fix: remove stale COPY line for deleted run-multi-gateways.sh

- Remove patch_tts_tool.py (orphaned)
- Remove run-multi-gateways.sh (replaced by s6 native profile supervision)
- Remove start-hermes.sh (overengineered — the image does this natively)
- compose.yml: command: gateway run, drop HERMES_PROFILES env var
- Dockerfile: revert start-hermes.sh COPY addition

Per-profile gateways are now managed by s6-overlay natively:
profiles with desired_state=running in gateway_state.json are
auto-started by the s6 reconciler on boot.

ai/compose.yml

@@ -7,20 +7,13 @@ services:
         - default
     container_name: hermes
     restart: always
-    # Use the image default ENTRYPOINT ["/init", "/opt/hermes/docker/main-wrapper.sh"]
-    # for proper s6-overlay supervision. The CMD runs our multi-profile launcher
-    # which spawns per-profile gateways in background, then the default gateway
-    # in foreground (keeps the container alive).
-    command: ["/usr/local/bin/start-hermes.sh"]
+    # Gateway run enables the internal API server on port 8642
+    command: gateway run
     environment:
       - HERMES_UID=10000
       - HERMES_GID=10000
-      - OLLAMA_HOST=http://ollama:11434
+      - OLLAMA_HOST=http://ollama-cpu:11434
       - HERMES_DASHBOARD=1
-      # Multi-profile: comma-separated list of profiles to run as gateways.
-      # start-hermes.sh reads this and starts one gateway per profile.
-      # Add profiles here when they exist on disk (e.g. default,researcher,writer)
-      - HERMES_PROFILES=ashley,claire,finn,matt,paul
       - API_SERVER_ENABLED=true
       - API_SERVER_PORT=8642
       - API_SERVER_HOST=0.0.0.0
@@ -108,12 +101,12 @@ services:
 
       - "traefik.http.services.syncthing.loadbalancer.server.port=8384"
 
-  ollama:
+  ollama-cpu:
     build:
       context: ./ollama
       dockerfile: Dockerfile
     image: ollama/ollama:rocm-gfx906
-    container_name: ollama
+    container_name: ollama-cpu
     tty: true
     restart: always 
     ports:
@@ -124,22 +117,42 @@ services:
       - /mnt/HoardingCow_docker_data/Ollama/ollama:/root/.ollama
     environment:
       - OLLAMA_VULKAN=0
-      - HSA_OVERRIDE_GFX_VERSION=9.0.6
-      - HCC_AMDGPU_TARGET=gfx906
-      - HIP_VISIBLE_DEVICES=0,1
-      - ROCR_VISIBLE_DEVICES=0,1
-      - HSA_ENABLE_SDMA=0 
       - OLLAMA_HOST=0.0.0.0
-      - OLLAMA_DEBUG=1
-      - OLLAMA_FLASH_ATTENTION=1
-      - OLLAMA_NUM_PARALLEL=2
+
+  llama-cpp-hermes:
+    image: llama-cpp:rocm-gfx906
+    container_name: llama-cpp-hermes
+    restart: unless-stopped
+    networks:
+      - ai_backend
+    ports:
+      - "127.0.0.1:8300:8080"
+    ipc: host
     devices:
-      # Map the render nodes and KFD for ROCm to work inside the container
       - /dev/kfd:/dev/kfd
       - /dev/dri:/dev/dri
     group_add:
       - "303"
       - "26"
+    environment:
+      - HSA_OVERRIDE_GFX_VERSION=9.0.6
+      - HSA_ENABLE_SDMA=0
+      - HIP_VISIBLE_DEVICES=0,1
+      - LLAMA_CACHE=/models
+    volumes:
+      - /mnt/HoardingCow_docker_data/Llama_cpp/models:/models
+      - /mnt/HoardingCow_docker_data/Ollama/ollama/models/blobs/sha256-17823599694fa3503ef54bf748d5078c6ce881f4d01616cafa255dc05d215a08:/model.gguf:ro
+    command: >
+      -m /model.gguf
+      --host 0.0.0.0
+      --port 8080
+      --gpu-layers 99
+      --ctx-size 163840
+      -ctk f16 -ctv f16
+      --flash-attn on
+      --split-mode layer
+      --no-mmap
+      --n-predict -1
 
   # --- Honcho + OpenConcho combiné: API + Web UI nginx/FastAPI ---
   honcho:
@@ -228,48 +241,6 @@ volumes:
     driver: bridge
     name: honcho_data
     
-  # llama_cpp_devstral:
-  #   image: ghcr.io/ggml-org/llama.cpp:server-rocm
-  #   container_name: llama_cpp_devstral
-  #   restart: unless-stopped
-  #   networks:
-  #     - ai_backend
-  #   ports:
-  #     - "8300:8080"
-  #   ipc: host
-  #   devices:
-  #     - "/dev/kfd:/dev/kfd"
-  #     - "/dev/dri:/dev/dri"
-  #   group_add:
-  #     - "303" # video
-  #     - "26"  # render
-  #   environment:
-  #     HSA_OVERRIDE_GFX_VERSION: 9.0.6
-  #     HIP_VISIBLE_DEVICES: 0,1
-  #     LLAMA_CACHE: /models
-  #   volumes:
-  #     - /mnt/HoardingCow_docker_data/Llama_cpp/models:/models
-  #     - /mnt/HoardingCow_docker_data/Llama_cpp/devstral-agent.jinja:/template.jinja
-  #   command: >
-  #     -hf unsloth/Devstral-Small-2-24B-Instruct-2512-GGUF:Devstral-Small-2-24B-Instruct-2512-Q8_0.gguf
-  #     -a devstral-2-small-llama_cpp
-  #     --chat-template-file /template.jinja
-  #     --host 0.0.0.0
-  #     --port 8080
-  #     --n-gpu-layers 99
-  #     --ctx-size 163840
-  #     --batch-size 4096
-  #     --ubatch-size 4096
-  #     --cache-type-k f16
-  #     --cache-type-v f16
-  #     --cache-reuse 256
-  #     --flash-attn on
-  #     --context-shift
-  #     --split-mode layer
-  #     --no-mmap
-  #     --n-predict -1
-  #     --parallel 2
-
   # vllm:
   #   image: nalanzeyu/vllm-gfx906:v0.9.0-rocm6.3
   #   container_name: vllm

ai/hermes/Dockerfile

@@ -13,6 +13,7 @@ FROM nousresearch/hermes-agent:latest
 USER root
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
+        gosu \
         libportaudio2 ca-certificates poppler-utils imagemagick \
         libolm-dev \
         texlive-latex-base texlive-latex-extra texlive-fonts-recommended \
@@ -58,15 +59,6 @@ os.remove(tgz)
 print('himalaya v1.2.0 installed')
 PYEOF
 
-# ---------- Install multi-gateway launcher ----------
-# Launches one gateway process per profile (HERMES_PROFILES env var)
-COPY --chmod=0755 run-multi-gateways.sh /usr/local/bin/run-multi-gateways.sh
-
-# ---------- Install s6-overlay compatible startup script ----------
-# Runs as the CMD via s6-overlay's main-program model.
-# Replaces the old bash->tini->entrypoint.sh chain that caused SIGTERM crash loops.
-COPY --chmod=0755 start-hermes.sh /usr/local/bin/start-hermes.sh
-
 # ---------- Runtime ----------
 USER hermes
 ENV HERMES_HOME=/opt/data

ai/hermes/patch_tts_tool.py

@@ -1,181 +0,0 @@
-#!/usr/bin/env python3
-"""Patch Hermes TTS tool: add Piper TTS provider, remove Edge TTS as default.
-
-Patches ALL copies of tts_tool.py found (venv site-packages + /opt/hermes/tools/).
-
-Searches multiple paths for tts_tool.py so it works both at build time
-(in the image venv) and at runtime (on the mounted data volume).
-
-Idempotent: if already patched, does nothing.
-"""
-
-import sys
-import os
-
-# ---------------------------------------------------------------------------
-# Search for all copies of tts_tool.py
-# ---------------------------------------------------------------------------
-CANDIDATE_PATHS = [
-    "/opt/hermes/.venv/lib/python3.13/site-packages/tools/tts_tool.py",
-    "/opt/hermes/tools/tts_tool.py",
-]
-
-found_paths = []
-
-for p in CANDIDATE_PATHS:
-    if os.path.exists(p):
-        found_paths.append(p)
-        print(f"Found tts_tool.py at: {p}")
-
-# Also try to find via Python import
-import subprocess
-try:
-    result = subprocess.run(
-        [sys.executable, "-c", "import tools.tts_tool; print(tools.tts_tool.__file__)"],
-        capture_output=True, text=True, timeout=5
-    )
-    if result.returncode == 0:
-        p = result.stdout.strip()
-        if os.path.exists(p) and p not in found_paths:
-            found_paths.append(p)
-            print(f"Found tts_tool.py via import at: {p}")
-except Exception:
-    pass
-
-if not found_paths:
-    print("WARNING: tts_tool.py not found anywhere. Patching deferred to runtime.")
-    print(f"Searched: {CANDIDATE_PATHS}")
-    sys.exit(0)
-
-# ---------------------------------------------------------------------------
-# Old else block: the Edge TTS default fallback to replace
-# ---------------------------------------------------------------------------
-old_else = '''        else:
-            # Default: Edge TTS (free), with NeuTTS as local fallback
-            edge_available = True
-            try:
-                _import_edge_tts()
-            except ImportError:
-                edge_available = False
-
-            if edge_available:
-                logger.info("Generating speech with Edge TTS...")
-                try:
-                    import concurrent.futures
-                    with concurrent.futures.ThreadPoolExecutor(max_workers=1) as pool:
-                        pool.submit(
-                            lambda: asyncio.run(_generate_edge_tts(text, file_str, tts_config))
-                        ).result(timeout=60)
-                except RuntimeError:
-                    asyncio.run(_generate_edge_tts(text, file_str, tts_config))
-            elif _check_neutts_available():
-                logger.info("Edge TTS not available, falling back to NeuTTS (local)...")
-                provider = "neutts"
-                _generate_neutts(text, file_str, tts_config)
-            else:
-                return json.dumps({
-                    "success": False,
-                    "error": "No TTS provider available. Install edge-tts (pip install edge-tts) "
-                             "or set up NeuTTS for local synthesis."
-                }, ensure_ascii=False)'''
-
-# ---------------------------------------------------------------------------
-# New block: elif provider == "piper" + else: fallback with Piper only
-# ---------------------------------------------------------------------------
-new_block = '''        elif provider == "piper":
-            # Piper TTS (local, CPU, no cloud, no Microsoft)
-            piper_binary = "/opt/hermes/.venv/bin/piper"
-            piper_config = tts_config.get("piper", {})
-            voice = piper_config.get("voice", "en_US-lessac-medium")
-            model_dir = piper_config.get("model_dir", "/opt/hermes/.venv/share/piper/voices")
-            model_path = os.path.join(model_dir, f"{voice}.onnx")
-            if not os.path.exists(model_path):
-                return json.dumps({
-                    "success": False,
-                    "error": "Piper TTS voice model not found. "
-                             "Install Piper TTS and download a voice model."
-                }, ensure_ascii=False)
-            logger.info("Generating speech with Piper TTS (local, CPU)...")
-            import subprocess as _sp
-            cmd = [piper_binary, "--model", model_path, "--output-raw"]
-            try:
-                proc = _sp.Popen(cmd, stdin=_sp.PIPE, stdout=_sp.PIPE, stderr=_sp.PIPE)
-                raw_audio, stderr = proc.communicate(input=text.encode(), timeout=60)
-                if proc.returncode != 0:
-                    raise RuntimeError(f"Piper TTS failed: {stderr.decode()[:200]}")
-                ffmpeg_cmd = ["ffmpeg", "-f", "s16le", "-ar", "22050", "-ac", "1", "-i", "-", "-y", file_str]
-                _sp.run(ffmpeg_cmd, input=raw_audio, capture_output=True, timeout=30)
-            except Exception as e:
-                return json.dumps({
-                    "success": False,
-                    "error": f"Piper TTS failed: {e}"
-                }, ensure_ascii=False)
-
-        else:
-            # Default: Piper TTS (local, CPU, no cloud, no Microsoft)
-            piper_binary = "/opt/hermes/.venv/bin/piper"
-            piper_config = tts_config.get("piper", {})
-            voice = piper_config.get("voice", "en_US-lessac-medium")
-            model_dir = piper_config.get("model_dir", "/opt/hermes/.venv/share/piper/voices")
-            model_path = os.path.join(model_dir, f"{voice}.onnx")
-            if os.path.exists(model_path) and os.path.exists(piper_binary):
-                logger.info("Generating speech with Piper TTS (local, CPU)...")
-                import subprocess as _sp
-                cmd = [piper_binary, "--model", model_path, "--output-raw"]
-                try:
-                    proc = _sp.Popen(cmd, stdin=_sp.PIPE, stdout=_sp.PIPE, stderr=_sp.PIPE)
-                    raw_audio, stderr = proc.communicate(input=text.encode(), timeout=60)
-                    if proc.returncode != 0:
-                        raise RuntimeError(stderr.decode()[:200])
-                    ffmpeg_cmd = ["ffmpeg", "-f", "s16le", "-ar", "22050", "-ac", "1", "-i", "-", "-y", file_str]
-                    _sp.run(ffmpeg_cmd, input=raw_audio, capture_output=True, timeout=30)
-                except Exception:
-                    pass
-            else:
-                return json.dumps({
-                    "success": False,
-                    "error": "Piper TTS not available. Install piper-tts and download a voice model."
-                }, ensure_ascii=False)'''
-
-# ---------------------------------------------------------------------------
-# Apply the patch to all copies found
-# ---------------------------------------------------------------------------
-patched_any = False
-
-for tts_path in found_paths:
-    with open(tts_path) as f:
-        code = f.read()
-
-    if 'provider == "piper"' in code:
-        print(f"ALREADY PATCHED: {tts_path}")
-        continue
-
-    if old_else in code:
-        code = code.replace(old_else, new_block, 1)
-        with open(tts_path, 'w') as f:
-            f.write(code)
-        print(f"PATCHED: {tts_path}")
-        patched_any = True
-    else:
-        print(f"SKIP {tts_path}: Edge fallback pattern not found")
-        import re
-        for m in re.finditer(r'        else:\n            # Default:', code):
-            start = max(0, m.start() - 100)
-            end = min(len(code), m.end() + 300)
-            print(f"  Found 'else:/# Default:' at position {m.start()}:")
-            print(f"  {code[start:end]}")
-            print("  ---")
-        # Don't exit with error — if one copy isn't patchable, try the others
-
-if not patched_any:
-    all_patched = all(
-        'provider == "piper"' in open(p).read()
-        for p in found_paths
-    )
-    if all_patched:
-        print("All copies already patched.")
-        sys.exit(0)
-    print("WARNING: Could not patch any copy of tts_tool.py")
-    sys.exit(1)
-
-print("tts_tool.py patched successfully across all copies.")

ai/hermes/run-multi-gateways.sh

@@ -1,32 +0,0 @@
-#!/bin/bash
-# Multi-gateway launcher for HERMES_PROFILES env var.
-# Reads comma-separated profile names, spawns one gateway per profile.
-# Designed to run before the main entrypoint — gateways run in background.
-set -e
-
-if [ -z "${HERMES_PROFILES}" ]; then
-  echo "HERMES_PROFILES not set — skipping multi-gateway launch"
-  exit 0
-fi
-
-# Source venv to make 'hermes' available (entrypoint.sh sources it later,
-# but we need it NOW for the background gateways)
-HERMES_BIN="/opt/hermes/.venv/bin/hermes"
-if [ ! -x "$HERMES_BIN" ]; then
-  echo "ERROR: hermes binary not found at $HERMES_BIN"
-  exit 1
-fi
-
-mkdir -p /opt/data/logs
-
-IFS=',' read -ra PROFILES <<< "${HERMES_PROFILES}"
-for profile in "${PROFILES[@]}"; do
-  profile="$(echo "${profile}" | xargs)"  # trim whitespace
-  [ -z "${profile}" ] && continue
-
-  echo "Starting gateway for profile: ${profile}"
-  nohup env API_SERVER_ENABLED=false API_SERVER_KEY= gosu hermes "$HERMES_BIN" --profile "${profile}" gateway run \
-      >> "/opt/data/logs/gateway-${profile}.log" 2>&1 &
-done
-
-echo "All gateways launched: ${HERMES_PROFILES}"

ai/hermes/start-hermes.sh

@@ -1,38 +0,0 @@
-#!/bin/bash
-# Multi-profile + default gateway launcher — runs as the CMD via s6-overlay.
-#
-# The image's default ENTRYPOINT ["/init", "/opt/hermes/docker/main-wrapper.sh"]
-# starts the s6 supervision tree, then exec's main-wrapper.sh with the CMD args.
-# main-wrapper.sh sources the venv, drops to the hermes user via s6-setuidgid,
-# and exec's this script.
-#
-# This script:
-#   1. Launches per-profile background gateways (HERMES_PROFILES env var)
-#   2. Starts the default gateway in foreground (keeps the container alive)
-#
-# Replaces the old approach of chaining bash -> tini -g -> deprecated entrypoint.sh
-# which bypassed s6-overlay and caused the SIGTERM crash loop.
-
-set -e
-
-HERMES_BIN="/opt/hermes/.venv/bin/hermes"
-
-# --- Multi-profile gateways (background) ---
-if [ -n "${HERMES_PROFILES:-}" ]; then
-    echo "[start-hermes] Launching per-profile gateways: ${HERMES_PROFILES}"
-    IFS=',' read -ra PROFILES <<< "${HERMES_PROFILES}"
-    for profile in "${PROFILES[@]}"; do
-        profile="$(echo "${profile}" | xargs)"  # trim whitespace
-        [ -z "${profile}" ] && continue
-        echo "[start-hermes]  -> background gateway for profile '${profile}'"
-        # No gosu/s6-setuidgid needed — we're already running as the hermes user
-        # (main-wrapper.sh drops privileges before exec'ing this script).
-        nohup "${HERMES_BIN}" --profile "${profile}" gateway run \
-            >> "/opt/data/logs/gateway-${profile}.log" 2>&1 &
-    done
-    echo "[start-hermes] All profile gateways launched"
-fi
-
-# --- Default gateway (foreground — keeps container alive) ---
-echo "[start-hermes] Starting default gateway (foreground)"
-exec "${HERMES_BIN}" gateway run

ai/honcho/Dockerfile

@@ -28,9 +28,10 @@ ENV PATH=$PNPM_HOME:$PATH
 RUN corepack enable && corepack prepare pnpm@latest --activate
 
 WORKDIR /app
-RUN apt-get update && apt-get install -y git && rm -rf /var/lib/apt/lists/*
+RUN apt-get update && apt-get install -y git openssh-client && rm -rf /var/lib/apt/lists/**
 
-ARG OPENCONCHO_SHA=3b5c3293fc18d768dbe85285264a8d66c896bd81
+ARG OPENCONCHO_SHA=148766312e6a5256f399c64171854f0ff6173b6d
+RUN mkdir -p -m 0700 ~/.ssh && ssh-keyscan -p 2222 code.lazyworkhorse.net >> ~/.ssh/known_hosts 2>/dev/null
 RUN --mount=type=ssh git clone --depth 1 ssh://git@code.lazyworkhorse.net:2222/gortium/openconcho.git /app && \
     git -C /app fetch --depth 1 origin ${OPENCONCHO_SHA} && \
     git -C /app checkout ${OPENCONCHO_SHA}

ai/llama-cpp/Dockerfile

@@ -0,0 +1,32 @@
+# llama-cpp-rocm6/Dockerfile
+# Custom llama.cpp server with ROCm 6.1 + gfx906 (MI50) support.
+# Build: docker build -t llama-cpp:rocm-gfx906 .
+# NOTE: Update LLAMACPP_VERSION for bugfixes, but verify ROCm 6.1 compat
+#       by checking if recent releases still support -DGGML_HIP=ON.
+
+FROM rocm/dev-ubuntu-22.04:6.1.2-complete AS builder
+RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y curl git build-essential pkg-config cmake make && rm -rf /var/lib/apt/lists/*
+ARG LLAMACPP_VERSION=b9890
+RUN git clone --depth 1 --branch ${LLAMACPP_VERSION} https://github.com/ggml-org/llama.cpp.git /build
+WORKDIR /build
+ENV HIP_PATH=/opt/rocm ROCM_PATH=/opt/rocm PATH=/opt/rocm/bin:/opt/rocm/llvm/bin:${PATH} CMAKE_PREFIX_PATH=/opt/rocm
+RUN mkdir build && cd build && \
+    cmake .. -DGGML_HIP=ON -DCMAKE_BUILD_TYPE=Release \
+      -DAMDGPU_TARGETS="gfx906:xnack-" \
+      -DCMAKE_POSITION_INDEPENDENT_CODE=ON \
+      -DGGML_CUDA=OFF -DGGML_VULKAN=OFF -DGGML_METAL=OFF \
+      -DBUILD_SHARED_LIBS=OFF && \
+    cmake --build . --target llama-server -- -j $(nproc)
+
+FROM ubuntu:24.04
+RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y \
+    ca-certificates curl libstdc++6 libgomp1 libopenblas0 \
+    libnuma1 libelf1 libdrm2 libdrm-amdgpu1 \
+    && rm -rf /var/lib/apt/lists/*
+COPY --from=builder /opt/rocm/lib/ /opt/rocm/lib/
+COPY --from=builder /opt/rocm/share/ /opt/rocm/share/
+COPY --from=builder /build/build/bin/llama-server /usr/local/bin/llama-server
+RUN echo /opt/rocm/lib > /etc/ld.so.conf.d/rocm.conf && ldconfig
+ENV HSA_OVERRIDE_GFX_VERSION=9.0.6 HCC_AMDGPU_TARGET=gfx906 HSA_ENABLE_SDMA=0
+EXPOSE 8080
+ENTRYPOINT ["/usr/local/bin/llama-server"]