fix: add start-hermes.sh baked into image, replace broken ENTRYPOINT

- New start-hermes.sh: multi-profile launcher that works with s6-overlay's
  main-program model (replaces bash->tini->entrypoint.sh chain)
- Dockerfile: COPY start-hermes.sh into /usr/local/bin/ alongside
  run-multi-gateways.sh (which is now unused but kept for reference)
- compose.yml: remove entrypoint override, CMD now points at
  /usr/local/bin/start-hermes.sh via the image default ENTRYPOINT

Fixes the SIGTERM crash loop caused by S6_CMD_ARG0 being unset when the
deprecated entrypoint.sh shim bypassed s6-overlay's /init.

ai/compose.yml

@@ -6,19 +6,19 @@ services:
       ssh:
         - default
     container_name: hermes
-    entrypoint: ["/bin/bash", "-c",
-      "bash /usr/local/bin/run-multi-gateways.sh && exec /usr/bin/tini -g -- /opt/hermes/docker/entrypoint.sh \"$@\"",
-      "hermes-entrypoint"]
     restart: always
-    # Gateway run enables the internal API server on port 8642
-    command: gateway run
+    # 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"]
     environment:
       - HERMES_UID=10000
       - HERMES_GID=10000
-      - OLLAMA_HOST=http://ollama-cpu:11434
+      - OLLAMA_HOST=http://ollama:11434
       - HERMES_DASHBOARD=1
       # Multi-profile: comma-separated list of profiles to run as gateways.
-      # The entrypoint reads this and starts one gateway per profile.
+      # 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
@@ -108,12 +108,12 @@ services:
 
       - "traefik.http.services.syncthing.loadbalancer.server.port=8384"
 
-  ollama-cpu:
+  ollama:
     build:
       context: ./ollama
       dockerfile: Dockerfile
     image: ollama/ollama:rocm-gfx906
-    container_name: ollama-cpu
+    container_name: ollama
     tty: true
     restart: always 
     ports:
@@ -124,42 +124,22 @@ 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
-
-  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
+      - OLLAMA_DEBUG=1
+      - OLLAMA_FLASH_ATTENTION=1
+      - OLLAMA_NUM_PARALLEL=2
     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:
@@ -248,6 +228,48 @@ 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

@@ -62,6 +62,11 @@ PYEOF
 # 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/start-hermes.sh

@@ -0,0 +1,38 @@
+#!/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/llama-cpp/Dockerfile

@@ -1,30 +0,0 @@
-# llama-cpp-rocm6/Dockerfile
-# Custom llama.cpp server with ROCm 6.1 + gfx906 (MI50) support.
-# Build: docker build -t llama-cpp:rocm-gfx906 .
-
-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=b9596
-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"]