diff --git a/stacks/mOrpheus/.env.example b/stacks/mOrpheus/.env.example
new file mode 100644
index 0000000..e41021e
--- /dev/null
+++ b/stacks/mOrpheus/.env.example
@@ -0,0 +1,23 @@
+# mOrpheus stack tunables (irv-ml1). Copy to .env on the host; edit values there.
+
+# GPU pin (UUID). Default = the 3090 (the stabler audio-zoo GPU; A6000 is comfy's + spiky).
+#   irv-ml1: 3090 = GPU-12c15b3c-0827-0bab-b0cd-31aaa37a8e72
+#            A6000 = GPU-9672f0d5-3caf-b15c-898a-cd9d653c654b
+MORPHEUS_GPU_UUID=GPU-12c15b3c-0827-0bab-b0cd-31aaa37a8e72
+
+# Weights (stage once; not in git). mOrpheus = LlamaForCausalLM 3.3B bf16 (6.6GB).
+MORPHEUS_MODEL_DIR=/home/lkraven/morpheus/models/mOrpheus
+SNAC_DIR=/home/lkraven/morpheus/models/snac_24khz
+
+# Engine. bf16 REQUIRED — FP8 destroys audio-token generation on this model (0 valid SNAC
+#   frames even at greedy). bf16 weights ~6.6GB; --enforce-eager (in compose) skips CUDA-
+#   graph capture so it fits the shared 3090. util 0.32 (~7.7GB of the 24GB card); raise if
+#   you free the 3090, lower only if you can (bf16 weights alone are 6.6GB => >=~0.30 floor).
+MORPHEUS_GPU_UTIL=0.32
+MORPHEUS_MAX_LEN=4096
+MORPHEUS_MAX_SEQS=8
+
+# TTS endpoint (host port). Public /tts + /docs live here.
+MORPHEUS_TTS_PORT=8299
+MORPHEUS_DEFAULT_VOICE=baddy
+MORPHEUS_VOICES=baddy
diff --git a/stacks/mOrpheus/README.md b/stacks/mOrpheus/README.md
new file mode 100644
index 0000000..153cf62
--- /dev/null
+++ b/stacks/mOrpheus/README.md
@@ -0,0 +1,54 @@
+# mOrpheus — uncensored Orpheus TTS (irv-ml1)
+
+Permanent serving stack for **`MrDragonFox/mOrpheus_3B-1Base_early_preview-v1-25000`** — an
+uncensored Orpheus TTS finetune (Llama-3.2-3B LLM → SNAC 24 kHz audio). Trained speaker
+**"baddy"**; supports **zero-shot voice cloning** from a reference clip.
+
+> **INTERNAL RESEARCH ONLY.** License is **CC-BY-NC-4.0** (non-commercial). Do **not** expose
+> this endpoint externally or use it in any commercial-facing product.
+
+## Shape
+
+Two containers (see `compose.yaml`):
+
+| service | where | role |
+|---|---|---|
+| `vllm-morpheus` | GPU (3090), FP8 | serves the mOrpheus LLM; emits Orpheus audio tokens |
+| `morpheus-tts` | CPU | SNAC-decodes tokens → 24 kHz WAV; the public `/tts` endpoint |
+
+**Real-time:** ~165 tok/s single-stream on the 3090 (FP8) ⇒ **RTF ≈ 0.50 (2× real-time)**,
+measured. A ~4 s clip generates in ~2 s. (Whole-clip decode in v1; chunked streaming for
+lower time-to-first-audio is a future enhancement.)
+
+## Endpoints (`http://10.100.79.3:8299`)
+
+- `POST /tts` → `audio/wav`. Body: `{"text": "...", "voice": "baddy", "temperature": 0.6,
+  "max_tokens": 1200, "repetition_penalty": 1.1}`.
+  - **Zero-shot clone:** add `"reference_audio_b64": "<base64 WAV>"` + `"reference_text":
+    "<its transcript>"`. Keep `repetition_penalty <= 1.1` for cloning (higher penalizes the
+    in-context reference audio tokens and breaks generation).
+- `GET /voices`, `GET /health`, `GET /docs` (OpenAPI UI).
+
+**Expressive tags** (baddy is trained for these): `<sigh> <gasp> <laugh> <chuckle> <pant>
+<groan> <moan>` etc. Use **real carrier sentences with sparse, sentence-boundary tags** —
+stacking many tags with little text sends this early checkpoint into a repeat-loop.
+
+## Deploy (irv-ml1, as `lkraven` — docker-group, no sudo)
+
+```bash
+# one-time: stage weights (from the audition dir or a fresh pull-hf-repo) + copy the stack
+mkdir -p /home/lkraven/morpheus/models
+mv /home/lkraven/orpheus-audition/models/mOrpheus     /home/lkraven/morpheus/models/
+mv /home/lkraven/orpheus-audition/models/snac_24khz   /home/lkraven/morpheus/models/
+# copy compose.yaml + tts/ to /home/lkraven/morpheus/, cp .env.example .env
+cd /home/lkraven/morpheus && docker compose build && docker compose up -d
+```
+
+## Gotchas
+
+- **Pin `vllm/vllm-openai:v0.23.0`** — `latest` ships Blackwell-only kernels (oink/aiter)
+  that crash on Ampere *import*. Do not bump to `latest` on this box.
+- **GPU = 3090, not the A6000** — the A6000 is comfy's and spikes to ~41 GB without warning
+  (OOM'd two launches). FP8's ~5 GB footprint coexists with the 3090 audio zoo.
+- **FP8** on Ampere is a VRAM save (upcast), no compute speedup — real-time comes from vLLM.
+- Canonical copy lives here; deployed copy is `/home/lkraven/morpheus/` on irv-ml1.
diff --git a/stacks/mOrpheus/compose.yaml b/stacks/mOrpheus/compose.yaml
new file mode 100644
index 0000000..b8206f1
--- /dev/null
+++ b/stacks/mOrpheus/compose.yaml
@@ -0,0 +1,91 @@
+# mOrpheus — uncensored Orpheus TTS, permanent stack on irv-ml1.
+#
+# Two services:
+#   vllm-morpheus (GPU, 3090, FP8) — serves the mOrpheus 3B LLM; emits Orpheus audio
+#     tokens. Real-time: ~165 tok/s single-stream => RTF ~0.50 (2x real-time), measured.
+#   morpheus-tts  (CPU) — SNAC-decodes those tokens to 24kHz WAV; the public /tts endpoint
+#     (baddy voice + optional zero-shot cloning). Kept off the GPU to avoid contention.
+#
+# PLACEMENT / GOTCHAS (irv-ml1, INTERNAL RESEARCH — CC-BY-NC, do NOT expose externally):
+#   - GPU = the 3090 (audio-zoo GPU, stabler); the A6000 is comfy's and spikes to ~41GB
+#     without warning (it OOM'd two launch attempts). Pin by UUID via MORPHEUS_GPU_UUID.
+#   - Image PINNED to vllm/vllm-openai:v0.23.0 — `latest` ships Blackwell-only kernels
+#     (oink/aiter) that crash on Ampere import. Do NOT bump to latest on this Ampere box.
+#   - bf16 REQUIRED, NOT FP8. FP8 (--quantization fp8) destroys audio-token generation on
+#     this model — 0 valid SNAC frames even at greedy (quant-breaks-TTS). bf16 weights are
+#     ~6.6GB; with --enforce-eager (no CUDA-graph capture) it fits the shared 3090 (~8GB
+#     free after nothing else vLLM). util 0.32 (~7.7GB of the 24GB 3090).
+#   All tunables in .env — edit that, not this file.
+
+name: morpheus
+
+services:
+  vllm-morpheus:
+    image: vllm/vllm-openai:v0.23.0
+    container_name: vllm-morpheus
+    restart: unless-stopped
+    runtime: nvidia
+    ipc: host
+    environment:
+      - NVIDIA_VISIBLE_DEVICES=${MORPHEUS_GPU_UUID:-GPU-12c15b3c-0827-0bab-b0cd-31aaa37a8e72}
+      - CUDA_DEVICE_ORDER=PCI_BUS_ID
+    volumes:
+      - ${MORPHEUS_MODEL_DIR:-/home/lkraven/morpheus/models/mOrpheus}:/model:ro
+    command:
+      - /model
+      - --served-model-name
+      - morpheus
+      - --dtype
+      - bfloat16
+      # NO --quantization: FP8 breaks audio-token generation on this model (see header).
+      - --enforce-eager
+      - --max-model-len
+      - "${MORPHEUS_MAX_LEN:-4096}"
+      - --gpu-memory-utilization
+      - "${MORPHEUS_GPU_UTIL:-0.32}"
+      - --max-num-seqs
+      - "${MORPHEUS_MAX_SEQS:-8}"
+    healthcheck:
+      test: ["CMD", "python3", "-c", "import urllib.request; urllib.request.urlopen('http://localhost:8000/health')"]
+      interval: 30s
+      timeout: 10s
+      retries: 3
+      start_period: 300s
+    networks: [morpheus-net]
+
+  morpheus-tts:
+    build: ./tts
+    image: morpheus-tts:local
+    container_name: morpheus-tts
+    restart: unless-stopped
+    depends_on:
+      vllm-morpheus:
+        condition: service_healthy
+    environment:
+      - VLLM_URL=http://vllm-morpheus:8000/v1/completions
+      - VLLM_MODEL=morpheus
+      - SNAC_DEVICE=cpu
+      - MORPHEUS_DEFAULT_VOICE=${MORPHEUS_DEFAULT_VOICE:-baddy}
+      - MORPHEUS_VOICES=${MORPHEUS_VOICES:-baddy}
+    volumes:
+      - ${MORPHEUS_MODEL_DIR:-/home/lkraven/morpheus/models/mOrpheus}:/model:ro
+      - ${SNAC_DIR:-/home/lkraven/morpheus/models/snac_24khz}:/snac:ro
+    ports:
+      - "${MORPHEUS_TTS_PORT:-8299}:8000"
+    healthcheck:
+      test: ["CMD", "python3", "-c", "import urllib.request; urllib.request.urlopen('http://localhost:8000/health')"]
+      interval: 30s
+      timeout: 10s
+      retries: 3
+      start_period: 60s
+    labels:
+      - homepage.group=AI Systems
+      - homepage.name=mOrpheus TTS
+      - homepage.icon=mdi-account-voice
+      - homepage.description=Uncensored Orpheus TTS (baddy + zero-shot clone), real-time, irv-ml1
+      - homepage.href=http://10.100.79.3:${MORPHEUS_TTS_PORT:-8299}/docs
+    networks: [morpheus-net]
+
+networks:
+  morpheus-net:
+    name: morpheus-net
diff --git a/stacks/mOrpheus/tts/Dockerfile b/stacks/mOrpheus/tts/Dockerfile
new file mode 100644
index 0000000..fc62f45
--- /dev/null
+++ b/stacks/mOrpheus/tts/Dockerfile
@@ -0,0 +1,14 @@
+# mOrpheus TTS wrapper — CPU-only (SNAC decode + FastAPI /tts). Calls the vLLM engine.
+FROM python:3.11-slim
+
+RUN apt-get update && apt-get install -y --no-install-recommends libsndfile1 && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /app
+COPY requirements.txt .
+# CPU torch (SNAC decode is small; keeps this container off the GPU / out of contention)
+RUN pip install --no-cache-dir torch --index-url https://download.pytorch.org/whl/cpu \
+    && pip install --no-cache-dir -r requirements.txt
+
+COPY app.py .
+EXPOSE 8000
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "8000"]
diff --git a/stacks/mOrpheus/tts/app.py b/stacks/mOrpheus/tts/app.py
new file mode 100644
index 0000000..8b62b2f
--- /dev/null
+++ b/stacks/mOrpheus/tts/app.py
@@ -0,0 +1,125 @@
+#!/usr/bin/env python3
+"""mOrpheus TTS wrapper — turns the vLLM engine's Orpheus audio tokens into 24kHz WAV.
+
+Architecture: this CPU service builds the Orpheus prompt (as raw token ids), calls the
+vLLM engine (which serves the mOrpheus LLM), recovers the generated audio-token ids by
+re-tokenizing the returned text (vLLM emits them as `<custom_token_N>` strings when
+skip_special_tokens=false), then SNAC-decodes them to audio.
+
+Token scheme (verified): audio-base 128266, 7-token SNAC frames (pos0->L1, pos1/4->L2,
+pos2/3/5/6->L3, offsets k*4096). Named-speaker prompt: [SOH] "voice: text" [EOT][SOA].
+Zero-shot cloning: [BOS][SOH] ref_text [EOT][SOA][SOS] <ref audio tokens> [EOS_sp] then
+[SOH] text [EOT][SOA]. Generation stops at end-of-speech (128258).
+"""
+import os, io, base64
+import numpy as np, torch, soundfile as sf, requests
+from scipy.signal import resample_poly
+from fastapi import FastAPI, HTTPException
+from fastapi.responses import Response
+from pydantic import BaseModel
+from transformers import AutoTokenizer
+from snac import SNAC
+
+MODEL_DIR = os.environ.get("MORPHEUS_MODEL_DIR", "/model")
+SNAC_DIR = os.environ.get("SNAC_DIR", "/snac")
+VLLM_URL = os.environ.get("VLLM_URL", "http://vllm-morpheus:8000/v1/completions")
+VLLM_MODEL = os.environ.get("VLLM_MODEL", "morpheus")
+SNAC_DEVICE = os.environ.get("SNAC_DEVICE", "cpu")
+DEFAULT_VOICE = os.environ.get("MORPHEUS_DEFAULT_VOICE", "baddy")
+VOICES = [v for v in os.environ.get("MORPHEUS_VOICES", "baddy").split(",") if v]
+AUDIO_MAX = 156937
+AUDIO_BASE, SOS, EOS_SP, SOH, SOA, EOT, BOS = 128266, 128257, 128258, 128259, 128260, 128009, 128000
+
+tok = AutoTokenizer.from_pretrained(MODEL_DIR)
+snac_model = SNAC.from_pretrained(SNAC_DIR).to(SNAC_DEVICE).eval()
+app = FastAPI(title="mOrpheus TTS", version="0.1.0")
+
+
+class TTSReq(BaseModel):
+    text: str
+    voice: str = DEFAULT_VOICE
+    temperature: float = 0.6
+    top_p: float = 0.95
+    max_tokens: int = 1200
+    repetition_penalty: float = 1.1  # keep <=1.1 for cloning (higher penalizes ref audio tokens)
+    reference_audio_b64: str | None = None  # optional zero-shot clone: base64 WAV
+    reference_text: str | None = None       # transcript of the reference
+
+
+def _encode_ref(wav_bytes: bytes, ref_text: str) -> list[int]:
+    wav, sr = sf.read(io.BytesIO(wav_bytes))
+    if wav.ndim > 1:
+        wav = wav.mean(1)
+    wav = wav.astype(np.float32)
+    if sr != 24000:
+        wav = resample_poly(wav, 24000, sr).astype(np.float32)
+    wt = torch.tensor(wav, device=SNAC_DEVICE).view(1, 1, -1)
+    with torch.inference_mode():
+        codes = snac_model.encode(wt)
+    L1, L2, L3 = [c.squeeze(0).tolist() for c in codes]
+    ids = []
+    for i in range(len(L1)):
+        ids += [L1[i] + AUDIO_BASE, L2[2 * i] + AUDIO_BASE + 4096, L3[4 * i] + AUDIO_BASE + 8192,
+                L3[4 * i + 1] + AUDIO_BASE + 12288, L2[2 * i + 1] + AUDIO_BASE + 16384,
+                L3[4 * i + 2] + AUDIO_BASE + 20480, L3[4 * i + 3] + AUDIO_BASE + 24576]
+    return [BOS, SOH] + tok(ref_text, add_special_tokens=False).input_ids + [EOT, SOA, SOS] + ids + [EOS_SP]
+
+
+def _build_prompt(req: TTSReq) -> list[int]:
+    if req.reference_audio_b64 and req.reference_text:
+        ref = _encode_ref(base64.b64decode(req.reference_audio_b64), req.reference_text)
+        return ref + [SOH] + tok(req.text, add_special_tokens=False).input_ids + [EOT, SOA]
+    return [SOH] + tok(f"{req.voice}: {req.text}").input_ids + [EOT, SOA]
+
+
+def _decode(ids: list[int]):
+    if SOS in ids:
+        ids = ids[len(ids) - 1 - ids[::-1].index(SOS) + 1:]
+    codes = [t - AUDIO_BASE for t in ids if AUDIO_BASE <= t <= AUDIO_MAX]
+    l1, l2, l3 = [], [], []
+    for i in range(len(codes) // 7):
+        f = codes[7 * i:7 * i + 7]
+        c = [f[0], f[1] - 4096, f[2] - 8192, f[3] - 12288, f[4] - 16384, f[5] - 20480, f[6] - 24576]
+        if any(x < 0 or x > 4095 for x in c):
+            continue
+        l1.append(c[0]); l2 += [c[1], c[4]]; l3 += [c[2], c[3], c[5], c[6]]
+    if not l1:
+        return None
+    ct = [torch.tensor(l1).unsqueeze(0).to(SNAC_DEVICE),
+          torch.tensor(l2).unsqueeze(0).to(SNAC_DEVICE),
+          torch.tensor(l3).unsqueeze(0).to(SNAC_DEVICE)]
+    with torch.inference_mode():
+        return snac_model.decode(ct).squeeze().cpu().numpy()
+
+
+@app.get("/health")
+def health():
+    return {"status": "ok", "voices": VOICES, "default": DEFAULT_VOICE, "engine": VLLM_URL}
+
+
+@app.get("/voices")
+def voices():
+    return {"voices": VOICES, "default": DEFAULT_VOICE}
+
+
+@app.post("/tts")
+def tts(req: TTSReq):
+    prompt_ids = _build_prompt(req)
+    payload = {"model": VLLM_MODEL, "prompt": prompt_ids, "max_tokens": req.max_tokens,
+               "temperature": req.temperature, "top_p": req.top_p, "skip_special_tokens": False,
+               "stop_token_ids": [EOS_SP], "repetition_penalty": req.repetition_penalty}
+    try:
+        r = requests.post(VLLM_URL, json=payload, timeout=180)
+    except requests.RequestException as e:
+        raise HTTPException(502, f"engine unreachable: {e}")
+    if r.status_code != 200:
+        raise HTTPException(502, f"engine {r.status_code}: {r.text[:200]}")
+    text = r.json()["choices"][0]["text"]
+    gen_ids = tok(text, add_special_tokens=False).input_ids
+    audio = _decode(gen_ids)
+    if audio is None:
+        raise HTTPException(500, "no audio tokens generated")
+    buf = io.BytesIO()
+    sf.write(buf, audio, 24000, format="WAV", subtype="PCM_16")
+    return Response(content=buf.getvalue(), media_type="audio/wav",
+                    headers={"X-Audio-Seconds": f"{len(audio)/24000:.2f}"})
diff --git a/stacks/mOrpheus/tts/requirements.txt b/stacks/mOrpheus/tts/requirements.txt
new file mode 100644
index 0000000..3e72f19
--- /dev/null
+++ b/stacks/mOrpheus/tts/requirements.txt
@@ -0,0 +1,8 @@
+snac
+transformers
+soundfile
+scipy
+numpy
+fastapi
+uvicorn[standard]
+requests