line.partition("=")
+        os.environ.setdefault(key.strip(), value.strip())
+
+
+_load_env()
+
+
+def _flag(name: str, default: str) -> bool:
+    return os.environ.get(name, default).strip().lower() in ("1", "true", "yes", "on")
+
+
+def _device(raw: str):
+    """An input/output device: blank = system default, digits = an index,
+    anything else = a name substring (sounddevice's rules)."""
+    raw = (raw or "").strip()
+    if not raw:
+        return None
+    return int(raw) if raw.lstrip("-").isdigit() else raw
+
+
+# ── server ──────────────────────────────────────────────────────────────────
+
+SERVER_URL = os.environ.get("BOLT_SERVER_URL", "").rstrip("/")
+API_KEY = os.environ.get("DESK_API_KEY", "")
+# Its own conversation, separate from the pet's and the console's. The "pet-"
+# prefix gets the desk API's voice-client behaviour (short, speakable replies).
+SESSION_ID = os.environ.get("DESK_SESSION_ID") or "pet-assistant-" + (platform.node() or "desktop")
+# The voice the server's /desk/tts speaks in (an ElevenLabs voice id).
+VOICE_ID = os.environ.get("ELEVENLABS_VOICE_ID", "")
+TTS_SAMPLE_RATE = int(os.environ.get("TTS_SAMPLE_RATE", "16000"))
+STREAMING_REPLIES = _flag("STREAMING_REPLIES", "true")
+
+# ── audio devices ───────────────────────────────────────────────────────────
+
+MIC_DEVICE = _device(os.environ.get("MIC_DEVICE", ""))
+SPEAKER_DEVICE = _device(os.environ.get("SPEAKER_DEVICE", ""))
+
+# ── wake phrases ────────────────────────────────────────────────────────────
+# 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.
+
+WAKE_WORDS = [w.strip().lower() for w in os.environ.get(
+    "WAKE_WORDS",
+    "hey bolt,bolt,thunderbolt,hey thunderbolt,ok bolt,okay bolt,yo bolt,hi bolt,"
+    "hey assistant,ok assistant,okay assistant,"
+    "hey bolts,hey volt,hey bold,hey boat,okay volt,hey thunder bolt,thunder bolt",
+).split(",") if w.strip()]
+
+# ── listening (local voice-activity detection, before anything is sent) ────
+# Nothing leaves the machine until this says someone is talking: silence and
+# background hum are never sent (or billed). Raise VAD_RMS_THRESHOLD if
+# background noise keeps getting transcribed; lower it if you have to shout.
+
+VAD_RMS_THRESHOLD = int(os.environ.get("VAD_RMS_THRESHOLD", "500"))
+VAD_START_FRAMES = int(os.environ.get("VAD_START_FRAMES", "3"))           # 80 ms each
+VAD_SILENCE_END_SEC = float(os.environ.get("VAD_SILENCE_END_SEC", "0.8"))
+MIN_UTTERANCE_SEC = float(os.environ.get("MIN_UTTERANCE_SEC", "0.35"))     # shorter: never sent
+MAX_UTTERANCE_SEC = float(os.environ.get("MAX_UTTERANCE_SEC", "12"))
+# After a bare "Hey Bolt", how long to wait for the actual request.
+COMMAND_WAIT_SEC = float(os.environ.get("COMMAND_WAIT_SEC", "8"))
+# When a reply ends on a question, keep listening for the answer without the
+# wake word, for this long, at most this many times in a row.
+FOLLOW_UP_SEC = float(os.environ.get("FOLLOW_UP_SEC", "8"))
+FOLLOW_UP_MAX_TURNS = int(os.environ.get("FOLLOW_UP_MAX_TURNS", "5"))
+CHIMES = _flag("CHIMES", "true")
+
+# ── what Bolt may do on this machine ────────────────────────────────────────
+# The server can relay commands back. With no screen to approve them on:
+#   off   nothing runs; Bolt is told it can't here
+#   read  (default) listing and reading files only, via filectl
+#   all   everything, including shell commands and file writes — like
+#         Bolt-Pet. Only if you trust every request that reaches your mic.
+
+COMMANDS = os.environ.get("ASSISTANT_COMMANDS", "read").strip().lower()
+COMMAND_TIMEOUT_SECONDS = int(os.environ.get("COMMAND_TIMEOUT_SECONDS", "60"))
+
+SAMPLE_RATE = 16000
+FRAME_SAMPLES = 1280  # 80 ms
+
+
+def is_configured() -> bool:
+    return bool(SERVER_URL and API_KEY)
+
+
+def missing_config() -> list:
+    return [name for name, value in (("BOLT_SERVER_URL", SERVER_URL), ("DESK_API_KEY", API_KEY),
+                                     ("ELEVENLABS_VOICE_ID", VOICE_ID)) if not value]