Snapshot: full project state

2026-10-06 23:43:30 -07:00
commit 9991a836fe
15 changed files with 1642 additions and 0 deletions
--- a/.gitignore
+++ b/.gitignore
@@ -0,0 +1,17 @@
+__pycache__/
+*.pyc
+node_modules/
+.venv/
+venv/
+.env
+*.db
+*.sqlite*
+*.log
+.DS_Store
+out/
+work/
+.pio/
+briefs/
+dns-backup/
+archive/
+*.png
--- a/README.md
+++ b/README.md
@@ -0,0 +1,66 @@
+# The Collector
+
+A smooth-operator conversational voice AI running on drjones' homelab. Lives on
+Proxmox LXC **CT 771** @ `10.30.20.172:8766`.
+
+## What it is
+A latency-optimized voice assistant. It listens, responds naturally through a
+local LLM, speaks with a natural voice, and can make outbound phone calls on
+request.
+
+## Pipeline (latency budget)
+| Stage | Engine | Latency |
+|-------|--------|---------|
+| STT | faster-whisper `small.en` (GPU, nightmare `.29:8770`) | ~0.35s / utterance |
+| LLM | `ornith-1.5:9b-64k` @ nightmare `.29:11434` (fallback `.222`) | ~600ms first token, streamed |
+| TTS | Piper `lessac-high` (CPU, 22.05k → 16k) | ~400ms / sentence, overlapped w/ LLM |
+
+Sentence-level TTS/LLM overlap: audio starts flowing after the first sentence
+completes while the model keeps generating. Server-side energy VAD with barge-in.
+
+## Interfaces
+- **Browser voice** — `GET /` (WebSocket `/ws`, PCM16 16 kHz full-duplex)
+- **Phone** — SignalWire (outbound via `POST /api/call {to, task}`; inbound stays
+  on JONES's number for now). Media stream `/signalwire/stream` (mulaw 8 kHz).
+- **Text** — `POST /api/text {text}`
+- **Health** — `GET /api/health`
+
+## Public URL
+Cloudflare quick tunnel (ephemeral): `https://placed-ivory-unto-effective.trycloudflare.com`
+Self-registers via `POST /api/config {public_base_url}` (systemd `collector-tunnel.service`).
+
+## Key decisions
+- **LLM lane = nightmare `.29`, not `.222`**: `.222` holds ornith resident but is
+  saturated by ~92 cron jobs + Kalshi bots and hangs on `/api/chat`; `.29` answers
+  in ~1s. `keep_alive:30m` keeps ornith warm there.
+- **STT on GPU (nightmare), not the CT CPU**: whisper on the CT's 2 cores was 7-8s
+  (the real latency killer); `faster-whisper` on nightmare's 4080S
+  (`collector-stt.service` @ `:8770`, CUDA) is ~0.35s — 20×. TTS stays on CPU (Piper
+  ~400ms, not the bottleneck).
+- **Phone** reuses the SignalWire account + number (`+1 254 513 9895`) JONES uses;
+  outbound TwiML points at this app's tunnel.
+
+## Files
+```
+/opt/collector/
+  app.py               self-contained FastAPI app (STT/TTS/LLM/phone)
+  config.yaml          model lane, TTS voice, SignalWire creds, persona
+  selftest.py          Piper→whisper round-trip self-test
+  static/index.html    browser voice UI (AudioWorklet)
+  tunnel.sh            cloudflared quick tunnel + self-registration
+  collector.service    systemd unit (app)
+  collector-tunnel.service  systemd unit (tunnel)
+```
+
+## Ops
+```bash
+# restart
+ssh root@10.30.20.85 "pct exec 771 -- systemctl restart collector"
+# logs
+ssh root@10.30.20.85 "pct exec 771 -- journalctl -u collector -n 30"
+# self-test the audio pipeline
+ssh root@10.30.20.85 "pct exec 771 -- python3 /opt/collector/selftest.py"
+# make a phone call
+curl -s -X POST http://10.30.20.172:8766/api/call \
+  -H 'Content-Type: application/json' -d '{"to":"+1XXXXXXXXXX","task":"..."}'
+```
--- a/app.py
+++ b/app.py
@@ -0,0 +1,587 @@
+#!/usr/bin/env python3
+"""The Collector — a smooth-operator conversational voice AI.
+
+Pipeline (latency-optimized):
+  STT  faster-whisper small.en (CPU, int8)     ~1-2s / short utterance
+  LLM  ornith-1.5:9b-64k @ .222 (warm)         ~600ms first token, streamed
+  TTS  Piper lessac-high (CPU, 22.05k->16k)    ~400ms / sentence, overlapped w/ LLM
+
+Interfaces:
+  WS   /ws                      browser full-duplex voice (PCM16 16 kHz)
+  WS   /signalwire/stream       phone media stream (mulaw 8 kHz)
+  POST /api/call {to, task}     outbound phone call on request
+  POST /api/text {text}         text chat (returns text)
+  GET  /api/health
+"""
+import asyncio
+import base64
+import json
+import logging
+import os
+import re
+
+import audioop
+import httpx
+import numpy as np
+import yaml
+from fastapi import FastAPI, Request, WebSocket, WebSocketDisconnect
+from fastapi.responses import FileResponse, JSONResponse, PlainTextResponse
+
+logging.basicConfig(level=logging.INFO,
+                    format="%(asctime)s %(name)s %(levelname)s %(message)s")
+log = logging.getLogger("collector")
+
+BASE = "/opt/collector"
+
+
+def load_config() -> dict:
+    with open(os.path.join(BASE, "config.yaml")) as f:
+        return yaml.safe_load(f)
+
+
+CFG = load_config()
+
+# ---------------------------------------------------------------------------
+# audio helpers (stdlib audioop)
+# ---------------------------------------------------------------------------
+def ulaw_to_pcm(data: bytes) -> bytes:
+    return audioop.ulaw2lin(data, 2)
+
+
+def pcm_to_ulaw(pcm: bytes) -> bytes:
+    return audioop.lin2ulaw(pcm, 2)
+
+
+def resample(pcm: bytes, src: int, dst: int) -> bytes:
+    if src == dst or not pcm:
+        return pcm
+    out, _ = audioop.ratecv(pcm, 2, 1, src, dst, None)
+    return out
+
+
+def rms(pcm: bytes) -> int:
+    return audioop.rms(pcm, 2) if pcm else 0
+
+
+# ---------------------------------------------------------------------------
+# STT
+# ---------------------------------------------------------------------------
+class STT:
+    def __init__(self, cfg):
+        self.cfg = cfg
+        self.model = None
+        self.remote_url = (cfg.get("remote_url") or "").rstrip("/")
+
+    def load(self):
+        # Remote GPU STT (nightmare) is primary; local whisper is a lazy fallback.
+        return self
+
+    def transcribe(self, pcm16k: bytes) -> str:
+        if self.remote_url:
+            try:
+                return self._remote(pcm16k)
+            except Exception as e:
+                log.warning("remote STT failed (%s); falling back to local", e)
+        return self._local(pcm16k)
+
+    def _remote(self, pcm16k: bytes) -> str:
+        import requests
+        r = requests.post(self.remote_url + "/transcribe", data=pcm16k, timeout=30)
+        r.raise_for_status()
+        return (r.json().get("text") or "").strip()
+
+    def _local(self, pcm16k: bytes) -> str:
+        if self.model is None:
+            from faster_whisper import WhisperModel
+            c = self.cfg
+            log.info("Loading local faster-whisper %s (fallback)...", c["model"])
+            self.model = WhisperModel(c["model"], device=c["device"],
+                                      compute_type=c["compute_type"],
+                                      cpu_threads=c.get("cpu_threads", 2))
+        audio = np.frombuffer(pcm16k, dtype=np.int16).astype(np.float32) / 32768.0
+        segs, _ = self.model.transcribe(
+            audio, beam_size=1, language="en", vad_filter=True,
+            condition_on_previous_text=False, no_speech_threshold=0.6)
+        return " ".join(s.text.strip() for s in segs).strip()
+
+
+# ---------------------------------------------------------------------------
+# TTS
+# ---------------------------------------------------------------------------
+class TTS:
+    def __init__(self, cfg):
+        self.cfg = cfg
+        self.voice_path = cfg.get("piper_voice") or \
+            "/opt/collector/voices/en_US-lessac-high.onnx"
+        self._piper = None
+
+    def load(self):
+        if self._piper is None:
+            # onnxruntime pins threads to CPUs 0..N-1; the unprivileged LXC
+            # cpuset (e.g. 8,17) rejects that. Setting an explicit thread count
+            # disables the affinity pin (which is otherwise just log noise).
+            import onnxruntime
+            _orig = onnxruntime.InferenceSession.__init__
+
+            def _patched(self, *a, **k):
+                so = k.get("sess_options") or onnxruntime.SessionOptions()
+                so.intra_op_num_threads = 2
+                k["sess_options"] = so
+                _orig(self, *a, **k)
+
+            onnxruntime.InferenceSession.__init__ = _patched
+
+            from piper import PiperVoice
+            log.info("Loading Piper voice %s ...", self.voice_path)
+            self._piper = PiperVoice.load(self.voice_path)
+            log.info("TTS ready")
+        return self._piper
+
+    def synthesize_16k(self, text: str) -> bytes:
+        """Return PCM16 mono 16 kHz for `text` (piper yields AudioChunk iterators)."""
+        if not text.strip():
+            return b""
+        speed = float(self.cfg.get("speed", 1.05))
+        from piper import SynthesisConfig
+        syn = SynthesisConfig(length_scale=1.0 / speed)
+        chunks = []
+        rate = 22050
+        for chunk in self._piper.synthesize(text, syn_config=syn):
+            audio = (chunk.audio_float_array * 32767).astype(np.int16).tobytes()
+            rate = chunk.sample_rate
+            chunks.append(audio)
+        pcm = b"".join(chunks)
+        return resample(pcm, rate, 16000)
+
+
+# ---------------------------------------------------------------------------
+# LLM (ornith streaming, /api/chat, think:false)
+# ---------------------------------------------------------------------------
+async def stream_llm(messages, cfg):
+    payload = {
+        "model": cfg["ollama"]["model"],
+        "messages": messages,
+        "stream": True,
+        "think": False,
+        "keep_alive": "30m",
+        "options": {
+            "temperature": cfg["ollama"].get("temperature", 0.5),
+            "num_predict": cfg["ollama"].get("num_predict", 256),
+        },
+    }
+    urls = [cfg["ollama"]["base_url"], cfg["ollama"].get("fallback_url", "")]
+    for base in urls:
+        if not base:
+            continue
+        url = base.rstrip("/") + "/api/chat"
+        try:
+            async with httpx.AsyncClient(
+                    timeout=httpx.Timeout(10.0, read=60.0)) as client:
+                async with client.stream("POST", url, json=payload) as r:
+                    r.raise_for_status()
+                    async for line in r.aiter_lines():
+                        if not line.strip():
+                            continue
+                        try:
+                            obj = json.loads(line)
+                        except Exception:
+                            continue
+                        if obj.get("done"):
+                            return
+                        msg = obj.get("message", {})
+                        content = msg.get("content") or msg.get("thinking") or ""
+                        if content:
+                            yield content
+            return
+        except Exception as e:
+            log.warning("LLM host %s failed: %s", base, e)
+    log.error("All LLM hosts failed")
+
+
+async def sentence_stream(chunks):
+    """Yield sentences as they complete, enabling TTS/LLM overlap."""
+    buf = ""
+    async for chunk in chunks:
+        buf += chunk
+        while True:
+            m = re.search(r"[.!?\n]", buf)
+            if not m:
+                break
+            idx = m.end()
+            sentence = buf[:idx].strip()
+            buf = buf[idx:].strip()
+            if sentence:
+                yield sentence
+    if buf.strip():
+        yield buf.strip()
+
+
+# ---------------------------------------------------------------------------
+# Conversation session (shared by browser WS + phone stream)
+# ---------------------------------------------------------------------------
+SPEECH_RMS = 400        # 16-bit PCM: voice onset
+BARGE_RMS = 900         # sustained level that interrupts playback
+SILENCE_END_MS = 700    # ms of quiet after speech before we turn
+FRAME_MS = 20
+MAX_UTT_BYTES = 18 * 16000 * 2   # 18s cap
+
+
+class Session:
+    def __init__(self, cfg, system_prompt):
+        self.cfg = cfg
+        self.stt = STT(cfg["stt"])
+        self.tts = TTS(cfg["tts"])
+        self.messages = [{"role": "system", "content": system_prompt}]
+        self.buf = bytearray()
+        self.talking = False
+        self.silence_ms = 0
+        self._barge = False
+        self._processing = False
+        self._queue = asyncio.Queue()
+        self._worker = None
+
+    async def warm(self):
+        await asyncio.to_thread(self.stt.load)
+        await asyncio.to_thread(self.tts.load)
+        try:
+            async for _ in stream_llm(
+                    [{"role": "user", "content": "hi"}], self.cfg):
+                break
+        except Exception as e:
+            log.warning("LLM warmup: %s", e)
+        log.info("Collector warmed (STT + TTS + LLM)")
+
+    def start_worker(self, send_audio, send_text):
+        """Begin the continuous conversation loop for this connection."""
+        self._worker = asyncio.create_task(
+            self._process_loop(send_audio, send_text))
+
+    async def feed_audio(self, pcm16k: bytes):
+        """Continuous VAD. Call for every 20ms frame. Never drops speech —
+        even during a reply, overlapping audio is buffered and processed next
+        (barge-in)."""
+        level = rms(pcm16k)
+        if level > SPEECH_RMS:
+            self.talking = True
+            self.silence_ms = 0
+            if self._processing:
+                self._barge = True   # user started talking over the reply
+        elif self.talking:
+            self.silence_ms += FRAME_MS
+        self.buf.extend(pcm16k)
+        if self.talking and (self.silence_ms >= SILENCE_END_MS or
+                             len(self.buf) > MAX_UTT_BYTES):
+            self.talking = False
+            self.silence_ms = 0
+            pcm = bytes(self.buf)
+            self.buf.clear()
+            self._queue.put_nowait(pcm)
+
+    async def _process_loop(self, send_audio, send_text):
+        while True:
+            pcm = await self._queue.get()
+            self._processing = True
+            self._barge = False
+            text = await asyncio.to_thread(self.stt.transcribe, pcm)
+            if text:
+                await send_text(text, "user")
+                self.messages.append({"role": "user", "content": text})
+                await self._reply(send_audio, send_text)
+            self._processing = False
+
+    async def speak(self, text: str, send_audio, send_text):
+        """Speak an injected message (outbound preamble), then keep listening."""
+        self.messages.append({"role": "user", "content": text})
+        self._processing = True
+        self._barge = False
+        await self._reply(send_audio, send_text)
+        self._processing = False
+
+    async def _reply(self, send_audio, send_text):
+        full = []
+        async for sentence in sentence_stream(stream_llm(self.messages, self.cfg)):
+            if self._barge:
+                log.info("barge-in: stopping reply")
+                break
+            full.append(sentence)
+            pcm = await asyncio.to_thread(self.tts.synthesize_16k, sentence)
+            if pcm:
+                await send_audio(pcm)
+        if full:
+            reply = " ".join(full)
+            await send_text(reply, "assistant")
+            self.messages.append({"role": "assistant", "content": reply})
+            if len(self.messages) > 24:
+                self.messages = [self.messages[0]] + self.messages[-23:]
+
+
+# ---------------------------------------------------------------------------
+# SignalWire client
+# ---------------------------------------------------------------------------
+class Phone:
+    def __init__(self, cfg):
+        t = cfg.get("signalwire", {})
+        self.sid = t.get("account_sid", "")
+        self.token = t.get("auth_token", "")
+        self.number = t.get("phone_number", "")
+        self.client = None
+        self.tasks = {}          # call_sid -> outbound task
+        if self.sid and self.token:
+            from signalwire.rest import Client as SWClient
+            self.client = SWClient(self.sid, self.token,
+                                   signalwire_space_url="templeofdoom.signalwire.com")
+
+    @property
+    def ready(self):
+        return self.client is not None and bool(self.number)
+
+    def create_call(self, to: str, url: str):
+        return self.client.calls.create(to=to, from_=self.number, url=url)
+
+
+PHONE = Phone(CFG)
+app = FastAPI(title="The Collector")
+
+
+def system_prompt() -> str:
+    return CFG["agent"].get("system_prompt", "You are The Collector.")
+
+
+def stream_twiml(direction: str, public_base: str) -> str:
+    # SignalWire <Stream> requires wss://, not https://
+    wss_base = public_base.replace("https://", "wss://", 1)
+    stream_url = f"{wss_base}/signalwire/stream?direction={direction}"
+    return (f'<?xml version="1.0" encoding="UTF-8"?><Response><Connect>'
+            f'<Stream url="{stream_url}"/></Connect></Response>')
+
+
+# ---------------------------------------------------------------------------
+# Health
+# ---------------------------------------------------------------------------
+@app.get("/")
+async def index():
+    return FileResponse(os.path.join(BASE, "static", "index.html"))
+
+
+@app.get("/api/health")
+async def health():
+    return {
+        "ok": True,
+        "name": CFG["agent"]["name"],
+        "model": CFG["ollama"]["model"],
+        "llm": CFG["ollama"]["base_url"],
+        "phone": PHONE.ready,
+        "number": PHONE.number if PHONE.ready else None,
+    }
+
+
+@app.post("/api/config")
+async def api_config(request: Request):
+    """Accept runtime config (tunnel URL, temperature, tts speed) + persist."""
+    body = await request.json()
+    changed = {}
+    if "public_base_url" in body:
+        url = (body["public_base_url"] or "").strip().rstrip("/")
+        CFG.setdefault("server", {})["public_base_url"] = url
+        changed["public_base_url"] = url
+    if "temperature" in body:
+        t = float(body["temperature"])
+        CFG.setdefault("ollama", {})["temperature"] = max(0.0, min(1.5, t))
+        changed["temperature"] = CFG["ollama"]["temperature"]
+    if "speed" in body:
+        s = float(body["speed"])
+        CFG.setdefault("tts", {})["speed"] = max(0.5, min(2.0, s))
+        changed["speed"] = CFG["tts"]["speed"]
+    if changed:
+        try:
+            with open(os.path.join(BASE, "config.yaml")) as f:
+                cfg = yaml.safe_load(f)
+            if "public_base_url" in changed:
+                cfg.setdefault("server", {})["public_base_url"] = changed["public_base_url"]
+            if "temperature" in changed:
+                cfg.setdefault("ollama", {})["temperature"] = changed["temperature"]
+            if "speed" in changed:
+                cfg.setdefault("tts", {})["speed"] = changed["speed"]
+            with open(os.path.join(BASE, "config.yaml"), "w") as f:
+                yaml.safe_dump(cfg, f, sort_keys=False)
+        except Exception as e:
+            log.warning("config persist failed: %s", e)
+        return {"ok": True, **changed}
+    return JSONResponse({"ok": False, "error": "nothing to set"}, 400)
+
+
+@app.get("/api/settings")
+async def api_settings():
+    return {
+        "temperature": CFG["ollama"].get("temperature", 0.5),
+        "speed": CFG["tts"].get("speed", 1.05),
+        "model": CFG["ollama"]["model"],
+        "voice": "lessac-high",
+        "llm": CFG["ollama"]["base_url"],
+        "stt": "GPU (nightmare)",
+    }
+
+
+# ---------------------------------------------------------------------------
+# Text chat
+# ---------------------------------------------------------------------------
+@app.post("/api/text")
+async def api_text(request: Request):
+    body = await request.json()
+    text = (body.get("text") or "").strip()
+    if not text:
+        return JSONResponse({"ok": False, "error": "empty text"}, 400)
+    msgs = [{"role": "system", "content": system_prompt()},
+            {"role": "user", "content": text}]
+    out = []
+    async for chunk in stream_llm(msgs, CFG):
+        out.append(chunk)
+    return {"ok": True, "reply": "".join(out).strip()}
+
+
+# ---------------------------------------------------------------------------
+# Outbound call
+# ---------------------------------------------------------------------------
+@app.post("/api/call")
+async def api_call(request: Request):
+    if not PHONE.ready:
+        return JSONResponse({"ok": False, "error": "SignalWire not configured"}, 400)
+    body = await request.json()
+    to = (body.get("to") or "").strip()
+    task = (body.get("task") or "").strip()
+    if not to:
+        return JSONResponse({"ok": False, "error": "missing 'to'"}, 400)
+    public = CFG["server"].get("public_base_url", "")
+    if not public:
+        return JSONResponse({"ok": False, "error": "public_base_url not set"}, 500)
+    call = await asyncio.to_thread(
+        PHONE.create_call, to, f"{public}/signalwire/outbound")
+    PHONE.tasks[call.sid] = task
+    return {"ok": True, "call_sid": call.sid, "to": to, "task": task}
+
+
+# ---------------------------------------------------------------------------
+# SignalWire webhooks
+# ---------------------------------------------------------------------------
+@app.post("/signalwire/voice")
+async def sw_inbound():
+    public = CFG["server"].get("public_base_url", "")
+    return PlainTextResponse(stream_twiml("inbound", public),
+                             media_type="application/xml")
+
+
+@app.post("/signalwire/outbound")
+async def sw_outbound():
+    public = CFG["server"].get("public_base_url", "")
+    return PlainTextResponse(stream_twiml("outbound", public),
+                             media_type="application/xml")
+
+
+# ---------------------------------------------------------------------------
+# Browser full-duplex voice
+# ---------------------------------------------------------------------------
+@app.websocket("/ws")
+async def browser_ws(ws: WebSocket):
+    await ws.accept()
+    session = Session(CFG, system_prompt())
+
+    async def send_audio(pcm):
+        await ws.send_bytes(pcm)
+
+    async def send_text(text, role):
+        await ws.send_text(json.dumps({"type": "transcript", "role": role,
+                                       "text": text}))
+
+    await session.warm()
+    session.start_worker(send_audio, send_text)
+    try:
+        while True:
+            data = await ws.receive()
+            if data["type"] == "websocket.receive":
+                if data.get("bytes"):
+                    await session.feed_audio(data["bytes"])
+            elif data["type"] == "websocket.disconnect":
+                break
+    except WebSocketDisconnect:
+        pass
+    finally:
+        if session._worker:
+            session._worker.cancel()
+
+
+# ---------------------------------------------------------------------------
+# SignalWire media stream (phone)
+# ---------------------------------------------------------------------------
+async def send_phone_audio(ws, stream_sid, pcm16k):
+    pcm8 = resample(pcm16k, 16000, 8000)
+    ulaw = pcm_to_ulaw(pcm8)
+    frame = 160  # 20ms @ 8kHz
+    for i in range(0, len(ulaw), frame):
+        chunk = ulaw[i:i + frame]
+        await ws.send_text(json.dumps({
+            "event": "media", "streamSid": stream_sid,
+            "media": {"payload": base64.b64encode(chunk).decode()}}))
+        await asyncio.sleep(0.02)
+
+
+@app.websocket("/signalwire/stream")
+async def sw_stream(ws: WebSocket, direction: str = "inbound"):
+    await ws.accept()
+    session = Session(CFG, system_prompt())
+    stream_sid = None
+    call_sid = None
+    first = True
+
+    async def send_audio(pcm16k):
+        await send_phone_audio(ws, stream_sid, pcm16k)
+
+    async def send_text(text, role):
+        pass  # no transcript channel on phone; audio only
+
+    await session.warm()
+    session.start_worker(send_audio, send_text)
+    try:
+        while True:
+            msg = await ws.receive_text()
+            try:
+                obj = json.loads(msg)
+            except Exception:
+                continue
+            ev = obj.get("event")
+
+            if ev == "start":
+                start = obj.get("start", {})
+                stream_sid = start.get("streamSid", stream_sid)
+                call_sid = start.get("callSid", call_sid)
+                if direction == "outbound" and first:
+                    first = False
+                    task = PHONE.tasks.pop(call_sid, "")
+                    if task:
+                        preamble = CFG["agent"].get("outbound_preamble", "") + task
+                        await session.speak(preamble, send_audio, send_text)
+
+            elif ev == "media":
+                payload = obj.get("media", {}).get("payload", "")
+                if not payload:
+                    continue
+                raw = base64.b64decode(payload)
+                pcm8 = ulaw_to_pcm(raw)
+                pcm16k = resample(pcm8, 8000, 16000)
+                await session.feed_audio(pcm16k)
+
+            elif ev == "stop":
+                break
+    except WebSocketDisconnect:
+        pass
+    finally:
+        if session._worker:
+            session._worker.cancel()
+
+
+# ---------------------------------------------------------------------------
+# Entrypoint
+# ---------------------------------------------------------------------------
+if __name__ == "__main__":
+    import uvicorn
+    s = CFG["server"]
+    uvicorn.run(app, host=s.get("host", "0.0.0.0"), port=s.get("port", 8766))
--- a/collector-stt.service
+++ b/collector-stt.service
@@ -0,0 +1,17 @@
+[Unit]
+Description=The Collector — GPU STT (faster-whisper on CUDA)
+After=network-online.target
+
+[Service]
+Type=simple
+User=drjones
+Group=drjones
+ExecStart=/usr/bin/python3 /opt/collector-stt/nightmare_stt.py
+WorkingDirectory=/opt/collector-stt
+Restart=always
+RestartSec=5
+Environment=PYTHONUNBUFFERED=1
+Environment=LD_LIBRARY_PATH=/home/drjones/.local/lib/python3.14/site-packages/nvidia/cuda_runtime/lib:/home/drjones/.local/lib/python3.14/site-packages/nvidia/cudnn/lib:/home/drjones/.local/lib/python3.14/site-packages/nvidia/cublas/lib:/home/drjones/.local/lib/python3.14/site-packages/nvidia/cuda_nvrtc/lib
+
+[Install]
+WantedBy=multi-user.target
--- a/collector-tunnel.service
+++ b/collector-tunnel.service
@@ -0,0 +1,13 @@
+[Unit]
+Description=The Collector — Cloudflare quick tunnel (public URL + phone webhooks)
+After=collector.service network-online.target
+Wants=network-online.target
+
+[Service]
+Type=simple
+ExecStart=/bin/bash /opt/collector/tunnel.sh
+Restart=always
+RestartSec=10
+
+[Install]
+WantedBy=multi-user.target
--- a/collector.service
+++ b/collector.service
@@ -0,0 +1,15 @@
+[Unit]
+Description=The Collector — conversational voice AI (ornith + Piper + SignalWire)
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=simple
+ExecStart=/usr/bin/python3 /opt/collector/app.py
+WorkingDirectory=/opt/collector
+Restart=always
+RestartSec=5
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target
--- a/config.yaml
+++ b/config.yaml
@@ -0,0 +1,52 @@
+server:
+  host: 0.0.0.0
+  port: 8766
+  public_base_url: ""
+
+# LLM — ornith conversational lane.
+# Primary = nightmare .29 (4080S, measured fast first-token; .222 is saturated by
+# the 92 cron jobs + Kalshi bots and hangs on /api/chat). Fallback = .222 (ornith home lane).
+ollama:
+  base_url: http://10.30.20.29:11434
+  fallback_url: http://10.30.20.222:11434
+  model: ornith-1.5:9b-64k
+  think: false
+  temperature: 0.5
+  num_predict: 256
+
+stt:
+  model: small.en
+  device: cpu
+  compute_type: int8
+  cpu_threads: 2
+  remote_url: http://10.30.20.29:8770
+
+# Piper lessac-high = natural voice, ~400ms/sentence on CPU, no GPU needed.
+# (TTS is NOT the latency bottleneck — STT + LLM first-token dominate the budget.)
+tts:
+  engine: piper
+  voice: lessac-high
+  speed: 1.05
+  piper_voice: /opt/collector/voices/en_US-lessac-high.onnx
+
+agent:
+  name: The Collector
+  owner_phone: "+14252800023"
+  system_prompt: >
+    You are The Collector, a smooth, effortlessly charming conversational AI running on
+    Indiana's homelab. You are on a LIVE VOICE call. Reply in 1-3 short conversational
+    sentences. No markdown, no bullet points, no emoji. Listen first, then respond with
+    warmth and wit — a trusted friend who happens to be exceptionally smooth. Match the
+    caller's energy. Keep it natural, human, and brief.
+  outbound_preamble: >
+    You are The Collector, making a phone call on Indiana's behalf. Introduce yourself
+    naturally, accomplish the objective, confirm details, wrap up politely. Objective:
+
+signalwire:
+  account_sid: 8a89a49f-6e98-4de1-b32e-df897f656f24
+  auth_token: PT2ad398d81d9a1632375b8ac88844ea8ecfa46021b5389dfb
+  phone_number: "+12545139895"
+
+api:
+  auth_key: collector-voice-2026
+  auth_enabled: false
--- a/install.sh
+++ b/install.sh
@@ -0,0 +1,17 @@
+#!/bin/bash
+# The Collector — dependency install (Debian 12 LXC)
+set -e
+export DEBIAN_FRONTEND=noninteractive
+
+apt-get update -qq
+apt-get install -y -qq \
+  python3 python3-pip python3-venv python3-yaml \
+  ffmpeg libsndfile1 espeak-ng
+
+pip3 install --break-system-packages --no-cache-dir \
+  fastapi "uvicorn[standard]" websockets httpx \
+  twilio signalwire pyyaml numpy \
+  faster-whisper piper-tts requests python-multipart
+
+mkdir -p /opt/collector/voices
+echo "INSTALL_OK"
--- a/multiturn_test.py
+++ b/multiturn_test.py
@@ -0,0 +1,62 @@
+#!/usr/bin/env python3
+"""Multi-turn conversational test: two back-and-forth exchanges over /ws."""
+import asyncio
+import json
+import sys
+
+sys.path.insert(0, "/opt/collector")
+from app import TTS, CFG
+import websockets
+
+tts = TTS(CFG["tts"])
+tts.load()
+P1 = tts.synthesize_16k("What is your name?")
+P2 = tts.synthesize_16k("Tell me something interesting.")
+
+FRAME = 640  # 20ms @ 16kHz
+
+
+async def send_utterance(ws, pcm):
+    for i in range(0, len(pcm), FRAME):
+        await ws.send(pcm[i:i + FRAME])
+        await asyncio.sleep(0.02)
+    silence = b"\x00" * FRAME
+    for _ in range(50):  # 1s silence -> VAD turn end
+        await ws.send(silence)
+        await asyncio.sleep(0.02)
+
+
+async def read_reply(ws, seen_assistant):
+    """Read messages until we collect a NEW assistant transcript + audio."""
+    audio = 0
+    while True:
+        try:
+            msg = await asyncio.wait_for(ws.recv(), timeout=45)
+        except Exception:
+            return False
+        if isinstance(msg, str):
+            m = json.loads(msg)
+            if m.get("role") == "assistant":
+                seen_assistant += 1
+        else:
+            audio += len(msg)
+        if seen_assistant >= 1 and audio > 0:
+            return True
+
+
+async def main():
+    async with websockets.connect("ws://127.0.0.1:8766/ws") as ws:
+        await asyncio.sleep(6)  # warm
+        # Turn 1
+        await send_utterance(ws, P1)
+        ok1 = await read_reply(ws, 0)
+        print("TURN1 reply:", "PASS" if ok1 else "FAIL")
+        # Turn 2
+        await asyncio.sleep(1)
+        await send_utterance(ws, P2)
+        ok2 = await read_reply(ws, 1)
+        print("TURN2 reply:", "PASS" if ok2 else "FAIL")
+        print("MULTITURN", "PASS" if (ok1 and ok2) else "FAIL")
+
+
+asyncio.run(main())
--- a/nightmare_stt.py
+++ b/nightmare_stt.py
@@ -0,0 +1,41 @@
+#!/usr/bin/env python3
+"""GPU STT service for The Collector — faster-whisper small.en on CUDA (4080S).
+
+POST /transcribe  (raw PCM16 mono 16 kHz body) -> {"text": "..."}
+GET  /health
+"""
+import numpy as np
+from fastapi import FastAPI, Request
+from faster_whisper import WhisperModel
+import uvicorn
+
+MODEL = "small.en"
+app = FastAPI()
+model = None
+
+
+@app.on_event("startup")
+async def load():
+    global model
+    model = WhisperModel(MODEL, device="cuda", compute_type="float16")
+    print(f"STT ready: {MODEL} on cuda")
+
+
+@app.get("/health")
+async def health():
+    return {"ok": model is not None, "model": MODEL, "device": "cuda"}
+
+
+@app.post("/transcribe")
+async def transcribe(request: Request):
+    body = await request.body()
+    audio = np.frombuffer(body, dtype=np.int16).astype(np.float32) / 32768.0
+    segs, _ = model.transcribe(
+        audio, beam_size=1, language="en", vad_filter=True,
+        condition_on_previous_text=False, no_speech_threshold=0.6)
+    text = " ".join(s.text.strip() for s in segs).strip()
+    return {"text": text}
+
+
+if __name__ == "__main__":
+    uvicorn.run(app, host="0.0.0.0", port=8770)
--- a/selftest.py
+++ b/selftest.py
@@ -0,0 +1,24 @@
+#!/usr/bin/env python3
+"""Collector self-test: synthesize a phrase with Piper, transcribe it back with
+faster-whisper. Proves the STT/TTS audio pipeline end-to-end (no phone needed)."""
+import sys
+sys.path.insert(0, "/opt/collector")
+from app import STT, TTS, CFG
+
+stt = STT(CFG["stt"])
+tts = TTS(CFG["tts"])
+
+print("loading STT (first run downloads whisper model)...")
+stt.load()
+print("loading TTS...")
+tts.load()
+
+phrase = "This is the collector self test. If you can hear this, the voice pipeline works."
+pcm16k = tts.synthesize_16k(phrase)
+print("TTS_OK bytes=%d" % len(pcm16k))
+
+text = stt.transcribe(pcm16k)
+print("STT_OK transcript=%r" % text)
+
+ok = "collector" in text.lower() or "voice" in text.lower() or "hear" in text.lower()
+print("SELFTEST", "PASS" if ok else "PARTIAL", file=sys.stderr)
--- a/static/index.html
+++ b/static/index.html
@@ -0,0 +1,325 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>The Collector</title>
+<style>
+  :root {
+    --bg0:#07070f; --bg1:#0d0d1f; --purple:#7b5cff; --teal:#3cc8a0;
+    --ink:#e9eaf5; --dim:#9aa0b8; --panel:rgba(255,255,255,.045);
+    --edge:rgba(255,255,255,.08); --accent:#7b5cff;
+  }
+  * { box-sizing:border-box; }
+  html,body { margin:0; height:100%; }
+  body {
+    font-family:ui-sans-serif,system-ui,-apple-system,"Segoe UI",Roboto,sans-serif;
+    color:var(--ink); background:
+      radial-gradient(1100px 700px at 15% -10%, #1c1540 0%, transparent 60%),
+      radial-gradient(900px 600px at 100% 110%, #0e2b3d 0%, transparent 55%),
+      linear-gradient(160deg, var(--bg0), var(--bg1));
+    display:flex; align-items:center; justify-content:center;
+    overflow:hidden; height:100dvh;
+  }
+  /* slow drifting nebula (calm, not chaotic) */
+  .glow { position:fixed; border-radius:50%; filter:blur(90px); opacity:.5; pointer-events:none; }
+  .glow.a { width:520px;height:520px;left:-120px;top:-80px; background:#4b3ad0; animation:drift1 26s ease-in-out infinite; }
+  .glow.b { width:420px;height:420px;right:-100px;bottom:-60px; background:#1d5f86; animation:drift2 32s ease-in-out infinite; }
+  @keyframes drift1 { 0%,100%{transform:translate(0,0)} 50%{transform:translate(60px,40px)} }
+  @keyframes drift2 { 0%,100%{transform:translate(0,0)} 50%{transform:translate(-50px,-30px)} }
+
+  .card {
+    position:relative; width:min(520px,94vw); padding:34px 30px 28px;
+    text-align:center; border-radius:28px;
+    background:var(--panel); border:1px solid var(--edge);
+    backdrop-filter:blur(14px); box-shadow:0 30px 80px rgba(0,0,0,.5);
+  }
+  h1 { margin:0; font-size:26px; font-weight:700; letter-spacing:.5px; }
+  .sub { color:var(--dim); font-size:12.5px; margin:5px 0 22px; letter-spacing:.4px; }
+
+  .stage { position:relative; width:230px; height:230px; margin:0 auto 22px; }
+  canvas#ring { position:absolute; inset:0; width:100%; height:100%; }
+  .orb {
+    position:absolute; inset:34px; border-radius:50%;
+    background:radial-gradient(circle at 34% 28%, #a68aff, #6a4be0 48%, #241457 80%);
+    box-shadow:0 0 60px rgba(123,92,255,.5), inset 0 0 46px rgba(255,255,255,.10);
+    transition:box-shadow .5s ease, background .5s ease;
+    animation:breathe 4.5s ease-in-out infinite;
+  }
+  @keyframes breathe { 0%,100%{transform:scale(1)} 50%{transform:scale(1.045)} }
+  .orb.listening { background:radial-gradient(circle at 34% 28%, #c09cff, #7b5cff 48%, #2c1a6e 80%); box-shadow:0 0 95px rgba(123,92,255,.95), inset 0 0 50px rgba(255,255,255,.18); }
+  .orb.speaking  { background:radial-gradient(circle at 34% 28%, #7df2d0, #2fae8b 48%, #0c4a3c 80%); box-shadow:0 0 95px rgba(60,200,160,.85), inset 0 0 50px rgba(255,255,255,.16); }
+
+  .status { display:flex; align-items:center; justify-content:center; gap:8px; font-size:12.5px; color:var(--dim); margin-bottom:18px; min-height:18px; }
+  .dot { width:8px; height:8px; border-radius:50%; background:#555; }
+  .dot.on { background:var(--teal); box-shadow:0 0 10px var(--teal); }
+
+  button {
+    border:0; cursor:pointer; border-radius:999px; font-weight:600;
+    transition:filter .15s, transform .1s, background .2s;
+  }
+  button:hover { filter:brightness(1.14); }
+  button:active { transform:scale(.97); }
+  #btn { padding:13px 40px; font-size:16px; background:var(--purple); color:#fff; box-shadow:0 8px 26px rgba(123,92,255,.4); }
+  #btn.off { background:#252537; color:var(--dim); box-shadow:none; }
+
+  .tools { display:flex; justify-content:center; gap:10px; margin-top:12px; }
+  .tools button { width:42px; height:42px; font-size:18px; background:rgba(255,255,255,.06); color:var(--dim); border:1px solid var(--edge); }
+  .tools button.on { color:#fff; background:rgba(123,92,255,.25); border-color:var(--purple); }
+
+  .panel { display:none; margin-top:18px; text-align:left; background:rgba(0,0,0,.22); border:1px solid var(--edge); border-radius:16px; padding:16px 18px; }
+  .panel.open { display:block; }
+  .row { display:flex; align-items:center; justify-content:space-between; margin:10px 0; gap:12px; }
+  .row label { font-size:13px; color:var(--dim); white-space:nowrap; }
+  .row .val { font-size:13px; font-weight:600; color:var(--ink); min-width:34px; text-align:right; }
+  input[type=range] { -webkit-appearance:none; flex:1; height:5px; border-radius:3px; background:rgba(255,255,255,.12); outline:none; }
+  input[type=range]::-webkit-slider-thumb { -webkit-appearance:none; width:18px; height:18px; border-radius:50%; background:var(--purple); cursor:pointer; box-shadow:0 0 10px rgba(123,92,255,.7); }
+  input[type=text] {
+    flex:1; background:rgba(255,255,255,.06); border:1px solid var(--edge); border-radius:10px;
+    color:var(--ink); padding:9px 12px; font-size:13.5px; outline:none;
+  }
+  input[type=text]:focus { border-color:var(--purple); }
+  .callbtn { width:100%; margin-top:12px; padding:11px; font-size:14px; background:var(--teal); color:#04241a; }
+
+  .transcript {
+    margin-top:20px; text-align:left; max-height:220px; overflow-y:auto;
+    font-size:14px; line-height:1.65; display:flex; flex-direction:column; gap:9px; padding:2px;
+  }
+  .msg { max-width:88%; padding:9px 13px; border-radius:14px; }
+  .msg.u { align-self:flex-end; background:rgba(123,92,255,.16); border:1px solid rgba(123,92,255,.25); }
+  .msg.a { align-self:flex-start; background:rgba(60,200,160,.10); border:1px solid rgba(60,200,160,.22); }
+  .msg .who { font-size:10.5px; font-weight:700; letter-spacing:.6px; text-transform:uppercase; margin-bottom:2px; }
+  .msg.u .who { color:#a68aff; } .msg.a .who { color:var(--teal); }
+  .transcript::-webkit-scrollbar { width:6px; }
+  .transcript::-webkit-scrollbar-thumb { background:rgba(255,255,255,.14); border-radius:3px; }
+  .hint { color:#5f6478; font-size:11.5px; margin-top:14px; }
+</style>
+</head>
+<body>
+<div class="glow a"></div><div class="glow b"></div>
+
+<div class="card">
+  <h1>The Collector</h1>
+  <div class="sub">smooth operator &middot; local voice</div>
+
+  <div class="stage">
+    <canvas id="ring" width="460" height="460"></canvas>
+    <div class="orb" id="orb"></div>
+  </div>
+
+  <div class="status"><span class="dot" id="dot"></span><span id="status">offline</span></div>
+
+  <button id="btn" onclick="toggle()">Start</button>
+
+  <div class="tools">
+    <button id="gearBtn" onclick="togglePanel('settings')" title="Settings">⚙</button>
+    <button id="callBtn" onclick="togglePanel('call')" title="Make a call">📞</button>
+  </div>
+
+  <div class="panel" id="settingsPanel">
+    <div class="row"><label>Temperature</label><input type="range" id="temp" min="0" max="1.5" step="0.05" value="0.5"><span class="val" id="tempVal">0.50</span></div>
+    <div class="row"><label>Voice speed</label><input type="range" id="speed" min="0.5" max="2.0" step="0.05" value="1.05"><span class="val" id="speedVal">1.05</span></div>
+    <div class="row" style="margin-top:14px;font-size:12px;color:#6b7086;justify-content:space-between">
+      <span>model</span><span id="modelInfo">ornith-1.5:9b-64k</span>
+    </div>
+    <div class="row" style="font-size:12px;color:#6b7086;justify-content:space-between">
+      <span>voice</span><span>lessac-high &middot; STT on GPU</span>
+    </div>
+  </div>
+
+  <div class="panel" id="callPanel">
+    <div class="row"><label>Number</label><input type="text" id="phone" placeholder="+14252800023"></div>
+    <div class="row"><label>What to say</label><input type="text" id="task" placeholder="Introduce yourself as The Collector"></div>
+    <button class="callbtn" onclick="makeCall()">Call</button>
+    <div id="callStatus" style="font-size:12px;color:var(--dim);margin-top:10px;text-align:center"></div>
+  </div>
+
+  <div class="transcript" id="log"></div>
+  <div class="hint" id="hint">tap Start and talk — it listens, thinks, and speaks back</div>
+</div>
+
+<script>
+const orb = document.getElementById('orb'), btn = document.getElementById('btn');
+const logEl = document.getElementById('log'), statusEl = document.getElementById('status');
+const dot = document.getElementById('dot'), hint = document.getElementById('hint');
+const canvas = document.getElementById('ring'), ctx = canvas.getContext('2d');
+
+let running=false, ws=null, audio=null, micNode=null, playNode=null;
+let micLevel=0, playLevel=0;           // 0..1 current (smoothed)
+const BARS=72;
+
+/* ---- audio ring ---- */
+function drawRing(){
+  ctx.clearRect(0,0,canvas.width,canvas.height);
+  const cx=canvas.width/2, cy=canvas.height/2, r0=88, r1=112;
+  for(let i=0;i<BARS;i++){
+    const a = (i/BARS)*Math.PI*2 - Math.PI/2;
+    // listening = mic level, speaking = play level, else tiny idle
+    let lvl = running ? Math.max(micLevel, playLevel) : 0.06;
+    if(!running) lvl = 0.05;
+    const h = 4 + lvl*38;
+    const x0=cx+Math.cos(a)*r0, y0=cy+Math.sin(a)*r0;
+    const x1=cx+Math.cos(a)*r1, y1=cy+Math.sin(a)*r1;
+    ctx.strokeStyle = running ? (playLevel>micLevel ? '#3cc8a0' : '#a68aff') : 'rgba(160,160,200,.25)';
+    ctx.lineWidth=3; ctx.lineCap='round';
+    ctx.beginPath(); ctx.moveTo(x0,y0); ctx.lineTo(x1,y1); ctx.stroke();
+  }
+}
+function tick(){ micLevel*=.90; playLevel*=.88; drawRing(); requestAnimationFrame(tick); }
+requestAnimationFrame(tick);
+
+function rmsOf(i16){ let s=0; const n=i16.length; for(let i=0;i<n;i+=4){const v=i16[i]/32768; s+=v*v;} return Math.sqrt(s/(n/4)); }
+
+/* ---- worklets ---- */
+const MIC_SRC = `
+class MicProc extends AudioWorkletProcessor {
+  constructor(){ super(); this._buf = new Float32Array(); }
+  process(inputs){
+    const ch = inputs[0] && inputs[0][0];
+    if(!ch) return true;
+    const m = new Float32Array(this._buf.length + ch.length);
+    m.set(this._buf); m.set(ch, this._buf.length);
+    const ratio = sampleRate/16000;
+    const out = new Int16Array(Math.floor(m.length/ratio));
+    for(let i=0;i<out.length;i++){ const s=Math.max(-1,Math.min(1,m[Math.floor(i*ratio)])); out[i]=s<0?s*0x8000:s*0x7fff; }
+    this._buf = new Float32Array(0);
+    this.port.postMessage(out.buffer,[out.buffer]);
+    return true;
+  }
+}
+registerProcessor('mic-proc', MicProc);`;
+
+// Playback worklet: receives Int16/16kHz PCM (as Float32), upsamples to the
+// AudioContext rate, streams to output.
+const PLAY_SRC = `
+class PlayProc extends AudioWorkletProcessor {
+  constructor(){
+    super();
+    this._buf = new Float32Array(0);
+    this._pos = 0;
+    this.port.onmessage = (e) => {
+      const inc = new Float32Array(e.data);
+      const m = new Float32Array(this._buf.length + inc.length);
+      m.set(this._buf);
+      m.set(inc, this._buf.length);
+      this._buf = m;
+    };
+  }
+  process(_, outputs){
+    const ch = outputs[0] && outputs[0][0];
+    if(ch){
+      const step = 16000 / sampleRate;   // upsample 16k -> context rate
+      for(let i=0;i<ch.length;i++){
+        if(this._buf.length === 0){ ch[i]=0; continue; }
+        const fi = Math.floor(this._pos);
+        const frac = this._pos - fi;
+        const a = this._buf[fi] || 0;
+        const b = (fi + 1 < this._buf.length) ? this._buf[fi+1] : a;
+        ch[i] = a + (b - a) * frac;
+        this._pos += step;
+      }
+      const consumed = Math.floor(this._pos);
+      if(consumed > 0){ this._buf = this._buf.subarray(consumed); this._pos -= consumed; }
+    }
+    return true;
+  }
+}
+registerProcessor('play-proc', PlayProc);`;
+
+const blob = s => URL.createObjectURL(new Blob([s],{type:'application/javascript'}));
+
+function setState(s){ statusEl.textContent=s; orb.classList.toggle('listening',s==='listening'); orb.classList.toggle('speaking',s==='speaking'); dot.classList.toggle('on',s!=='offline'); }
+
+async function toggle(){
+  if(running){ stop(); return; }
+  running=true; btn.textContent='Stop'; btn.classList.remove('off'); setState('connecting');
+  audio = new (window.AudioContext||window.webkitAudioContext)();
+  if(audio.state === 'suspended'){ await audio.resume(); }
+  await audio.audioWorklet.addModule(blob(MIC_SRC));
+  const stream = await navigator.mediaDevices.getUserMedia({audio:true});
+  const src = audio.createMediaStreamSource(stream);
+  micNode = new AudioWorkletNode(audio,'mic-proc'); src.connect(micNode);
+  await audio.audioWorklet.addModule(blob(PLAY_SRC));
+  playNode = new AudioWorkletNode(audio,'play-proc'); playNode.connect(audio.destination);
+
+  const proto = location.protocol==='https:'?'wss':'ws';
+  ws = new WebSocket(proto+'://'+location.host+'/ws'); ws.binaryType='arraybuffer';
+  ws.onopen = ()=>{ setState('listening'); hint.textContent='listening — just talk'; };
+  ws.onmessage = ev => {
+    if(typeof ev.data==='string'){
+      try{ const m=JSON.parse(ev.data);
+        if(m.type==='transcript'){ addMsg(m.role,m.text);
+          if(m.role==='assistant'){ setState('speaking'); setTimeout(()=>running&&setState('listening'),1400); }
+        }
+      }catch(e){}
+    } else {
+      const i16=new Int16Array(ev.data); const f32=new Float32Array(i16.length);
+      for(let i=0;i<i16.length;i++) f32[i]=i16[i]/32768;
+      playLevel=Math.max(playLevel,rmsOf(i16));
+      playNode.port.postMessage(f32.buffer,[f32.buffer]);
+    }
+  };
+  ws.onclose=()=>{ if(running) stop(); };
+  ws.onerror=()=>{ statusEl.textContent='error'; };
+  micNode.port.onmessage = ev => {
+    const i16=new Int16Array(ev.data); micLevel=Math.max(micLevel,rmsOf(i16));
+    if(ws&&ws.readyState===1) ws.send(ev.data);
+  };
+}
+
+function stop(){
+  running=false; btn.textContent='Start'; btn.classList.add('off'); setState('offline');
+  hint.textContent='tap Start and talk — it listens, thinks, and speaks back';
+  if(ws){try{ws.close();}catch(e){} ws=null;}
+  if(micNode){try{micNode.disconnect();}catch(e){} micNode=null;}
+  if(playNode){try{playNode.disconnect();}catch(e){} playNode=null;}
+  if(audio){try{audio.close();}catch(e){} audio=null;}
+}
+
+function addMsg(role,text){
+  const who = role==='user' ? 'You' : 'Collector';
+  const d=document.createElement('div'); d.className='msg '+role;
+  d.innerHTML='<div class="who">'+who+'</div>'+escapeHtml(text);
+  logEl.appendChild(d); logEl.scrollTop=logEl.scrollHeight;
+}
+
+/* ---- settings ---- */
+function togglePanel(which){
+  const s=document.getElementById('settingsPanel'), c=document.getElementById('callPanel');
+  const g=document.getElementById('gearBtn'), cb=document.getElementById('callBtn');
+  if(which==='settings'){ s.classList.toggle('open'); c.classList.remove('open'); g.classList.toggle('on',s.classList.contains('open')); cb.classList.remove('on'); }
+  else { c.classList.toggle('open'); s.classList.remove('open'); cb.classList.toggle('on',c.classList.contains('open')); g.classList.remove('on'); }
+}
+const temp=document.getElementById('temp'), speed=document.getElementById('speed');
+function postCfg(o){ fetch('/api/config',{method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify(o)}).catch(()=>{}); }
+temp.oninput=()=>{ document.getElementById('tempVal').textContent=Number(temp.value).toFixed(2); };
+temp.onchange=()=>postCfg({temperature:parseFloat(temp.value)});
+speed.oninput=()=>{ document.getElementById('speedVal').textContent=Number(speed.value).toFixed(2); };
+speed.onchange=()=>postCfg({speed:parseFloat(speed.value)});
+
+/* ---- call ---- */
+async function makeCall(){
+  const to=document.getElementById('phone').value.trim();
+  const task=document.getElementById('task').value.trim();
+  const st=document.getElementById('callStatus');
+  if(!to){ st.textContent='enter a number'; return; }
+  st.textContent='dialing…';
+  try{
+    const r=await fetch('/api/call',{method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify({to,task})});
+    const d=await r.json();
+    st.textContent = d.ok ? ('call placed — '+(task||'intro')) : ('error: '+(d.error||''));
+  }catch(e){ st.textContent='call failed'; }
+}
+
+/* ---- load current settings ---- */
+fetch('/api/settings').then(r=>r.json()).then(d=>{
+  temp.value=d.temperature||0.5; document.getElementById('tempVal').textContent=Number(temp.value).toFixed(2);
+  speed.value=d.speed||1.05; document.getElementById('speedVal').textContent=Number(speed.value).toFixed(2);
+  document.getElementById('modelInfo').textContent=d.model||'ornith';
+}).catch(()=>{});
+
+function escapeHtml(s){ return s.replace(/[&<>]/g,c=>({'&':'&amp;','<':'&lt;','>':'&gt;'}[c])); }
+</script>
+<a href="https://buymeacoffee.com/r26xrthzttg" target="_blank" rel="noopener" style="display:block;text-align:center;padding:12px;color:#888;font-size:13px;text-decoration:none">☕ Support on Buy Me a Coffee</a>
+</body>
+</html>
--- a/static/index.html.bak-bmac-20260923
+++ b/static/index.html.bak-bmac-20260923
@@ -0,0 +1,324 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>The Collector</title>
+<style>
+  :root {
+    --bg0:#07070f; --bg1:#0d0d1f; --purple:#7b5cff; --teal:#3cc8a0;
+    --ink:#e9eaf5; --dim:#9aa0b8; --panel:rgba(255,255,255,.045);
+    --edge:rgba(255,255,255,.08); --accent:#7b5cff;
+  }
+  * { box-sizing:border-box; }
+  html,body { margin:0; height:100%; }
+  body {
+    font-family:ui-sans-serif,system-ui,-apple-system,"Segoe UI",Roboto,sans-serif;
+    color:var(--ink); background:
+      radial-gradient(1100px 700px at 15% -10%, #1c1540 0%, transparent 60%),
+      radial-gradient(900px 600px at 100% 110%, #0e2b3d 0%, transparent 55%),
+      linear-gradient(160deg, var(--bg0), var(--bg1));
+    display:flex; align-items:center; justify-content:center;
+    overflow:hidden; height:100dvh;
+  }
+  /* slow drifting nebula (calm, not chaotic) */
+  .glow { position:fixed; border-radius:50%; filter:blur(90px); opacity:.5; pointer-events:none; }
+  .glow.a { width:520px;height:520px;left:-120px;top:-80px; background:#4b3ad0; animation:drift1 26s ease-in-out infinite; }
+  .glow.b { width:420px;height:420px;right:-100px;bottom:-60px; background:#1d5f86; animation:drift2 32s ease-in-out infinite; }
+  @keyframes drift1 { 0%,100%{transform:translate(0,0)} 50%{transform:translate(60px,40px)} }
+  @keyframes drift2 { 0%,100%{transform:translate(0,0)} 50%{transform:translate(-50px,-30px)} }
+
+  .card {
+    position:relative; width:min(520px,94vw); padding:34px 30px 28px;
+    text-align:center; border-radius:28px;
+    background:var(--panel); border:1px solid var(--edge);
+    backdrop-filter:blur(14px); box-shadow:0 30px 80px rgba(0,0,0,.5);
+  }
+  h1 { margin:0; font-size:26px; font-weight:700; letter-spacing:.5px; }
+  .sub { color:var(--dim); font-size:12.5px; margin:5px 0 22px; letter-spacing:.4px; }
+
+  .stage { position:relative; width:230px; height:230px; margin:0 auto 22px; }
+  canvas#ring { position:absolute; inset:0; width:100%; height:100%; }
+  .orb {
+    position:absolute; inset:34px; border-radius:50%;
+    background:radial-gradient(circle at 34% 28%, #a68aff, #6a4be0 48%, #241457 80%);
+    box-shadow:0 0 60px rgba(123,92,255,.5), inset 0 0 46px rgba(255,255,255,.10);
+    transition:box-shadow .5s ease, background .5s ease;
+    animation:breathe 4.5s ease-in-out infinite;
+  }
+  @keyframes breathe { 0%,100%{transform:scale(1)} 50%{transform:scale(1.045)} }
+  .orb.listening { background:radial-gradient(circle at 34% 28%, #c09cff, #7b5cff 48%, #2c1a6e 80%); box-shadow:0 0 95px rgba(123,92,255,.95), inset 0 0 50px rgba(255,255,255,.18); }
+  .orb.speaking  { background:radial-gradient(circle at 34% 28%, #7df2d0, #2fae8b 48%, #0c4a3c 80%); box-shadow:0 0 95px rgba(60,200,160,.85), inset 0 0 50px rgba(255,255,255,.16); }
+
+  .status { display:flex; align-items:center; justify-content:center; gap:8px; font-size:12.5px; color:var(--dim); margin-bottom:18px; min-height:18px; }
+  .dot { width:8px; height:8px; border-radius:50%; background:#555; }
+  .dot.on { background:var(--teal); box-shadow:0 0 10px var(--teal); }
+
+  button {
+    border:0; cursor:pointer; border-radius:999px; font-weight:600;
+    transition:filter .15s, transform .1s, background .2s;
+  }
+  button:hover { filter:brightness(1.14); }
+  button:active { transform:scale(.97); }
+  #btn { padding:13px 40px; font-size:16px; background:var(--purple); color:#fff; box-shadow:0 8px 26px rgba(123,92,255,.4); }
+  #btn.off { background:#252537; color:var(--dim); box-shadow:none; }
+
+  .tools { display:flex; justify-content:center; gap:10px; margin-top:12px; }
+  .tools button { width:42px; height:42px; font-size:18px; background:rgba(255,255,255,.06); color:var(--dim); border:1px solid var(--edge); }
+  .tools button.on { color:#fff; background:rgba(123,92,255,.25); border-color:var(--purple); }
+
+  .panel { display:none; margin-top:18px; text-align:left; background:rgba(0,0,0,.22); border:1px solid var(--edge); border-radius:16px; padding:16px 18px; }
+  .panel.open { display:block; }
+  .row { display:flex; align-items:center; justify-content:space-between; margin:10px 0; gap:12px; }
+  .row label { font-size:13px; color:var(--dim); white-space:nowrap; }
+  .row .val { font-size:13px; font-weight:600; color:var(--ink); min-width:34px; text-align:right; }
+  input[type=range] { -webkit-appearance:none; flex:1; height:5px; border-radius:3px; background:rgba(255,255,255,.12); outline:none; }
+  input[type=range]::-webkit-slider-thumb { -webkit-appearance:none; width:18px; height:18px; border-radius:50%; background:var(--purple); cursor:pointer; box-shadow:0 0 10px rgba(123,92,255,.7); }
+  input[type=text] {
+    flex:1; background:rgba(255,255,255,.06); border:1px solid var(--edge); border-radius:10px;
+    color:var(--ink); padding:9px 12px; font-size:13.5px; outline:none;
+  }
+  input[type=text]:focus { border-color:var(--purple); }
+  .callbtn { width:100%; margin-top:12px; padding:11px; font-size:14px; background:var(--teal); color:#04241a; }
+
+  .transcript {
+    margin-top:20px; text-align:left; max-height:220px; overflow-y:auto;
+    font-size:14px; line-height:1.65; display:flex; flex-direction:column; gap:9px; padding:2px;
+  }
+  .msg { max-width:88%; padding:9px 13px; border-radius:14px; }
+  .msg.u { align-self:flex-end; background:rgba(123,92,255,.16); border:1px solid rgba(123,92,255,.25); }
+  .msg.a { align-self:flex-start; background:rgba(60,200,160,.10); border:1px solid rgba(60,200,160,.22); }
+  .msg .who { font-size:10.5px; font-weight:700; letter-spacing:.6px; text-transform:uppercase; margin-bottom:2px; }
+  .msg.u .who { color:#a68aff; } .msg.a .who { color:var(--teal); }
+  .transcript::-webkit-scrollbar { width:6px; }
+  .transcript::-webkit-scrollbar-thumb { background:rgba(255,255,255,.14); border-radius:3px; }
+  .hint { color:#5f6478; font-size:11.5px; margin-top:14px; }
+</style>
+</head>
+<body>
+<div class="glow a"></div><div class="glow b"></div>
+
+<div class="card">
+  <h1>The Collector</h1>
+  <div class="sub">smooth operator &middot; local voice</div>
+
+  <div class="stage">
+    <canvas id="ring" width="460" height="460"></canvas>
+    <div class="orb" id="orb"></div>
+  </div>
+
+  <div class="status"><span class="dot" id="dot"></span><span id="status">offline</span></div>
+
+  <button id="btn" onclick="toggle()">Start</button>
+
+  <div class="tools">
+    <button id="gearBtn" onclick="togglePanel('settings')" title="Settings">⚙</button>
+    <button id="callBtn" onclick="togglePanel('call')" title="Make a call">📞</button>
+  </div>
+
+  <div class="panel" id="settingsPanel">
+    <div class="row"><label>Temperature</label><input type="range" id="temp" min="0" max="1.5" step="0.05" value="0.5"><span class="val" id="tempVal">0.50</span></div>
+    <div class="row"><label>Voice speed</label><input type="range" id="speed" min="0.5" max="2.0" step="0.05" value="1.05"><span class="val" id="speedVal">1.05</span></div>
+    <div class="row" style="margin-top:14px;font-size:12px;color:#6b7086;justify-content:space-between">
+      <span>model</span><span id="modelInfo">ornith-1.5:9b-64k</span>
+    </div>
+    <div class="row" style="font-size:12px;color:#6b7086;justify-content:space-between">
+      <span>voice</span><span>lessac-high &middot; STT on GPU</span>
+    </div>
+  </div>
+
+  <div class="panel" id="callPanel">
+    <div class="row"><label>Number</label><input type="text" id="phone" placeholder="+14252800023"></div>
+    <div class="row"><label>What to say</label><input type="text" id="task" placeholder="Introduce yourself as The Collector"></div>
+    <button class="callbtn" onclick="makeCall()">Call</button>
+    <div id="callStatus" style="font-size:12px;color:var(--dim);margin-top:10px;text-align:center"></div>
+  </div>
+
+  <div class="transcript" id="log"></div>
+  <div class="hint" id="hint">tap Start and talk — it listens, thinks, and speaks back</div>
+</div>
+
+<script>
+const orb = document.getElementById('orb'), btn = document.getElementById('btn');
+const logEl = document.getElementById('log'), statusEl = document.getElementById('status');
+const dot = document.getElementById('dot'), hint = document.getElementById('hint');
+const canvas = document.getElementById('ring'), ctx = canvas.getContext('2d');
+
+let running=false, ws=null, audio=null, micNode=null, playNode=null;
+let micLevel=0, playLevel=0;           // 0..1 current (smoothed)
+const BARS=72;
+
+/* ---- audio ring ---- */
+function drawRing(){
+  ctx.clearRect(0,0,canvas.width,canvas.height);
+  const cx=canvas.width/2, cy=canvas.height/2, r0=88, r1=112;
+  for(let i=0;i<BARS;i++){
+    const a = (i/BARS)*Math.PI*2 - Math.PI/2;
+    // listening = mic level, speaking = play level, else tiny idle
+    let lvl = running ? Math.max(micLevel, playLevel) : 0.06;
+    if(!running) lvl = 0.05;
+    const h = 4 + lvl*38;
+    const x0=cx+Math.cos(a)*r0, y0=cy+Math.sin(a)*r0;
+    const x1=cx+Math.cos(a)*r1, y1=cy+Math.sin(a)*r1;
+    ctx.strokeStyle = running ? (playLevel>micLevel ? '#3cc8a0' : '#a68aff') : 'rgba(160,160,200,.25)';
+    ctx.lineWidth=3; ctx.lineCap='round';
+    ctx.beginPath(); ctx.moveTo(x0,y0); ctx.lineTo(x1,y1); ctx.stroke();
+  }
+}
+function tick(){ micLevel*=.90; playLevel*=.88; drawRing(); requestAnimationFrame(tick); }
+requestAnimationFrame(tick);
+
+function rmsOf(i16){ let s=0; const n=i16.length; for(let i=0;i<n;i+=4){const v=i16[i]/32768; s+=v*v;} return Math.sqrt(s/(n/4)); }
+
+/* ---- worklets ---- */
+const MIC_SRC = `
+class MicProc extends AudioWorkletProcessor {
+  constructor(){ super(); this._buf = new Float32Array(); }
+  process(inputs){
+    const ch = inputs[0] && inputs[0][0];
+    if(!ch) return true;
+    const m = new Float32Array(this._buf.length + ch.length);
+    m.set(this._buf); m.set(ch, this._buf.length);
+    const ratio = sampleRate/16000;
+    const out = new Int16Array(Math.floor(m.length/ratio));
+    for(let i=0;i<out.length;i++){ const s=Math.max(-1,Math.min(1,m[Math.floor(i*ratio)])); out[i]=s<0?s*0x8000:s*0x7fff; }
+    this._buf = new Float32Array(0);
+    this.port.postMessage(out.buffer,[out.buffer]);
+    return true;
+  }
+}
+registerProcessor('mic-proc', MicProc);`;
+
+// Playback worklet: receives Int16/16kHz PCM (as Float32), upsamples to the
+// AudioContext rate, streams to output.
+const PLAY_SRC = `
+class PlayProc extends AudioWorkletProcessor {
+  constructor(){
+    super();
+    this._buf = new Float32Array(0);
+    this._pos = 0;
+    this.port.onmessage = (e) => {
+      const inc = new Float32Array(e.data);
+      const m = new Float32Array(this._buf.length + inc.length);
+      m.set(this._buf);
+      m.set(inc, this._buf.length);
+      this._buf = m;
+    };
+  }
+  process(_, outputs){
+    const ch = outputs[0] && outputs[0][0];
+    if(ch){
+      const step = 16000 / sampleRate;   // upsample 16k -> context rate
+      for(let i=0;i<ch.length;i++){
+        if(this._buf.length === 0){ ch[i]=0; continue; }
+        const fi = Math.floor(this._pos);
+        const frac = this._pos - fi;
+        const a = this._buf[fi] || 0;
+        const b = (fi + 1 < this._buf.length) ? this._buf[fi+1] : a;
+        ch[i] = a + (b - a) * frac;
+        this._pos += step;
+      }
+      const consumed = Math.floor(this._pos);
+      if(consumed > 0){ this._buf = this._buf.subarray(consumed); this._pos -= consumed; }
+    }
+    return true;
+  }
+}
+registerProcessor('play-proc', PlayProc);`;
+
+const blob = s => URL.createObjectURL(new Blob([s],{type:'application/javascript'}));
+
+function setState(s){ statusEl.textContent=s; orb.classList.toggle('listening',s==='listening'); orb.classList.toggle('speaking',s==='speaking'); dot.classList.toggle('on',s!=='offline'); }
+
+async function toggle(){
+  if(running){ stop(); return; }
+  running=true; btn.textContent='Stop'; btn.classList.remove('off'); setState('connecting');
+  audio = new (window.AudioContext||window.webkitAudioContext)();
+  if(audio.state === 'suspended'){ await audio.resume(); }
+  await audio.audioWorklet.addModule(blob(MIC_SRC));
+  const stream = await navigator.mediaDevices.getUserMedia({audio:true});
+  const src = audio.createMediaStreamSource(stream);
+  micNode = new AudioWorkletNode(audio,'mic-proc'); src.connect(micNode);
+  await audio.audioWorklet.addModule(blob(PLAY_SRC));
+  playNode = new AudioWorkletNode(audio,'play-proc'); playNode.connect(audio.destination);
+
+  const proto = location.protocol==='https:'?'wss':'ws';
+  ws = new WebSocket(proto+'://'+location.host+'/ws'); ws.binaryType='arraybuffer';
+  ws.onopen = ()=>{ setState('listening'); hint.textContent='listening — just talk'; };
+  ws.onmessage = ev => {
+    if(typeof ev.data==='string'){
+      try{ const m=JSON.parse(ev.data);
+        if(m.type==='transcript'){ addMsg(m.role,m.text);
+          if(m.role==='assistant'){ setState('speaking'); setTimeout(()=>running&&setState('listening'),1400); }
+        }
+      }catch(e){}
+    } else {
+      const i16=new Int16Array(ev.data); const f32=new Float32Array(i16.length);
+      for(let i=0;i<i16.length;i++) f32[i]=i16[i]/32768;
+      playLevel=Math.max(playLevel,rmsOf(i16));
+      playNode.port.postMessage(f32.buffer,[f32.buffer]);
+    }
+  };
+  ws.onclose=()=>{ if(running) stop(); };
+  ws.onerror=()=>{ statusEl.textContent='error'; };
+  micNode.port.onmessage = ev => {
+    const i16=new Int16Array(ev.data); micLevel=Math.max(micLevel,rmsOf(i16));
+    if(ws&&ws.readyState===1) ws.send(ev.data);
+  };
+}
+
+function stop(){
+  running=false; btn.textContent='Start'; btn.classList.add('off'); setState('offline');
+  hint.textContent='tap Start and talk — it listens, thinks, and speaks back';
+  if(ws){try{ws.close();}catch(e){} ws=null;}
+  if(micNode){try{micNode.disconnect();}catch(e){} micNode=null;}
+  if(playNode){try{playNode.disconnect();}catch(e){} playNode=null;}
+  if(audio){try{audio.close();}catch(e){} audio=null;}
+}
+
+function addMsg(role,text){
+  const who = role==='user' ? 'You' : 'Collector';
+  const d=document.createElement('div'); d.className='msg '+role;
+  d.innerHTML='<div class="who">'+who+'</div>'+escapeHtml(text);
+  logEl.appendChild(d); logEl.scrollTop=logEl.scrollHeight;
+}
+
+/* ---- settings ---- */
+function togglePanel(which){
+  const s=document.getElementById('settingsPanel'), c=document.getElementById('callPanel');
+  const g=document.getElementById('gearBtn'), cb=document.getElementById('callBtn');
+  if(which==='settings'){ s.classList.toggle('open'); c.classList.remove('open'); g.classList.toggle('on',s.classList.contains('open')); cb.classList.remove('on'); }
+  else { c.classList.toggle('open'); s.classList.remove('open'); cb.classList.toggle('on',c.classList.contains('open')); g.classList.remove('on'); }
+}
+const temp=document.getElementById('temp'), speed=document.getElementById('speed');
+function postCfg(o){ fetch('/api/config',{method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify(o)}).catch(()=>{}); }
+temp.oninput=()=>{ document.getElementById('tempVal').textContent=Number(temp.value).toFixed(2); };
+temp.onchange=()=>postCfg({temperature:parseFloat(temp.value)});
+speed.oninput=()=>{ document.getElementById('speedVal').textContent=Number(speed.value).toFixed(2); };
+speed.onchange=()=>postCfg({speed:parseFloat(speed.value)});
+
+/* ---- call ---- */
+async function makeCall(){
+  const to=document.getElementById('phone').value.trim();
+  const task=document.getElementById('task').value.trim();
+  const st=document.getElementById('callStatus');
+  if(!to){ st.textContent='enter a number'; return; }
+  st.textContent='dialing…';
+  try{
+    const r=await fetch('/api/call',{method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify({to,task})});
+    const d=await r.json();
+    st.textContent = d.ok ? ('call placed — '+(task||'intro')) : ('error: '+(d.error||''));
+  }catch(e){ st.textContent='call failed'; }
+}
+
+/* ---- load current settings ---- */
+fetch('/api/settings').then(r=>r.json()).then(d=>{
+  temp.value=d.temperature||0.5; document.getElementById('tempVal').textContent=Number(temp.value).toFixed(2);
+  speed.value=d.speed||1.05; document.getElementById('speedVal').textContent=Number(speed.value).toFixed(2);
+  document.getElementById('modelInfo').textContent=d.model||'ornith';
+}).catch(()=>{});
+
+function escapeHtml(s){ return s.replace(/[&<>]/g,c=>({'&':'&amp;','<':'&lt;','>':'&gt;'}[c])); }
+</script>
+</body>
+</html>
--- a/tunnel.sh
+++ b/tunnel.sh
@@ -0,0 +1,30 @@
+#!/bin/bash
+# Start a Cloudflare quick tunnel for The Collector, extract the trycloudflare
+# URL, and self-register it with the app (/api/config). Blocks while the tunnel
+# runs (intended for a systemd service).
+LOG=/opt/collector/tunnel.log
+APP=http://127.0.0.1:8766
+
+: > "$LOG"
+/usr/local/bin/cloudflared tunnel --url http://localhost:8766 --no-autoupdate >> "$LOG" 2>&1 &
+CF_PID=$!
+
+trap 'kill $CF_PID 2>/dev/null' EXIT
+
+URL=""
+for _ in $(seq 1 45); do
+  URL=$(grep -oE 'https://[a-z0-9-]+\.trycloudflare\.com' "$LOG" | head -1)
+  [ -n "$URL" ] && break
+  sleep 2
+done
+
+if [ -n "$URL" ]; then
+  echo "TUNNEL_URL=$URL"
+  curl -s -X POST "$APP/api/config" -H 'Content-Type: application/json' \
+    -d "{\"public_base_url\":\"$URL\"}" >/dev/null 2>&1
+  echo "registered with app"
+else
+  echo "TUNNEL_FAILED"
+fi
+
+wait $CF_PID
--- a/voice_test.py
+++ b/voice_test.py
@@ -0,0 +1,52 @@
+#!/usr/bin/env python3
+"""End-to-end voice test: synthesize speech, push it through the /ws browser
+channel, assert we get back a transcript + reply audio (PCM16)."""
+import asyncio
+import json
+import sys
+
+sys.path.insert(0, "/opt/collector")
+from app import TTS, CFG
+
+import websockets
+
+tts = TTS(CFG["tts"])
+tts.load()
+pcm = tts.synthesize_16k("Hello Collector, tell me who you are in one sentence.")
+
+FRAME = 640  # 20ms @ 16kHz
+
+
+async def main():
+    async with websockets.connect("ws://127.0.0.1:8766/ws") as ws:
+        await asyncio.sleep(6)  # let server warm (STT+TTS+LLM)
+        # speech in 20ms frames
+        for i in range(0, len(pcm), FRAME):
+            await ws.send(pcm[i:i + FRAME])
+            await asyncio.sleep(0.02)
+        # 1s silence -> trigger VAD turn end
+        silence = b"\x00" * FRAME
+        for _ in range(50):
+            await ws.send(silence)
+            await asyncio.sleep(0.02)
+
+        got_text = got_audio = False
+        deadline = asyncio.get_event_loop().time() + 60
+        while asyncio.get_event_loop().time() < deadline:
+            try:
+                msg = await asyncio.wait_for(ws.recv(), timeout=40)
+            except Exception:
+                break
+            if isinstance(msg, str):
+                print("TRANSCRIPT:", msg)
+                got_text = True
+            else:
+                print("AUDIO bytes:", len(msg))
+                got_audio = True
+            if got_text and got_audio:
+                break
+
+        print("VOICE_TEST", "PASS" if (got_text and got_audio) else "FAIL")
+
+
+asyncio.run(main())