diff --git a/.env.example b/.env.example
new file mode 100644
index 0000000..40c20de
--- /dev/null
+++ b/.env.example
@@ -0,0 +1,62 @@
+# Copy to .env and fill in. run.sh auto-loads it.
+
+# ── Public ingress (Twilio dials this back) ──────────────────────────────────
+# Public hostname; nginx terminates TLS here and proxies to the app. Must match the
+# Twilio webhook host (Twilio signs https://PUBLIC_HOST/voice).
+PUBLIC_HOST=voip.activeblue.net
+PORT=8200
+# App bind address. Default 127.0.0.1 (nginx proxies in locally) — not exposed on LAN.
+BIND_HOST=127.0.0.1
+
+# ── Twilio ───────────────────────────────────────────────────────────────────
+# From console.twilio.com. Account SID + a Standard API Key (scoped to this app,
+# revocable independently). The Auth Token stays in the Twilio console only — never on
+# this server. Create the key under Account → API Keys → Create Standard key, name it
+# avc-phone-agent-prod; the Secret is shown once. Used to auto-hang-up the carrier leg
+# and validate inbound webhook signatures.
+TWILIO_ACCOUNT_SID=ACxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
+TWILIO_API_KEY_SID=SKxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
+TWILIO_API_KEY_SECRET=your_api_key_secret_here
+# Inbound webhook signature validation is ON whenever TWILIO_API_KEY_SECRET is set.
+# Set to false only for local testing without real Twilio requests.
+TWILIO_VALIDATE=true
+# Shared secret embedded in the Media Stream wss URL to gate /ws. Set a stable random
+# value (e.g. `openssl rand -base64 24`); if blank, one is generated per process start.
+STREAM_TOKEN=
+
+# ── Odoo appointment integration ─────────────────────────────────────────────
+# Leave ODOO_USER/ODOO_API_KEY blank to disable Odoo and log requests to JSONL only.
+# Same creds the activeblue-agent container uses (docker inspect activeblue-agent).
+# Verified working against db1 with ODOO_TARGET=crm.
+ODOO_URL=http://localhost:8069
+ODOO_DB=db1
+ODOO_USER=mr.garcia09@gmail.com
+ODOO_API_KEY=
+ODOO_TARGET=crm          # crm = callback lead (recommended) | calendar = tentative event
+
+# ── Capacity ─────────────────────────────────────────────────────────────────
+# Max simultaneous calls (each uses GPU; Ollama serializes generation). Over-cap
+# callers hear BUSY_MESSAGE and are hung up. Tune to your GPU headroom (2-3 typical).
+MAX_CONCURRENT_CALLS=2
+# BUSY_MESSAGE=Thank you for calling Advanced Vision Care. All of our lines are busy right now. Please call back in a few minutes. Goodbye.
+
+# ── Models (defaults are fine) ───────────────────────────────────────────────
+OLLAMA_MODEL=llama3.1:8b
+OLLAMA_URL=http://127.0.0.1:11434/v1
+# LLM provider: ollama (local, default) | anthropic (Claude API). Flip to A/B test Claude.
+LLM_PROVIDER=ollama
+ANTHROPIC_API_KEY=
+# Default is the most capable model; for low-latency phone voice prefer claude-haiku-4-5
+# (fastest) or claude-sonnet-4-6 (balance).
+ANTHROPIC_MODEL=claude-opus-4-8
+# ── STT: Deepgram (real-time, in-call only) ──────────────────────────────────
+# Nova-2 delivers end-of-utterance in <300ms (vs Whisper's 1-3s buffering). Key from
+# console.deepgram.com. Model is fixed to nova-2 in code; DEEPGRAM_MODEL is informational.
+DEEPGRAM_API_KEY=
+DEEPGRAM_MODEL=nova-2
+# Whisper is retained for POST-CALL transcription only (Phase 3), not the live pipeline.
+WHISPER_MODEL=base
+WHISPER_DEVICE=cuda
+WHISPER_COMPUTE=float16
+KOKORO_VOICE=af_heart
+KOKORO_MODEL_DIR=/home/tocmo0nlord/pipecat-run/models
diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..ff04b07
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,19 @@
+# Secrets — never commit
+.env
+
+# Recordings (local only, may contain PHI)
+recordings/
+
+# Model weights
+*.gguf
+
+# Python
+__pycache__/
+*.py[cod]
+*.egg-info/
+.venv/
+venv/
+
+# OS / editor
+.DS_Store
+*.swp
diff --git a/README.md b/README.md
new file mode 100644
index 0000000..1a4de53
--- /dev/null
+++ b/README.md
@@ -0,0 +1,132 @@
+# AVC Phone Agent — inbound optometry line (Pipecat + Twilio, fully local)
+
+A real phone number that callers dial; the agent answers in voice, handles hours /
+location / insurance / services questions, and **captures appointment requests** for
+staff callback. All AI runs **locally on this box**:
+
+```
+caller ─▶ Twilio ─▶ wss (Traefik TLS) ─▶ server.py ─▶ Pipecat pipeline:
+          Twilio Media Stream (8kHz µ-law)
+              │
+              ▼
+   Silero VAD ─▶ Whisper STT (GPU) ─▶ activeblue-avc (Ollama) ─▶ Kokoro TTS ─▶ back to caller
+```
+
+Inbound only. No cloud STT/TTS — audio stays on the machine except the Twilio carrier leg.
+
+## Files
+| File | Role |
+|---|---|
+| `server.py` | FastAPI: `POST /voice` (TwiML) + `WS /ws` (Twilio Media Stream) |
+| `bot.py` | The per-call Pipecat pipeline (VAD→STT→LLM→TTS) + tool wiring |
+| `practice.py` | **AVC business facts (PLACEHOLDERS — edit before go-live)** + appointment-capture tool |
+| `odoo_client.py` | Writes captured requests into Odoo (CRM lead by default) via XML-RPC |
+| `run.sh` | Launcher (reuses pipecat-run venv + sets CUDA lib path) |
+| `avc-phone.service` | systemd unit (install on this box) |
+| `deploy/setup-tls.sh` | One-shot: Let's Encrypt cert + nginx vhost install (run as root) |
+| `deploy/nginx-*.conf` | nginx TLS reverse-proxy vhost + WebSocket-upgrade map |
+| `traefik-avc-phone.yml` | Unused alternative (kept for a future multi-host/Traefik setup) |
+| `.env.example` | Copy to `.env`, fill Twilio creds + public host + Odoo creds |
+| `appointment_requests.jsonl` | Local fallback — only used if Odoo is unreachable/disabled |
+
+## What's done vs. what YOU must supply
+
+**Working / verified locally:**
+- Pipeline assembles; all services construct (smoke-tested).
+- GPU Whisper fixed — installed CUDA12 `cublas`+`cudnn` wheels into the venv; `run.sh`
+  sets `LD_LIBRARY_PATH` so faster-whisper finds them. Verified transcribe on GPU.
+- Local model `activeblue-avc:latest` is the brain; Kokoro voice; appointment tool.
+- **Odoo appointment integration wired + verified** against prod `db1`: a captured
+  request creates a `crm.lead` (callback to-do) via XML-RPC using the same API key the
+  `activeblue-agent` service uses. Verified create→read→delete (no residue left in db1).
+  If Odoo is unreachable or creds are blank, it falls back to `appointment_requests.jsonl`
+  and still confirms to the caller — a request is never lost.
+
+**You must supply (can't be done from this box):**
+1. **Twilio account + a Voice phone number.**
+2. **Port-forward 443** (and 80) from your router to this box, and run `deploy/setup-tls.sh`
+   for the nginx TLS reverse proxy (Twilio needs real TLS on 443 for the `wss` stream).
+3. **Real AVC facts** in `practice.py` (hours, address, insurance, services, phone).
+4. **Odoo creds in `.env`** (`ODOO_USER` + `ODOO_API_KEY`) to enable lead creation.
+   Set `ODOO_DB` (`db1` for prod) and `ODOO_TARGET` (`crm` lead, or `calendar` event).
+   Leave creds blank to disable Odoo and log to JSONL only.
+
+## Setup
+
+1. **Config**
+   ```bash
+   cd /home/tocmo0nlord/avc-phone
+   cp .env.example .env        # fill PUBLIC_HOST, TWILIO_ACCOUNT_SID, TWILIO_AUTH_TOKEN
+   $EDITOR practice.py         # replace PLACEHOLDER hours/address/insurance/services
+   ```
+
+2. **Run it**
+   ```bash
+   ./run.sh                    # listens plain HTTP on :8200 (Traefik terminates TLS)
+   curl localhost:8200/health  # {"status":"ok",...}
+   ```
+
+3. **TLS reverse proxy (nginx, on this box).** No Traefik — `voip.activeblue.net` points
+   at your WAN IP (`66.23.239.222`) which NATs to this box (`10.10.1.221`). nginx is
+   already installed and only serving the default site, so we add a vhost for the domain.
+   **Twilio's `wss` media stream needs real TLS on 443**, so:
+   - **Forward `443` (and `80`) on your router → `10.10.1.221`.** (80 is for the
+     Let's Encrypt challenge + the http→https redirect; 443 is the actual traffic.)
+   - Run the one-shot setup (gets a Let's Encrypt cert, installs the vhost + ws map,
+     reloads nginx):
+     ```bash
+     sudo bash deploy/setup-tls.sh
+     ```
+     It uses `deploy/nginx-voip.activeblue.net.conf` (proxies 443 → `127.0.0.1:8200`,
+     forwards the `/ws` upgrade, 1-hour stream timeout) and `deploy/nginx-ws-upgrade.conf`.
+   - Verify publicly: `curl https://voip.activeblue.net/health`.
+
+4. **Twilio number config** (console.twilio.com → your number → Voice):
+   - **A call comes in** → Webhook → `https://voip.activeblue.net/voice` → HTTP **POST**.
+   - Save. That's it — the TwiML we return tells Twilio to open the Media Stream to
+     `wss://voip.activeblue.net/ws`.
+
+5. **Call the number.** You should hear the greeting and be able to talk to it.
+
+## Security (built in)
+- **Webhook signature validation:** `POST /voice` verifies Twilio's `X-Twilio-Signature`
+  (HMAC-SHA1 over the public URL + sorted POST params, keyed by `TWILIO_AUTH_TOKEN`).
+  Enforced automatically whenever `TWILIO_AUTH_TOKEN` is set. Verified against Twilio's
+  published reference vector. Unsigned/forged requests get `403`. Set `TWILIO_VALIDATE=false`
+  only for local testing.
+  - The signed URL must match exactly, so **`PUBLIC_HOST` must equal the host on the number's
+    webhook** (`https://$PUBLIC_HOST/voice`). If Traefik rewrites host/path, signatures fail.
+- **Media-stream gate:** `/ws` can't carry a usable Twilio signature, so it's gated by a
+  shared `STREAM_TOKEN` embedded in the wss URL we hand Twilio. Bad/missing token → socket
+  closed. Set a stable `STREAM_TOKEN` in `.env` (`openssl rand -base64 24`).
+
+## Run it as a service (systemd)
+A unit is provided: `avc-phone.service` (runs as your user, `Restart=always`, ordered
+after `ollama.service`). Install (needs sudo — paste these in a `!` shell or a terminal):
+```bash
+sudo cp /home/tocmo0nlord/avc-phone/avc-phone.service /etc/systemd/system/
+sudo systemctl daemon-reload
+sudo systemctl enable --now avc-phone.service
+systemctl status avc-phone.service          # check it's running
+journalctl -u avc-phone.service -f          # follow logs
+```
+Restart after editing `.env` or `practice.py`: `sudo systemctl restart avc-phone.service`.
+(No-sudo alternative: a `systemctl --user` unit + `loginctl enable-linger tocmo0nlord` —
+ask and I'll convert it.)
+
+## Concurrency cap (built in)
+`MAX_CONCURRENT_CALLS` (default **2**) bounds simultaneous live calls. The count tracks
+active `/ws` pipelines (the real GPU consumers); when full, `/voice` speaks `BUSY_MESSAGE`
+and hangs up **before any GPU work**, so in-progress calls are never degraded. A hard
+reservation at `/ws` covers the rare race. `/health` reports `active_calls`/`max_calls`
+for monitoring. Tune the cap to your GPU headroom.
+
+## Known limits / next steps
+- **Per-call Whisper load:** each call currently constructs its own Whisper model on the
+  GPU. Fine within the cap; a future optimization is sharing one warm Whisper instance
+  across calls to cut memory + first-utterance latency.
+- **Latency:** first call after start pays one-time model loads (Whisper/Kokoro/Ollama).
+  Keep the process warm. Tune `WHISPER_MODEL=tiny` if you need faster STT.
+- **Function-calling reliability:** `activeblue-avc` is an 8B fine-tune; tool-calling
+  may need prompt tuning. If it's flaky, we can fall back to a deterministic slot-filling
+  flow for appointment capture.
diff --git a/avc-phone.service b/avc-phone.service
new file mode 100644
index 0000000..ada446d
--- /dev/null
+++ b/avc-phone.service
@@ -0,0 +1,34 @@
+[Unit]
+Description=AVC optometry phone agent (Pipecat + Twilio, local GPU)
+Documentation=file:///home/tocmo0nlord/avc-phone/README.md
+# Needs the network up and Ollama serving the activeblue-avc model. Docker (Odoo) is
+# only a soft dependency — the agent falls back to JSONL if Odoo is down — so it is
+# ordered after but not required.
+Wants=network-online.target
+After=network-online.target ollama.service docker.service
+# Restart rate limit (these are [Unit]-level directives).
+StartLimitIntervalSec=300
+StartLimitBurst=5
+
+[Service]
+Type=simple
+User=tocmo0nlord
+Group=tocmo0nlord
+WorkingDirectory=/home/tocmo0nlord/avc-phone
+# run.sh loads .env, sets LD_LIBRARY_PATH for the CUDA libs, and execs the server.
+ExecStart=/home/tocmo0nlord/avc-phone/run.sh
+# A phone line should always come back up.
+Restart=always
+RestartSec=5
+# Give model loads time before a failed start counts against the restart limit.
+TimeoutStartSec=120
+
+# Light hardening. Home is left writable on purpose: the app reads the shared venv +
+# Kokoro models + HF Whisper cache under /home and writes the JSONL fallback, so a
+# read-only home would risk breaking model-cache locks. /usr,/etc stay read-only.
+NoNewPrivileges=true
+ProtectSystem=full
+PrivateTmp=true
+
+[Install]
+WantedBy=multi-user.target
diff --git a/bot.py b/bot.py
new file mode 100644
index 0000000..cc2a717
--- /dev/null
+++ b/bot.py
@@ -0,0 +1,382 @@
+#!/usr/bin/env python3
+"""AVC optometry phone agent — the Pipecat pipeline for a single inbound call.
+
+Same VAD -> STT -> LLM -> TTS loop as pipecat-run/bot.py, but the ends are swapped
+for telephony: audio arrives/leaves as 8 kHz mu-law over a Twilio Media Stream
+(WebSocket), decoded by TwilioFrameSerializer. STT runs on the GPU; the LLM is the
+local `activeblue-avc` fine-tune via Ollama; TTS is local Kokoro.
+
+This module just builds + runs the pipeline for one connected call. server.py owns
+the FastAPI/TwiML/WebSocket side and calls run_call() once per call.
+"""
+
+import os
+
+import re
+import time
+
+from loguru import logger
+
+from pipecat.adapters.schemas.function_schema import FunctionSchema
+from pipecat.adapters.schemas.tools_schema import ToolsSchema
+from pipecat.audio.vad.silero import SileroVADAnalyzer
+from pipecat.audio.vad.vad_analyzer import VADParams
+from pipecat.frames.frames import (
+    BotStoppedSpeakingFrame,
+    EndFrame,
+    EndTaskFrame,
+    Frame,
+    InputAudioRawFrame,
+    LLMFullResponseEndFrame,
+    LLMTextFrame,
+    TTSSpeakFrame,
+)
+from pipecat.pipeline.pipeline import Pipeline
+from pipecat.pipeline.runner import PipelineRunner
+from pipecat.pipeline.task import PipelineParams, PipelineTask
+from pipecat.processors.aggregators.llm_context import LLMContext
+from pipecat.processors.aggregators.llm_response_universal import LLMContextAggregatorPair
+from pipecat.processors.audio.vad_processor import VADProcessor
+from pipecat.processors.frame_processor import FrameDirection, FrameProcessor
+from pipecat.serializers.twilio import TwilioFrameSerializer
+from pipecat.services.anthropic.llm import AnthropicLLMService
+from pipecat.services.deepgram.stt import DeepgramSTTService
+from pipecat.services.kokoro.tts import KokoroTTSService
+from pipecat.services.ollama.llm import OLLamaLLMService
+from pipecat.transports.websocket.fastapi import (
+    FastAPIWebsocketParams,
+    FastAPIWebsocketTransport,
+)
+
+from practice import practice_summary
+
+# ── Config (env-overridable) ─────────────────────────────────────────────────
+HERE = os.path.dirname(os.path.abspath(__file__))
+# Reuse the Kokoro model files already downloaded by the pipecat-run project.
+MODEL_DIR = os.environ.get("KOKORO_MODEL_DIR", "/home/tocmo0nlord/pipecat-run/models")
+
+OLLAMA_MODEL = os.environ.get("OLLAMA_MODEL", "activeblue-avc:latest")
+OLLAMA_URL = os.environ.get("OLLAMA_URL", "http://127.0.0.1:11434/v1")
+# Swappable LLM provider: "ollama" (local) or "anthropic" (Claude API). Same universal
+# LLMContext drives both — only the service construction differs (see build_llm_service).
+LLM_PROVIDER = os.environ.get("LLM_PROVIDER", "ollama").lower()
+ANTHROPIC_API_KEY = os.environ.get("ANTHROPIC_API_KEY", "")
+# Defaults to the most capable model. For low-latency PHONE voice, set ANTHROPIC_MODEL to
+# claude-haiku-4-5 (fastest) or claude-sonnet-4-6 (balance) — see notes in build_llm_service.
+ANTHROPIC_MODEL = os.environ.get("ANTHROPIC_MODEL", "claude-opus-4-8")
+# In-call function-calling: AUTO by provider — ON for Claude (reliable tool calls → real-time
+# Odoo booking), OFF for local Ollama (llama3.1:8b over-calls / leaks JSON). An explicit
+# ENABLE_TOOLS env overrides the auto choice either way.
+_enable_tools_env = os.environ.get("ENABLE_TOOLS")
+ENABLE_TOOLS = (
+    _enable_tools_env.lower() in ("1", "true", "yes")
+    if _enable_tools_env is not None
+    else (LLM_PROVIDER == "anthropic")
+)
+LLM_TEMPERATURE = float(os.environ.get("LLM_TEMPERATURE", "0.3"))
+LLM_MAX_TOKENS = int(os.environ.get("LLM_MAX_TOKENS", "160"))
+KOKORO_VOICE = os.environ.get("KOKORO_VOICE", "af_heart")
+# Real-time STT is Deepgram Nova-2: end-of-utterance events in <300ms (vs Whisper's
+# 1-3s of chunk buffering, the main cause of non-reply / repeat-yourself). Whisper
+# large-v3 is retained for post-call transcription only (Phase 3).
+DEEPGRAM_API_KEY = os.environ.get("DEEPGRAM_API_KEY", "")
+
+# Twilio sends 8 kHz mu-law on the wire — we run the PIPELINE at 16 kHz and let
+# TwilioFrameSerializer resample to/from the 8 kHz wire. (Silero VAD, Deepgram, and
+# Kokoro are all happy at 16 kHz.)
+WIRE_SAMPLE_RATE = 8000          # Twilio mu-law on the wire (serializer handles this)
+PIPELINE_SAMPLE_RATE = 16000     # internal rate Whisper/VAD actually need
+
+# VAD tuning. Defaults (confidence 0.7 / min_volume 0.6) are desktop-mic values that can
+# miss short/quiet 8 kHz telephony utterances like "yes" — loosen them for the phone.
+VAD_CONFIDENCE = float(os.environ.get("VAD_CONFIDENCE", "0.5"))
+VAD_MIN_VOLUME = float(os.environ.get("VAD_MIN_VOLUME", "0.3"))
+VAD_START_SECS = float(os.environ.get("VAD_START_SECS", "0.2"))
+VAD_STOP_SECS = float(os.environ.get("VAD_STOP_SECS", "0.5"))
+
+# Agent persona name — purely for warmth; change/remove freely.
+AGENT_NAME = os.environ.get("AGENT_NAME", "Sofia")
+
+SYSTEM_PROMPT = (
+    f"You are {AGENT_NAME}, a warm, friendly receptionist for Advanced Vision Care, an "
+    "optometry practice with eight offices in South Florida. You are on a real phone call, so "
+    "talk like a helpful human being: natural, relaxed, and genuinely conversational — usually "
+    "just one short sentence at a time. Speak in English. Say numbers, dates, and times as "
+    "words a person would say.\n\n"
+    "Your job is to answer callers' questions and to take appointment requests. To book a "
+    "visit you need four things: which office or city, the reason for the visit, a preferred "
+    "day and time, and their name. Gather these naturally as the conversation flows — don't "
+    "interrogate, and never ask for something the caller already told you (people often give "
+    "their name or reason in their first sentence). You already have their number from caller "
+    "ID, so never ask for a phone number. When you have the details, repeat them back in one "
+    "warm sentence to confirm, and let them know a staff member will call to finalize the time.\n\n"
+    "Stay truthful and within your limits:\n"
+    "- Use ONLY the facts below for addresses, phone numbers, insurance, and services. Never "
+    "make any of these up.\n"
+    "- To find the right office, ask what CITY or AREA is most convenient for the caller. Do "
+    "NOT suggest or name a specific office yourself — you don't know where they are. Only after "
+    "they tell you their area, name the matching office; and only list locations if they ask "
+    "what offices exist.\n"
+    "- You cannot see a calendar, so never say a time is open or available — take the time as "
+    "a request that staff will confirm.\n"
+    "- Insurance: only confirm a plan that is in the list below. For any plan that is not "
+    "listed (UnitedHealthcare, Aetna, Cigna, and so on), don't say yes or no — say our staff "
+    "will verify their coverage.\n"
+    "- Hours are not published — say they vary by office and staff will confirm; never give "
+    "specific hours.\n"
+    "- You don't give medical advice and can't transfer calls. If the caller mentions an eye "
+    "problem, just note it as the reason and say a staff member or doctor will follow up.\n"
+    "- If you're not sure you heard something, simply ask them to repeat it.\n"
+    "- When the caller is all set, give a brief, warm closing that ends with the word "
+    "'Goodbye' — that ends the call, so only say it when you truly mean to.\n\n"
+    "PRACTICE FACTS:\n" + practice_summary()
+)
+
+
+def _build_tools() -> ToolsSchema:
+    # Only the booking action is a tool. Practice facts already live in the system prompt,
+    # so no get_practice_info tool (avoids needless calls/latency). callback_number is NOT
+    # required — we have the caller-ID and inject it in the handler.
+    return ToolsSchema(
+        standard_tools=[
+            FunctionSchema(
+                name="record_appointment_request",
+                description=(
+                    "Record the caller's appointment request once you have their name and at "
+                    "least the office/city and reason. Call this when the caller wants to book "
+                    "a visit; staff will call back to confirm the exact time."
+                ),
+                properties={
+                    "patient_name": {"type": "string", "description": "Caller's full name"},
+                    "location": {"type": "string", "description": "Which office/city the caller wants, e.g. Hialeah, Kendall, Tamarac"},
+                    "reason": {"type": "string", "description": "Reason for the visit, e.g. annual exam, broken glasses, eye pain"},
+                    "preferred_time": {"type": "string", "description": "Preferred day/time in the caller's words, if given"},
+                },
+                required=["patient_name"],
+            ),
+        ]
+    )
+
+
+class EndCallProcessor(FrameProcessor):
+    """Lets Sofia hang up. MUST sit between the LLM and the TTS: there it sees her reply
+    text (LLMTextFrame, flowing downstream) AND the upstream copy of BotStoppedSpeakingFrame
+    the output transport emits. It accumulates each reply; if the finished reply contains a
+    closing ('goodbye'/'adiós'), it waits until she's done speaking, then pushes EndTaskFrame
+    upstream — the task ends and TwilioFrameSerializer (auto_hang_up) drops the call."""
+
+    _CLOSINGS = ("goodbye", "good-bye", "good bye", "adiós", "adios", "hasta luego")
+
+    def __init__(self):
+        super().__init__()
+        self._buf = ""
+        self._should_end = False
+
+    @classmethod
+    def _is_closing(cls, text: str) -> bool:
+        t = (text or "").lower()
+        return any(c in t for c in cls._CLOSINGS)
+
+    async def process_frame(self, frame: Frame, direction: FrameDirection):
+        await super().process_frame(frame, direction)
+        if isinstance(frame, LLMTextFrame):
+            self._buf += frame.text
+        elif isinstance(frame, LLMFullResponseEndFrame):
+            if self._is_closing(self._buf):
+                self._should_end = True
+                logger.info("Sofia signalled closing -- will hang up after she finishes speaking")
+            self._buf = ""
+        elif isinstance(frame, BotStoppedSpeakingFrame) and self._should_end:
+            self._should_end = False
+            logger.info("Sofia closed the call -- ending task / hanging up")
+            await self.push_frame(EndTaskFrame(), FrameDirection.UPSTREAM)
+        await self.push_frame(frame, direction)
+
+
+class AudioHeartbeat(FrameProcessor):
+    """Diagnostic: logs how many inbound audio frames arrive every ~5s. If this keeps
+    ticking but VAD never fires, the issue is VAD/threshold; if it drops to 0 after a
+    turn, inbound audio stalled at the transport. Cheap, leave it on while stabilizing."""
+
+    def __init__(self):
+        super().__init__()
+        self._n = 0
+        self._t = time.time()
+
+    async def process_frame(self, frame: Frame, direction: FrameDirection):
+        await super().process_frame(frame, direction)
+        if isinstance(frame, InputAudioRawFrame):
+            self._n += 1
+            now = time.time()
+            if now - self._t >= 5:
+                logger.info(f"[audio-in] {self._n} frames in last {now - self._t:.0f}s")
+                self._n = 0
+                self._t = now
+        await self.push_frame(frame, direction)
+
+
+def build_llm_service():
+    """Build the LLM service for the selected provider. The universal LLMContext +
+    aggregators work with either, so only this construction differs (true A/B swap)."""
+    if LLM_PROVIDER == "anthropic":
+        if not ANTHROPIC_API_KEY:
+            raise RuntimeError("LLM_PROVIDER=anthropic but ANTHROPIC_API_KEY is not set")
+        logger.info(f"LLM provider: anthropic ({ANTHROPIC_MODEL})")
+        # NOTE: Opus 4.8/4.7 reject temperature/top_p/top_k (HTTP 400), so we omit them —
+        # this keeps the default Opus model working. For low-latency phone voice, prefer
+        # claude-haiku-4-5 (fastest) or claude-sonnet-4-6 over Opus. enable_prompt_caching
+        # caches the system prompt + growing conversation prefix (helps multi-turn cost/latency).
+        return AnthropicLLMService(
+            api_key=ANTHROPIC_API_KEY,
+            settings=AnthropicLLMService.Settings(
+                model=ANTHROPIC_MODEL,
+                enable_prompt_caching=True,
+                max_tokens=LLM_MAX_TOKENS,
+            ),
+        )
+    logger.info(f"LLM provider: ollama ({OLLAMA_MODEL})")
+    return OLLamaLLMService(
+        settings=OLLamaLLMService.Settings(
+            model=OLLAMA_MODEL,
+            temperature=LLM_TEMPERATURE,
+            max_tokens=LLM_MAX_TOKENS,
+        ),
+        base_url=OLLAMA_URL,
+    )
+
+
+async def run_agent(transport, caller_number=None, call_sid=None, do_capture=True):
+    """Build + run the AVC voice agent on a given transport. Shared by the phone path
+    (Twilio Media Stream) and the browser path (WebRTC) — same prompt, model, voice, and
+    booking/hang-up logic; only the transport differs. do_capture writes the post-call
+    appointment to Odoo (on for phone; off for browser testing so it doesn't make cards)."""
+    stt = DeepgramSTTService(
+        api_key=DEEPGRAM_API_KEY,
+        settings=DeepgramSTTService.Settings(
+            model="nova-2",
+            language="en-US",
+            smart_format=True,
+            punctuate=True,
+            interim_results=False,      # final transcripts only — avoids double-firing
+            utterance_end_ms=1000,      # ms of silence before end-of-utterance fires
+        ),
+    )
+    llm = build_llm_service()
+    # In-call booking tool — only registered when ENABLE_TOOLS is on (auto: Claude yes,
+    # local Ollama no, since llama3.1:8b over-calls/leaks). The handler is a closure so it
+    # can stamp the verified caller-ID + call_sid onto the lead (the model never supplies a
+    # phone number — we don't ask for one). With tools on, this writes the Odoo lead IN-CALL,
+    # so the post-call extraction is skipped below to avoid a duplicate.
+    if ENABLE_TOOLS:
+        async def _record_appointment(params):
+            args = params.arguments or {}
+            if do_capture:
+                from practice import persist_appointment
+                persist_appointment({
+                    "call_sid": call_sid,
+                    "patient_name": args.get("patient_name"),
+                    "callback_number": caller_number,  # verified caller-ID, not model-supplied
+                    "location": args.get("location"),
+                    "reason": args.get("reason"),
+                    "preferred_time": args.get("preferred_time"),
+                    "source": "in_call_tool",
+                })
+            else:
+                logger.info(f"[capture off] would record appointment: {args.get('patient_name')} / {args.get('location')}")
+            await params.result_callback(
+                {"status": "recorded", "message": "Recorded — staff will call to confirm the time."}
+            )
+
+        llm.register_function("record_appointment_request", _record_appointment)
+
+    tts = KokoroTTSService(
+        model_path=os.path.join(MODEL_DIR, "kokoro-v1.0.onnx"),
+        voices_path=os.path.join(MODEL_DIR, "voices-v1.0.bin"),
+        settings=KokoroTTSService.Settings(voice=KOKORO_VOICE),
+    )
+    vad = VADProcessor(vad_analyzer=SileroVADAnalyzer(params=VADParams(
+        confidence=VAD_CONFIDENCE,
+        start_secs=VAD_START_SECS,
+        stop_secs=VAD_STOP_SECS,
+        min_volume=VAD_MIN_VOLUME,
+    )))
+    heartbeat = AudioHeartbeat()
+
+    context_kwargs = {"messages": [{"role": "system", "content": SYSTEM_PROMPT}]}
+    if ENABLE_TOOLS:
+        context_kwargs["tools"] = _build_tools()
+    context = LLMContext(**context_kwargs)
+    agg = LLMContextAggregatorPair(context)
+    endcall = EndCallProcessor()
+
+    pipeline = Pipeline(
+        [
+            transport.input(),
+            heartbeat,
+            vad,
+            stt,
+            agg.user(),
+            llm,
+            endcall,
+            tts,
+            transport.output(),
+            agg.assistant(),
+        ]
+    )
+
+    task = PipelineTask(
+        pipeline,
+        params=PipelineParams(
+            audio_in_sample_rate=PIPELINE_SAMPLE_RATE,
+            audio_out_sample_rate=PIPELINE_SAMPLE_RATE,
+            allow_interruptions=True,
+        ),
+    )
+
+    @transport.event_handler("on_client_connected")
+    async def on_client_connected(transport, client):
+        logger.info("Client connected -- greeting")
+        await task.queue_frames(
+            [TTSSpeakFrame(
+                f"Thank you for calling Advanced Vision Care, this is {AGENT_NAME}. "
+                "How can I help you today?"
+            )]
+        )
+
+    @transport.event_handler("on_client_disconnected")
+    async def on_client_disconnected(transport, client):
+        logger.info("Client disconnected -- ending task")
+        await task.queue_frame(EndFrame())
+
+    runner = PipelineRunner(handle_sigint=False)
+    await runner.run(task)
+
+    # Call is over. Post-call extraction is the capture path ONLY when in-call tools are
+    # off (local Ollama). With tools on (Claude), the booking was already written in-call,
+    # so skip extraction to avoid a duplicate lead.
+    if do_capture and not ENABLE_TOOLS:
+        try:
+            from extract import extract_and_record
+
+            await extract_and_record(
+                context.messages, OLLAMA_URL, OLLAMA_MODEL,
+                call_sid=call_sid, caller_number=caller_number,
+            )
+        except Exception:
+            logger.exception("Post-call appointment extraction failed")
+
+
+async def run_call(websocket, serializer: TwilioFrameSerializer, caller_number=None, call_sid=None):
+    """Phone entrypoint: wrap the Twilio Media Stream in a transport, run the shared agent."""
+    transport = FastAPIWebsocketTransport(
+        websocket=websocket,
+        params=FastAPIWebsocketParams(
+            audio_in_enabled=True,
+            audio_out_enabled=True,
+            audio_in_sample_rate=PIPELINE_SAMPLE_RATE,
+            audio_out_sample_rate=PIPELINE_SAMPLE_RATE,
+            add_wav_header=False,
+            serializer=serializer,
+        ),
+    )
+    await run_agent(transport, caller_number=caller_number, call_sid=call_sid, do_capture=True)
diff --git a/bot_web.py b/bot_web.py
new file mode 100644
index 0000000..e191da0
--- /dev/null
+++ b/bot_web.py
@@ -0,0 +1,63 @@
+#!/usr/bin/env python3
+"""Browser entrypoint for the SAME Sofia agent — for fast iteration without phoning in.
+
+Reuses bot.run_agent (identical prompt, model, voice, booking + hang-up logic) but over a
+browser WebRTC transport via Pipecat's dev runner. Serves a mic UI at http://localhost:7860.
+
+Caveat: this path is 16 kHz WebRTC, NOT 8 kHz telephony — great for testing conversation /
+prompt / voice / hang-up, but it does NOT reproduce phone-specific audio (µ-law, clipping,
+VAD-on-8kHz). Appointment capture to Odoo is OFF here (do_capture=False) so browser tests
+don't create CRM cards.
+
+Run:
+    ./run_web.sh         # then open http://localhost:7860, click Connect, allow mic, talk
+"""
+
+from pipecat.runner.utils import create_transport
+from pipecat.transports.base_transport import TransportParams
+
+from bot import run_agent
+
+
+import os
+
+# Browser tests don't write to Odoo by default (keeps test bookings out of the CRM). Set
+# WEB_ALLOW_CAPTURE=true for a one-off test that actually creates the lead.
+WEB_ALLOW_CAPTURE = os.environ.get("WEB_ALLOW_CAPTURE", "false").lower() in ("1", "true", "yes")
+
+
+async def bot(runner_args):
+    """Called by pipecat.runner.run.main for each browser connection."""
+    transport = await create_transport(
+        runner_args,
+        {"webrtc": lambda: TransportParams(audio_in_enabled=True, audio_out_enabled=True)},
+    )
+    await run_agent(transport, do_capture=WEB_ALLOW_CAPTURE)
+
+
+if __name__ == "__main__":
+    import os
+    import sys
+
+    import uvicorn
+
+    # Serve on the LAN over HTTPS so the mic works from another machine (browsers require
+    # a secure context off-localhost). Reuses the self-signed certs (valid for 10.10.1.221).
+    CERT = os.environ.get("WEB_CERT", "/home/tocmo0nlord/pipecat-run/certs/cert.pem")
+    KEY = os.environ.get("WEB_KEY", "/home/tocmo0nlord/pipecat-run/certs/key.pem")
+    if os.path.exists(CERT) and os.path.exists(KEY):
+        _orig_run = uvicorn.run
+
+        def _run_with_tls(app, *args, **kwargs):
+            kwargs.setdefault("ssl_certfile", CERT)
+            kwargs.setdefault("ssl_keyfile", KEY)
+            return _orig_run(app, *args, **kwargs)
+
+        uvicorn.run = _run_with_tls
+        if "--host" not in sys.argv:
+            sys.argv += ["--host", "0.0.0.0"]
+        print("Browser UI: https://10.10.1.221:7860  (accept the self-signed cert once)")
+
+    from pipecat.runner.run import main
+
+    main()
diff --git a/deploy/nginx-voip.activeblue.net.conf b/deploy/nginx-voip.activeblue.net.conf
new file mode 100644
index 0000000..e1d0376
--- /dev/null
+++ b/deploy/nginx-voip.activeblue.net.conf
@@ -0,0 +1,58 @@
+# nginx reverse proxy for the AVC phone agent.
+# Terminates TLS for voip.activeblue.net and proxies to the app on 127.0.0.1:8200,
+# forwarding the /ws WebSocket (Twilio Media Stream) with long timeouts.
+#
+# Install:
+#   sudo cp deploy/nginx-voip.activeblue.net.conf /etc/nginx/sites-available/voip.activeblue.net
+#   sudo cp deploy/nginx-ws-upgrade.conf           /etc/nginx/conf.d/ws-upgrade.conf
+#   sudo ln -s /etc/nginx/sites-available/voip.activeblue.net /etc/nginx/sites-enabled/
+#   sudo nginx -t && sudo systemctl reload nginx
+# (Get the cert FIRST — see README — or the 443 block fails to load.)
+
+# ── HTTP :80 — ACME challenge + redirect everything else to HTTPS ─────────────
+server {
+    listen 80;
+    listen [::]:80;
+    server_name voip.activeblue.net;
+
+    # Let's Encrypt HTTP-01 webroot challenge (served, not redirected).
+    location /.well-known/acme-challenge/ {
+        root /var/www/html;
+    }
+
+    location / {
+        return 301 https://$host$request_uri;
+    }
+}
+
+# ── HTTPS :443 — TLS termination + proxy to the app ──────────────────────────
+server {
+    listen 443 ssl;
+    listen [::]:443 ssl;
+    http2 on;
+    server_name voip.activeblue.net;
+
+    ssl_certificate     /etc/letsencrypt/live/voip.activeblue.net/fullchain.pem;
+    ssl_certificate_key /etc/letsencrypt/live/voip.activeblue.net/privkey.pem;
+    ssl_protocols       TLSv1.2 TLSv1.3;
+    ssl_ciphers         HIGH:!aNULL:!MD5;
+    ssl_prefer_server_ciphers on;
+
+    # Twilio Media Streams hold the WebSocket open for the whole call — allow it.
+    proxy_read_timeout  3600s;
+    proxy_send_timeout  3600s;
+
+    location / {
+        proxy_pass http://127.0.0.1:8200;
+        proxy_http_version 1.1;
+
+        # WebSocket upgrade (for /ws). $connection_upgrade comes from ws-upgrade.conf.
+        proxy_set_header Upgrade    $http_upgrade;
+        proxy_set_header Connection $connection_upgrade;
+
+        proxy_set_header Host              $host;
+        proxy_set_header X-Real-IP         $remote_addr;
+        proxy_set_header X-Forwarded-For   $proxy_add_x_forwarded_for;
+        proxy_set_header X-Forwarded-Proto $scheme;
+    }
+}
diff --git a/deploy/nginx-ws-upgrade.conf b/deploy/nginx-ws-upgrade.conf
new file mode 100644
index 0000000..eddf262
--- /dev/null
+++ b/deploy/nginx-ws-upgrade.conf
@@ -0,0 +1,8 @@
+# Maps the Connection header for WebSocket proxying. Goes in /etc/nginx/conf.d/ so it
+# lives in the http{} context (a `map` can't go inside a server/location block).
+# When a request carries `Upgrade: websocket`, send `Connection: upgrade`; otherwise
+# `Connection: close`. Used by the voip.activeblue.net vhost for the /ws media stream.
+map $http_upgrade $connection_upgrade {
+    default upgrade;
+    ''      close;
+}
diff --git a/deploy/setup-tls.sh b/deploy/setup-tls.sh
new file mode 100755
index 0000000..0c10260
--- /dev/null
+++ b/deploy/setup-tls.sh
@@ -0,0 +1,43 @@
+#!/usr/bin/env bash
+# One-shot TLS + nginx reverse-proxy setup for the AVC phone agent.
+# RUN AS ROOT:  sudo bash deploy/setup-tls.sh
+#
+# Prerequisites (must be true BEFORE running):
+#   - DNS: voip.activeblue.net -> your WAN IP        (done: 66.23.239.222)
+#   - Router forwards external 80 AND 443 -> this box (10.10.1.221)
+#   - nginx running with its default :80 site (used to answer the ACME challenge)
+#
+# What it does: installs certbot, gets a Let's Encrypt cert via the webroot challenge
+# (served by the existing default :80 site), installs the vhost + ws-upgrade map, then
+# tests and reloads nginx. Idempotent-ish; safe to re-run.
+set -euo pipefail
+
+DOMAIN="voip.activeblue.net"
+EMAIL="mr.garcia09@gmail.com"
+APP_DIR="/home/tocmo0nlord/avc-phone"
+WEBROOT="/var/www/html"
+
+if [ "$(id -u)" -ne 0 ]; then echo "Run as root (sudo)."; exit 1; fi
+
+echo "==> 1/4 install certbot"
+if ! command -v certbot >/dev/null 2>&1; then
+  apt-get update && apt-get install -y certbot
+fi
+
+echo "==> 2/4 obtain certificate for $DOMAIN (webroot challenge)"
+mkdir -p "$WEBROOT/.well-known/acme-challenge"
+certbot certonly --webroot -w "$WEBROOT" -d "$DOMAIN" \
+  --non-interactive --agree-tos -m "$EMAIL" --keep-until-expiring
+
+echo "==> 3/4 install nginx vhost + ws-upgrade map"
+cp "$APP_DIR/deploy/nginx-ws-upgrade.conf"            /etc/nginx/conf.d/ws-upgrade.conf
+cp "$APP_DIR/deploy/nginx-voip.activeblue.net.conf"   /etc/nginx/sites-available/voip.activeblue.net
+ln -sf /etc/nginx/sites-available/voip.activeblue.net /etc/nginx/sites-enabled/voip.activeblue.net
+
+echo "==> 4/4 test + reload nginx"
+nginx -t
+systemctl reload nginx
+
+echo
+echo "Done. Verify:  curl https://$DOMAIN/health"
+echo "Cert auto-renews via the certbot systemd timer; nginx reload on renew is handled by certbot's deploy hook."
diff --git a/extract.py b/extract.py
new file mode 100644
index 0000000..f33501d
--- /dev/null
+++ b/extract.py
@@ -0,0 +1,101 @@
+"""Post-call appointment extraction.
+
+Instead of unreliable in-call tool-calling (which made llama3.1:8b speak raw JSON),
+we let the agent gather appointment details conversationally, then run ONE structured
+extraction over the finished transcript and write it to Odoo. Reliable because it's a
+single JSON-mode completion, not mid-conversation tool emission.
+"""
+
+import json
+import re
+
+import httpx
+from loguru import logger
+
+from practice import persist_appointment
+
+_EXTRACT_INSTRUCTIONS = (
+    "You are reviewing a phone-call transcript between a caller and the receptionist "
+    "for an optometry practice. Extract any APPOINTMENT REQUEST the caller made.\n"
+    "Respond with ONLY a JSON object with these keys:\n"
+    '  "wants_appointment": boolean — true only if the caller asked to book/schedule a visit\n'
+    '  "patient_name": string or null\n'
+    '  "callback_number": string or null (digits the caller gave to be called back)\n'
+    '  "location": string or null (which office/city)\n'
+    '  "reason": string or null (e.g. eye exam, broken glasses)\n'
+    '  "preferred_time": string or null (day/time in the caller\'s words)\n'
+    "Use null for anything not clearly stated. Do not invent values."
+)
+
+
+async def extract_and_record(messages, ollama_url, model, call_sid=None, caller_number=None):
+    """Extract an appointment from the transcript and persist it. Returns the record
+    dict if one was saved, else None."""
+    # Build a plain transcript from the conversation (skip the system prompt).
+    turns = [
+        f"{m['role']}: {m['content']}"
+        for m in messages
+        if m.get("role") in ("user", "assistant") and isinstance(m.get("content"), str) and m["content"].strip()
+    ]
+    if not any(m.get("role") == "user" for m in messages):
+        return None  # nobody said anything
+    transcript = "\n".join(turns)
+
+    base = ollama_url.rstrip("/")
+    if base.endswith("/v1"):
+        base = base[:-3]
+
+    try:
+        async with httpx.AsyncClient(timeout=30) as client:
+            r = await client.post(
+                f"{base}/api/chat",
+                json={
+                    "model": model,
+                    "format": "json",
+                    "stream": False,
+                    "options": {"temperature": 0},
+                    "messages": [
+                        {"role": "system", "content": _EXTRACT_INSTRUCTIONS},
+                        {"role": "user", "content": f"Transcript:\n{transcript}"},
+                    ],
+                },
+            )
+            r.raise_for_status()
+            data = json.loads(r.json()["message"]["content"])
+    except Exception:
+        logger.exception("Appointment extraction call failed")
+        return None
+
+    if not data.get("wants_appointment"):
+        logger.info("Post-call extraction: no appointment requested")
+        return None
+
+    # Don't create near-empty cards from quick hang-ups: require at least a name or a
+    # reason. A bare location + caller-ID isn't enough to be worth a worklist card.
+    name = (data.get("patient_name") or "").strip()
+    reason_raw = (data.get("reason") or "").strip()
+    if not name and not reason_raw:
+        logger.info("Post-call extraction: appointment intent but no name/reason captured — skipping card")
+        return None
+
+    # Prefer the verified Twilio caller-ID over a number pulled from the transcript —
+    # the model sometimes invents/echoes a phone number. Keep a genuinely different
+    # spoken number as a note for staff.
+    spoken = (data.get("callback_number") or "").strip()
+    callback = caller_number or spoken or None
+    reason = data.get("reason")
+    if spoken and caller_number and re.sub(r"\D", "", spoken) != re.sub(r"\D", "", caller_number):
+        reason = f"{reason or ''} (caller mentioned alternate number: {spoken})".strip()
+
+    record = {
+        "call_sid": call_sid,
+        "patient_name": data.get("patient_name"),
+        "callback_number": callback,
+        "location": data.get("location"),
+        "reason": reason,
+        "preferred_time": data.get("preferred_time"),
+        "source": "post_call_extraction",
+    }
+    where = persist_appointment(record)
+    logger.info(f"Post-call appointment saved ({where}): {record['patient_name']} / {record['location']}")
+    return record
diff --git a/odoo_client.py b/odoo_client.py
new file mode 100644
index 0000000..768d288
--- /dev/null
+++ b/odoo_client.py
@@ -0,0 +1,106 @@
+"""Minimal Odoo XML-RPC client for the phone agent.
+
+Creates an appointment *request* in Odoo from a captured call. A request is NOT a
+confirmed booking — staff call the patient back to finalize — so by default we write
+a CRM lead (a clean "to-do" that doesn't occupy a real calendar slot). Set
+ODOO_TARGET=calendar to instead drop a tentative event on the calendar.
+
+Auth + target are all env-driven (see .env.example). Connection is lazy and every
+failure is swallowed by the caller's fallback, so a flaky Odoo never drops a request.
+"""
+
+import os
+import xmlrpc.client
+from datetime import datetime, timedelta
+from html import escape
+
+ODOO_URL = os.environ.get("ODOO_URL", "http://localhost:8069")
+ODOO_DB = os.environ.get("ODOO_DB", "db1")
+ODOO_USER = os.environ.get("ODOO_USER", "")
+ODOO_API_KEY = os.environ.get("ODOO_API_KEY", "")
+ODOO_TARGET = os.environ.get("ODOO_TARGET", "crm").lower()   # "crm" | "calendar"
+# Pipeline placement for crm target. If ODOO_STAGE_ID is set, the request is created as a
+# staged opportunity (shows up in the CRM pipeline as a worklist) instead of a bare lead.
+ODOO_STAGE_ID = int(os.environ["ODOO_STAGE_ID"]) if os.environ.get("ODOO_STAGE_ID") else None
+ODOO_TEAM_ID = int(os.environ["ODOO_TEAM_ID"]) if os.environ.get("ODOO_TEAM_ID") else None
+ODOO_USER_ID = int(os.environ["ODOO_USER_ID"]) if os.environ.get("ODOO_USER_ID") else None
+
+
+class OdooError(RuntimeError):
+    pass
+
+
+def _connect():
+    if not (ODOO_USER and ODOO_API_KEY):
+        raise OdooError("ODOO_USER / ODOO_API_KEY not set")
+    common = xmlrpc.client.ServerProxy(f"{ODOO_URL}/xmlrpc/2/common")
+    uid = common.authenticate(ODOO_DB, ODOO_USER, ODOO_API_KEY, {})
+    if not uid:
+        raise OdooError("Odoo authentication failed (check db/user/key)")
+    models = xmlrpc.client.ServerProxy(f"{ODOO_URL}/xmlrpc/2/object")
+    return uid, models
+
+
+def _exec(uid, models, model, method, *args, **kw):
+    return models.execute_kw(ODOO_DB, uid, ODOO_API_KEY, model, method, list(args), kw)
+
+
+def _find_or_create_partner(uid, models, name, phone):
+    """Return a res.partner id, matching on phone first, else creating one."""
+    domain = []
+    if phone:
+        domain = ["|", ["phone", "=", phone], ["mobile", "=", phone]]
+    if domain:
+        hit = _exec(uid, models, "res.partner", "search", domain, limit=1)
+        if hit:
+            return hit[0]
+    vals = {"name": name or "Phone caller", "phone": phone or False, "company_type": "person"}
+    return _exec(uid, models, "res.partner", "create", vals)
+
+
+def create_appointment_request(patient_name, callback_number, reason, preferred_time, call_sid=None):
+    """Create the request in Odoo. Returns (model, record_id) or raises OdooError."""
+    uid, models = _connect()
+    summary = f"📞 Phone appt request — {patient_name or 'caller'}"
+    # description is an Odoo HTML field — build with <br/> so it renders in the UI.
+    rows = [
+        ("Name", patient_name),
+        ("Callback", callback_number),
+        ("Reason", reason),
+        ("Preferred time (patient's words)", preferred_time),
+        ("Twilio call SID", call_sid),
+    ]
+    note = "<p><b>Captured by the AVC phone agent</b> (UNCONFIRMED — call patient to finalize).</p><p>" + \
+        "<br/>".join(f"{escape(k)}: {escape(str(v)) if v else '—'}" for k, v in rows) + "</p>"
+
+    if ODOO_TARGET == "calendar":
+        partner_id = _find_or_create_partner(uid, models, patient_name, callback_number)
+        # Tentative 30-min slot tomorrow 9:00 as a visible placeholder; real time set on callback.
+        start = (datetime.utcnow() + timedelta(days=1)).replace(hour=9, minute=0, second=0, microsecond=0)
+        vals = {
+            "name": summary,
+            "start": start.strftime("%Y-%m-%d %H:%M:%S"),
+            "stop": (start + timedelta(minutes=30)).strftime("%Y-%m-%d %H:%M:%S"),
+            "description": note,
+            "partner_ids": [(4, partner_id)],
+        }
+        rec = _exec(uid, models, "calendar.event", "create", vals)
+        return ("calendar.event", rec)
+
+    # CRM target. With a stage configured, create a staged opportunity (lands in the
+    # pipeline as a worklist staff act on); otherwise a plain lead.
+    vals = {
+        "name": summary,
+        "contact_name": patient_name or False,
+        "phone": callback_number or False,
+        "description": note,
+        "type": "opportunity" if ODOO_STAGE_ID else "lead",
+    }
+    if ODOO_STAGE_ID:
+        vals["stage_id"] = ODOO_STAGE_ID
+    if ODOO_TEAM_ID:
+        vals["team_id"] = ODOO_TEAM_ID
+    if ODOO_USER_ID:
+        vals["user_id"] = ODOO_USER_ID
+    rec = _exec(uid, models, "crm.lead", "create", vals)
+    return ("crm.lead", rec)
diff --git a/practice.py b/practice.py
new file mode 100644
index 0000000..3c6ec5d
--- /dev/null
+++ b/practice.py
@@ -0,0 +1,155 @@
+"""Advanced Vision Care practice facts + the phone agent's tools.
+
+Facts sourced from advancedvisioncareflorida.com (8 locations across Broward,
+Miami-Dade, Palm Beach). NOTE: the website does NOT publish office hours, so we do
+NOT assert hours — the agent must offer to have staff confirm them instead of
+inventing them. Fill HOURS in if/when you have them.
+"""
+
+import json
+import os
+import re
+from datetime import datetime, timezone
+
+from loguru import logger
+
+# ─────────────────────────────────────────────────────────────────────────────
+# Real facts from advancedvisioncareflorida.com
+# ─────────────────────────────────────────────────────────────────────────────
+LOCATIONS = [
+    # Broward County
+    {"city": "Hollywood / Fort Lauderdale", "address": "2873 Stirling Rd, Fort Lauderdale, FL 33312", "phone": "(954) 983-4969"},
+    {"city": "Tamarac", "address": "5865 N University Dr, Tamarac, FL 33321", "phone": "(954) 720-2720"},
+    {"city": "Pembroke Pines", "address": "246 S Flamingo Rd, Pembroke Pines, FL 33027", "phone": "(954) 443-1230"},
+    {"city": "Lauderdale Lakes", "address": "3682 W Oakland Park Blvd, Lauderdale Lakes, FL 33311", "phone": "(954) 730-8087"},
+    # Miami-Dade County
+    {"city": "Hialeah", "address": "1770 W 32nd Pl, Hialeah, FL 33012", "phone": "(305) 885-4477"},
+    {"city": "Kendall", "address": "11605 N Kendall Dr, Miami, FL 33176", "phone": "(305) 982-8927"},
+    {"city": "Miami Gardens", "address": "4771 NW 183rd St, Miami Gardens, FL 33055", "phone": "(305) 390-2467"},
+    # Palm Beach County
+    {"city": "Boca Raton", "address": "21673 State Road 7, Boca Raton, FL 33428", "phone": "(561) 470-2310"},
+]
+
+PRACTICE_FACTS = {
+    "name": "Advanced Vision Care",
+    "locations": LOCATIONS,
+    "insurance": [
+        "CarePlus", "Doctors Health", "Florida Blue Medicare", "Optum", "Spectera",
+        "Sunshine Health", "VSP", "WellCare",
+    ],
+    "services": (
+        "routine and medical eye exams, contact lens exams, pediatric eye exams, "
+        "and LASIK consultations"
+    ),
+    # Website does not publish hours — leave None so the agent won't invent them.
+    "hours": None,
+}
+
+REQUESTS_LOG = os.path.join(os.path.dirname(os.path.abspath(__file__)), "appointment_requests.jsonl")
+
+
+# Expand street abbreviations so the TTS speaks "North Kendall Drive", not "N … D-R".
+_ABBREV = {
+    "NW": "Northwest", "NE": "Northeast", "SW": "Southwest", "SE": "Southeast",
+    "N": "North", "S": "South", "E": "East", "W": "West",
+    "Dr": "Drive", "Rd": "Road", "Blvd": "Boulevard", "St": "Street",
+    "Ave": "Avenue", "Pl": "Place", "Ln": "Lane", "Ct": "Court", "Hwy": "Highway",
+    "FL": "Florida",
+}
+
+
+def _spoken_address(addr: str) -> str:
+    """Expand directional + street-type abbreviations for natural speech."""
+    return re.sub(
+        r"\b(" + "|".join(re.escape(k) for k in _ABBREV) + r")\b",
+        lambda m: _ABBREV[m.group(1)],
+        addr,
+    )
+
+
+def practice_summary() -> str:
+    """Compact facts block for the system prompt."""
+    f = PRACTICE_FACTS
+    loc_lines = "\n".join(f"  - {l['city']}: {_spoken_address(l['address'])} — {l['phone']}" for l in f["locations"])
+    hours = f["hours"] or (
+        "NOT published — do not state specific hours; offer to have the office confirm."
+    )
+    return (
+        f"Practice name: {f['name']}\n"
+        f"Locations ({len(f['locations'])} offices across South Florida):\n{loc_lines}\n"
+        f"Insurance accepted (these EXACT plans only): {', '.join(f['insurance'])}.\n"
+        f"Services: {f['services']}\n"
+        f"Hours: {hours}\n"
+    )
+
+
+def _find_location(name: str):
+    """Loose match a caller's city/location text to a known office."""
+    if not name:
+        return None
+    n = name.lower()
+    for l in LOCATIONS:
+        if n in l["city"].lower() or l["city"].lower() in n:
+            return l
+    return None
+
+
+# ─── Tools (used when ENABLE_TOOLS=true and the model supports tool-calling) ──
+
+def persist_appointment(record: dict) -> str:
+    """Write an appointment request to Odoo (a crm.lead) if configured, else to the
+    JSONL fallback so a request is never lost. Returns where it landed. Used by both
+    the post-call extraction and the (optional) in-call tool."""
+    record.setdefault("ts", datetime.now(timezone.utc).isoformat())
+    if os.environ.get("ODOO_USER") and os.environ.get("ODOO_API_KEY"):
+        try:
+            from odoo_client import create_appointment_request
+
+            model, rec_id = create_appointment_request(
+                patient_name=record.get("patient_name"),
+                callback_number=record.get("callback_number"),
+                reason=f"[{record.get('location') or 'location TBD'}] {record.get('reason') or ''}".strip(),
+                preferred_time=record.get("preferred_time"),
+                call_sid=record.get("call_sid"),
+            )
+            logger.info(f"Appointment -> Odoo {model} id={rec_id}: {record.get('patient_name')}")
+            return f"odoo:{model}:{rec_id}"
+        except Exception as e:
+            logger.warning(f"Odoo write failed ({e!r}); falling back to local log")
+            record["odoo_error"] = repr(e)
+
+    with open(REQUESTS_LOG, "a") as fh:
+        fh.write(json.dumps(record) + "\n")
+    logger.info(f"Appointment -> JSONL: {record.get('patient_name')}")
+    return "jsonl"
+
+
+async def record_appointment_request(params):
+    """In-call tool path (only used when ENABLE_TOOLS=true). Wraps persist_appointment."""
+    args = params.arguments or {}
+    persist_appointment({
+        "call_sid": getattr(params, "call_sid", None),
+        "patient_name": args.get("patient_name"),
+        "callback_number": args.get("callback_number"),
+        "location": args.get("location"),
+        "reason": args.get("reason"),
+        "preferred_time": args.get("preferred_time"),
+        "source": "in_call_tool",
+    })
+    await params.result_callback(
+        {"status": "captured", "message": "Got it — our staff will call you back to confirm the time."}
+    )
+
+
+async def get_practice_info(params):
+    """Return practice facts (optionally narrowed to one location) for accurate answers."""
+    args = params.arguments or {}
+    loc = _find_location(args.get("location", ""))
+    result = {
+        "name": PRACTICE_FACTS["name"],
+        "insurance": PRACTICE_FACTS["insurance"],
+        "services": PRACTICE_FACTS["services"],
+        "hours": "not published — offer to have the office confirm",
+    }
+    result["location"] = loc if loc else PRACTICE_FACTS["locations"]
+    await params.result_callback(result)
diff --git a/run.sh b/run.sh
new file mode 100755
index 0000000..9028f4c
--- /dev/null
+++ b/run.sh
@@ -0,0 +1,19 @@
+#!/usr/bin/env bash
+# Launch the AVC phone agent. Reuses the pipecat-run venv (has pipecat-ai 1.3.0,
+# faster-whisper, kokoro-onnx, fastapi) and the CUDA12 libs we installed into it.
+set -euo pipefail
+
+HERE="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+VENV="/home/tocmo0nlord/pipecat-run/.venv"
+
+# faster-whisper (ctranslate2) needs the CUDA12 cublas + cudnn shared libs at runtime.
+NV="$VENV/lib/python3.13/site-packages/nvidia"
+export LD_LIBRARY_PATH="$NV/cublas/lib:$NV/cudnn/lib:${LD_LIBRARY_PATH:-}"
+
+# Load .env if present (Twilio creds, PUBLIC_HOST, model overrides).
+if [ -f "$HERE/.env" ]; then
+  set -a; . "$HERE/.env"; set +a
+fi
+
+cd "$HERE"
+exec "$VENV/bin/python" server.py
diff --git a/run_web.sh b/run_web.sh
new file mode 100755
index 0000000..14ab8c9
--- /dev/null
+++ b/run_web.sh
@@ -0,0 +1,20 @@
+#!/usr/bin/env bash
+# Launch the browser (WebRTC) version of Sofia for fast iteration — same brain as the
+# phone agent, served at http://localhost:7860. Mirrors run.sh (venv + CUDA libs + .env).
+set -euo pipefail
+
+HERE="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+VENV="/home/tocmo0nlord/pipecat-run/.venv"
+
+# faster-whisper (ctranslate2) needs the CUDA12 cublas + cudnn shared libs at runtime.
+NV="$VENV/lib/python3.13/site-packages/nvidia"
+export LD_LIBRARY_PATH="$NV/cublas/lib:$NV/cudnn/lib:${LD_LIBRARY_PATH:-}"
+
+# Load .env (model, voice, VAD, etc.) — same config the phone agent uses.
+if [ -f "$HERE/.env" ]; then
+  set -a; . "$HERE/.env"; set +a
+fi
+
+cd "$HERE"
+echo "Sofia (browser) starting — open http://localhost:7860 once it's up."
+exec "$VENV/bin/python" bot_web.py
diff --git a/server.py b/server.py
new file mode 100644
index 0000000..db3c5a8
--- /dev/null
+++ b/server.py
@@ -0,0 +1,218 @@
+#!/usr/bin/env python3
+"""Twilio-facing web server for the AVC phone agent.
+
+Two endpoints, both reached by Twilio over your public Traefik domain:
+
+  POST /voice  -> returns TwiML telling Twilio to open a bidirectional Media Stream
+                  back to wss://<PUBLIC_HOST>/ws?token=<STREAM_TOKEN>
+  WS   /ws     -> Twilio's Media Stream. We check the stream token, read the opening
+                  'start' event for the SIDs, then hand the socket to the pipeline.
+
+Security:
+  - POST /voice is authenticated with Twilio's X-Twilio-Signature (HMAC-SHA1 over the
+    public URL + sorted POST params, keyed by the API Key Secret). Enforced whenever
+    TWILIO_API_KEY_SECRET is set; set TWILIO_VALIDATE=false to bypass for local testing.
+  - WS /ws can't carry an X-Twilio-Signature usefully, so we gate it with a shared
+    STREAM_TOKEN embedded in the wss URL we hand Twilio in the TwiML.
+
+Inbound only. Run behind Traefik (TLS terminated there); this app listens plain HTTP
+on $PORT. See README for the Twilio number + Traefik wiring.
+"""
+
+import asyncio
+import base64
+import hashlib
+import hmac
+import json
+import os
+import secrets
+
+from fastapi import FastAPI, Request, WebSocket
+from fastapi.responses import HTMLResponse
+from loguru import logger
+
+from bot import run_call
+from pipecat.serializers.twilio import TwilioFrameSerializer
+
+# Public hostname Twilio dials back into (your Traefik domain), e.g. phone.example.com
+PUBLIC_HOST = os.environ.get("PUBLIC_HOST", "CHANGE-ME.example.com")
+PORT = int(os.environ.get("PORT", "8200"))
+# Bind localhost by default: nginx terminates TLS and proxies in from 127.0.0.1, so the
+# app needn't be exposed on the LAN. Set BIND_HOST=0.0.0.0 only if a remote proxy needs it.
+BIND_HOST = os.environ.get("BIND_HOST", "127.0.0.1")
+
+# Twilio REST creds — let the serializer auto-hang-up the carrier leg on EndFrame,
+# and validate inbound webhook signatures.
+TWILIO_ACCOUNT_SID = os.environ.get("TWILIO_ACCOUNT_SID")
+# Standard API Key (scoped to this app, revocable independently) instead of the account
+# master Auth Token. The Secret is used both for HMAC webhook-signature validation and as
+# the serializer credential for auto-hang-up.
+TWILIO_API_KEY_SID = os.environ.get("TWILIO_API_KEY_SID")
+TWILIO_API_KEY_SECRET = os.environ.get("TWILIO_API_KEY_SECRET")
+# Signature validation is ON by default when the API key secret exists; explicit opt-out.
+TWILIO_VALIDATE = os.environ.get("TWILIO_VALIDATE", "true").lower() not in ("false", "0", "no")
+
+# Shared secret embedded in the Media Stream wss URL to gate /ws. Auto-generated if
+# unset (fine for a single process), but set it in .env for stability across restarts.
+STREAM_TOKEN = os.environ.get("STREAM_TOKEN") or secrets.token_urlsafe(24)
+
+# Max simultaneous live calls. Each call holds an Ollama context on the 16GB GPU and
+# Ollama serializes generation, so cap this to protect call quality.
+# Over-cap callers hear BUSY_MESSAGE and are hung up — existing calls are never degraded.
+MAX_CONCURRENT_CALLS = int(os.environ.get("MAX_CONCURRENT_CALLS", "2"))
+BUSY_MESSAGE = os.environ.get(
+    "BUSY_MESSAGE",
+    "Thank you for calling Advanced Vision Care. All of our lines are busy right now. "
+    "Please call back in a few minutes. Goodbye.",
+)
+
+app = FastAPI()
+
+# Live count of active /ws pipelines (the real GPU consumers), guarded by a lock.
+_active_calls = 0
+_active_lock = asyncio.Lock()
+
+
+async def _reserve_call_slot() -> bool:
+    """Atomically take a call slot. Returns False if at capacity."""
+    global _active_calls
+    async with _active_lock:
+        if _active_calls >= MAX_CONCURRENT_CALLS:
+            return False
+        _active_calls += 1
+        return True
+
+
+async def _release_call_slot():
+    global _active_calls
+    async with _active_lock:
+        _active_calls = max(0, _active_calls - 1)
+
+
+def _twilio_signature_ok(url: str, params: dict, header_sig: str) -> bool:
+    """Recompute Twilio's request signature and compare in constant time.
+
+    Algorithm (Twilio docs): take the full public URL, append each POST param as
+    key+value sorted by key, HMAC-SHA1 with the API Key Secret, base64-encode.
+    """
+    if not (TWILIO_API_KEY_SECRET and header_sig):
+        return False
+    payload = url + "".join(f"{k}{params[k]}" for k in sorted(params))
+    digest = hmac.new(TWILIO_API_KEY_SECRET.encode(), payload.encode("utf-8"), hashlib.sha1).digest()
+    expected = base64.b64encode(digest).decode()
+    return hmac.compare_digest(expected, header_sig)
+
+
+@app.get("/health")
+async def health():
+    return {
+        "status": "ok",
+        "public_host": PUBLIC_HOST,
+        "validate": TWILIO_VALIDATE and bool(TWILIO_API_KEY_SECRET),
+        "active_calls": _active_calls,
+        "max_calls": MAX_CONCURRENT_CALLS,
+    }
+
+
+@app.post("/voice")
+async def voice(request: Request):
+    """TwiML: connect the call to our Media Stream WebSocket (bidirectional)."""
+    form = dict(await request.form())
+    if TWILIO_VALIDATE and TWILIO_API_KEY_SECRET:
+        # Validate against the PUBLIC url Twilio actually signed, not the internal one.
+        public_url = f"https://{PUBLIC_HOST}/voice"
+        sig = request.headers.get("X-Twilio-Signature", "")
+        if not _twilio_signature_ok(public_url, form, sig):
+            logger.warning("Rejected /voice: bad or missing X-Twilio-Signature")
+            return HTMLResponse(status_code=403, content="forbidden")
+    elif not TWILIO_API_KEY_SECRET:
+        logger.warning("/voice signature validation DISABLED (no TWILIO_API_KEY_SECRET set)")
+
+    caller = form.get("From", "")  # caller-ID; passed through for appointment callback
+
+    # Capacity gate: if all slots are busy, speak the busy message and hang up here —
+    # before any GPU work — so in-progress calls are never degraded. (A reservation is
+    # taken at /ws, tied to the socket lifecycle; this is the live read of that count.)
+    if _active_calls >= MAX_CONCURRENT_CALLS:
+        logger.info(f"At capacity ({_active_calls}/{MAX_CONCURRENT_CALLS}) — returning busy")
+        busy = f"""<?xml version="1.0" encoding="UTF-8"?>
+<Response>
+  <Say>{BUSY_MESSAGE}</Say>
+  <Hangup/>
+</Response>"""
+        return HTMLResponse(content=busy, media_type="application/xml")
+
+    # NOTE: <Connect><Stream> is bidirectional (agent can speak back). <Start><Stream>
+    # would be one-way (listen only) — do not use that here.
+    # Token passed as a <Parameter> (Twilio does NOT preserve a query string on the
+    # wss URL); it arrives in the /ws 'start' message's customParameters.
+    twiml = f"""<?xml version="1.0" encoding="UTF-8"?>
+<Response>
+  <Connect>
+    <Stream url="wss://{PUBLIC_HOST}/ws">
+      <Parameter name="token" value="{STREAM_TOKEN}" />
+      <Parameter name="caller" value="{caller}" />
+    </Stream>
+  </Connect>
+</Response>"""
+    return HTMLResponse(content=twiml, media_type="application/xml")
+
+
+@app.websocket("/ws")
+async def media_stream(websocket: WebSocket):
+    # The stream token rides in the TwiML <Parameter>, which Twilio delivers inside the
+    # 'start' message's customParameters — so we must accept the socket to read it, then
+    # validate + capacity-gate before doing any real work.
+    await websocket.accept()
+
+    call_sid = None
+    reserved = False
+    try:
+        # Twilio sends a 'connected' frame, then a 'start' frame with SIDs + params.
+        msgs = websocket.iter_text()
+        await msgs.__anext__()                       # 'connected'
+        start = json.loads(await msgs.__anext__())   # 'start'
+        start_data = start["start"]
+
+        token = (start_data.get("customParameters") or {}).get("token")
+        if token != STREAM_TOKEN:
+            logger.warning("Rejected /ws: bad or missing stream token")
+            await websocket.close(code=1008)  # policy violation
+            return
+
+        # Capacity gate (hard safety net for the /voice→/ws race).
+        if not await _reserve_call_slot():
+            logger.warning(f"/ws over capacity ({_active_calls}/{MAX_CONCURRENT_CALLS}) — closing")
+            await websocket.close(code=1013)  # try again later
+            return
+        reserved = True
+
+        stream_sid = start_data["streamSid"]
+        call_sid = start_data["callSid"]
+        caller_number = (start_data.get("customParameters") or {}).get("caller") or None
+        logger.info(
+            f"Media stream start: call={call_sid} stream={stream_sid} caller={caller_number} "
+            f"({_active_calls}/{MAX_CONCURRENT_CALLS} slots in use)"
+        )
+
+        serializer = TwilioFrameSerializer(
+            stream_sid=stream_sid,
+            call_sid=call_sid,
+            account_sid=TWILIO_ACCOUNT_SID,
+            auth_token=TWILIO_API_KEY_SECRET,
+        )
+        await run_call(websocket, serializer, caller_number=caller_number, call_sid=call_sid)
+    except Exception:
+        logger.exception("Call pipeline error")
+    finally:
+        if reserved:
+            await _release_call_slot()
+        logger.info(f"Call ended: {call_sid} ({_active_calls}/{MAX_CONCURRENT_CALLS} slots in use)")
+
+
+if __name__ == "__main__":
+    import uvicorn
+
+    logger.info(f"AVC phone agent on {BIND_HOST}:{PORT} | public={PUBLIC_HOST} | "
+                f"sig_validation={'on' if (TWILIO_VALIDATE and TWILIO_API_KEY_SECRET) else 'OFF'}")
+    uvicorn.run(app, host=BIND_HOST, port=PORT)
diff --git a/traefik-avc-phone.yml b/traefik-avc-phone.yml
new file mode 100644
index 0000000..ca18c26
--- /dev/null
+++ b/traefik-avc-phone.yml
@@ -0,0 +1,71 @@
+# Traefik DYNAMIC configuration for the AVC phone agent.
+# File-provider snippet — Traefik does NOT run on the GPU box (miaai), so this routes
+# your existing Traefik host to the phone service over the network.
+#
+# ── Install ──────────────────────────────────────────────────────────────────
+# Drop this into the directory your Traefik watches as a file provider, e.g.
+#   traefik.yml (static):
+#     providers:
+#       file:
+#         directory: /etc/traefik/dynamic
+#         watch: true
+# then: cp traefik-avc-phone.yml /etc/traefik/dynamic/   (Traefik hot-reloads it)
+#
+# ── BEFORE IT WORKS, set these 4 things to match YOUR Traefik ────────────────
+#  1) HOST              -> the real domain (must equal PUBLIC_HOST in the app's .env,
+#                          because Twilio's signature is computed over https://<HOST>/voice)
+#  2) entryPoints       -> your HTTPS entrypoint name (commonly `websecure`; could be `https`)
+#  3) certResolver      -> your ACME resolver name (commonly `le`/`letsencrypt`/`myresolver`)
+#  4) the service url    -> reachable address of the GPU box from the Traefik host
+#                          (LAN 10.10.1.221:8200; swap for the NetBird overlay IP if that's
+#                           the path Traefik uses to reach miaai)
+#
+# WebSockets (the /ws Twilio Media Stream): Traefik forwards the Upgrade handshake
+# automatically — no special middleware needed. One router/service covers both the
+# /voice HTTPS POST and the /ws WSS upgrade because they share host + backend.
+
+http:
+  routers:
+    avc-phone:
+      rule: "Host(`phone.example.com`)"          # 1) <-- your domain
+      entryPoints:
+        - websecure                               # 2) <-- your HTTPS entrypoint
+      service: avc-phone
+      tls:
+        certResolver: le                          # 3) <-- your ACME resolver
+
+    # Optional: redirect plain HTTP -> HTTPS for this host. Omit if your Traefik does
+    # this globally already.
+    avc-phone-http:
+      rule: "Host(`phone.example.com`)"           # 1) <-- your domain
+      entryPoints:
+        - web                                     # your HTTP (:80) entrypoint name
+      middlewares:
+        - avc-phone-https-redirect
+      service: avc-phone                          # never reached (redirected first)
+
+  services:
+    avc-phone:
+      loadBalancer:
+        passHostHeader: true
+        servers:
+          - url: "http://10.10.1.221:8200"        # 4) <-- GPU box (miaai) : app port
+        # Twilio media streams are long-lived; keep generous timeouts on the
+        # entrypoint (static config) — see note at bottom.
+
+  middlewares:
+    avc-phone-https-redirect:
+      redirectScheme:
+        scheme: https
+        permanent: true
+
+# ── One static-config reminder (NOT part of this file) ───────────────────────
+# Twilio Media Streams hold the WebSocket open for the whole call. If your HTTPS
+# entrypoint has a short respondingTimeout/idleTimeout, long calls get cut. In the
+# STATIC traefik config, ensure the entrypoint allows long-lived streams, e.g.:
+#   entryPoints:
+#     websecure:
+#       address: ":443"
+#       transport:
+#         respondingTimeouts:
+#           idleTimeout: 3600s