refactor: split ai/ into hermes/ and ollama/ directories

- Rename ai/ to hermes/ (Hermes agent service) - Move ai/ollama/ to ollama/ (top-level, Ollama ROCm/gfx906 service) - Add ollama/compose.yml for standalone ollama deployment - Update hermes/compose.yml: remove ollama service (now in ollama/) - Update Makefile SERVICES list: ai -> hermes, add ollama - Bump ollama version from v0.13.5 to v0.23.2
2026-05-09 21:44:46 -04:00
18 changed files with 253 additions and 893 deletions
--- a/.gitea/workflows/build-hermes.yml
+++ b/.gitea/workflows/build-hermes.yml
@@ -1,31 +0,0 @@
-name: Build Hermes agent
-on:
-  pull_request:
-    branches: [ master ]
-    paths:
-      - 'ai/hermes/**'
-      - 'ai/compose.yml'
-  push:
-    branches: [ master ]
-    paths:
-      - 'ai/hermes/**'
-      - 'ai/compose.yml'
-
-jobs:
-  build:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout
-        run: |
-          git clone -b "${{ github.head_ref || github.ref_name }}" \
-            https://gitea:${{ secrets.GITHUB_TOKEN }}@code.lazyworkhorse.net/gortium/compose.git .
-          git log --oneline -3
-
-      - name: Build hermes image
-        run: |
-          cd ai
-          docker compose build hermes 2>&1
-
-      - name: Verify image
-        run: |
-          docker run --rm ai-hermes /opt/hermes/.venv/bin/python --version 2>&1
--- a/.gitea/workflows/build-ollama.yml
+++ b/.gitea/workflows/build-ollama.yml
@@ -1,31 +0,0 @@
-name: Build ollama (gfx906)
-on:
-  pull_request:
-    branches: [ master ]
-    paths:
-      - 'ai/ollama/**'
-      - 'ai/compose.yml'
-  push:
-    branches: [ master ]
-    paths:
-      - 'ai/ollama/**'
-      - 'ai/compose.yml'
-
-jobs:
-  build:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout
-        run: |
-          git clone -b "${{ github.head_ref || github.ref_name }}" \
-            https://gitea:${{ secrets.GITHUB_TOKEN }}@code.lazyworkhorse.net/gortium/compose.git .
-          git log --oneline -3
-
-      - name: Build ollama image
-        run: |
-          cd ai
-          docker compose build ollama --no-cache 2>&1
-
-      - name: Verify version
-        run: |
-          docker run --rm ollama/ollama:rocm-gfx906 ollama --version 2>&1
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -1,31 +0,0 @@
-# AGENTS.md
-
-Development conventions for the compose repository (Docker Compose stacks).
-
-## Build & Deploy
-
- The ai stack is managed via `systemctl restart ai_stack.service` on the NixOS host
- Compose files are built from the git repo; apply via `nh os switch` or `systemctl restart ai_stack.service`
- Never `docker compose up -d` directly — it bypasses the systemd env file and breaks secrets
-
-## Hermes Workers
-
- Paperclip Hermes workers are added via `ai/scripts/provision-hermes-worker.sh`
- The script appends only — never deletes or modifies existing content
- Workers are CPU-only containers on the `ai_backend` network with no GPU passthrough
-
-## Workflow
-
- New feature → clean branch from `origin/master` → push → PR on Gitea
- Branch naming: `feat/description` (features), `fix/description` (bugs/docs)
- Always branch from `origin/master`, never from another feature branch
- Submodule changes (when this repo is consumed as a submodule): commit the submodule update in the parent repo
- PR title should describe the change; body should explain motivation + summary
- After PR merge, delete the feature branch
-
-## YAML Conventions
-
- Use `x-*` extension fields for reusable anchors
- Comment out inactive services rather than deleting them
- Environment variables in dictionary format (`KEY: value`) for anchor compatibility
- List format (`- KEY=value`) works for standard services but doesn't merge with anchors
--- a/2
+++ b/2
@@ -2,7 +2,7 @@
 COMPOSE_PATH=~/Projects/AltNet/docker-compose

 # List of services (folder names)
-SERVICES=monitoring ai cloudstorage crm_tp crm_cf mediacenter homeautomation network backup homepage passwordmanager
+SERVICES=monitoring hermes ollama cloudstorage crm_tp crm_cf mediacenter homeautomation network backup homepage passwordmanager

 # Bring up all services
 all_up:
--- a/ai/compose.yml
+++ b/ai/compose.yml
@@ -1,357 +0,0 @@
-version: "3.8"
-
-# ── Hermes Worker Template ──────────────────────────────────
-# Used by paperclip-worker-* Hermes containers via YAML anchor.
-# Each worker = one isolated Hermes agent for a Paperclip employee.
-# Override at service level: container_name, API_SERVER_PORT,
-# API_SERVER_KEY, DISCORD_BOT_TOKEN, volumes.
-# Workers have NO GPU — they use OpenCode Go or remote providers.
-x-hermes-worker: &hermes-worker
-  build:
-    context: ./hermes
-    ssh:
-      - default
-  entrypoint: ["/bin/bash", "-c",
-    "bash /opt/data/hermes-tools/install.sh && exec /usr/bin/tini -g -- /opt/hermes/docker/entrypoint.sh \"$@\"",
-    "hermes-entrypoint"]
-  command: gateway run
-  restart: always
-  environment:
-    API_SERVER_ENABLED: "true"
-    API_SERVER_HOST: "0.0.0.0"
-    OLLAMA_HOST: "http://ollama:11434"
-    OPENROUTER_API_KEY: ${OPENROUTER_API_KEY}
-    # Each worker needs its own OpenCode Go API key in .env
-    OPENCODE_GO_API_KEY: ${OPENCODE_GO_API_KEY}
-    GATEWAY_ALLOW_ALL_USERS: "true"
-    TZ: "America/Montreal"
-  networks:
-    ai_backend:
-  # NO devices — workers are CPU-only, no GPU passthrough
-# ─────────────────────────────────────────────────────────────
-
-services:
-
-  # webui:
-  #   image: ghcr.io/open-webui/open-webui:main
-  #   volumes:
-  #     - /mnt/HoardingCow_docker_data/Ollama/open-webui:/app/backend/data
-  #   restart: always
-  #   environment:
-  #     - OLLAMA_API_BASE_URL=http://ollama:11434/api
-  #   networks:
-  #     - ai_net
-  #     - ai_backend
-  #   labels:
-  #     - "traefik.enable=true"
-
-  #     # Router for HTTP + redirection to HTTPS
-  #     - "traefik.http.routers.webui-http.rule=Host(`ai.lazyworkhorse.net`)"
-  #     - "traefik.http.routers.webui-http.entrypoints=web"
-  #     - "traefik.http.routers.webui-http.middlewares=redirect-to-https"
-
-  #     # Router for HTTPS with TLS
-  #     - "traefik.http.routers.webui-https.rule=Host(`ai.lazyworkhorse.net`)"
-  #     - "traefik.http.routers.webui-https.entrypoints=websecure"
-  #     - "traefik.http.routers.webui-https.tls=true"
-  #     - "traefik.http.routers.webui-https.tls.certresolver=njalla"
-
-  hermes:
-    build:
-      context: ./hermes
-      ssh:
-        - default
-    container_name: hermes
-    entrypoint: ["/bin/bash", "-c",
-      "bash /opt/data/hermes-tools/install.sh && exec /usr/bin/tini -g -- /opt/hermes/docker/entrypoint.sh \"$@\"",
-      "hermes-entrypoint"]
-    restart: always
-    # Gateway run enables the internal API server on port 8642
-    command: gateway run
-    environment:
-      - OLLAMA_HOST=http://ollama:11434
-      - API_SERVER_ENABLED=true
-      - API_SERVER_PORT=8642
-      - API_SERVER_HOST=0.0.0.0
-      - API_SERVER_KEY=hermes_local_key
-      - GATEWAY_ALLOW_ALL_USERS=true
-      - OPENROUTER_API_KEY=${OPENROUTER_API_KEY}
-      # ROCm for GPU-accelerated faster-whisper STT
-      - 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
-      - TZ=America/Montreal
-    volumes:
-      - /mnt/HoardingCow_docker_data/Hermes/data:/opt/data
-    devices:
-      - /dev/kfd:/dev/kfd
-      - /dev/dri:/dev/dri
-    group_add:
-      - "303"
-      - "26"
-    networks:
-      - ai_backend
-
-  ollama:
-    build:
-      context: ./ollama
-      dockerfile: Dockerfile
-    image: ollama/ollama:rocm-gfx906
-    container_name: ollama
-    tty: true
-    restart: always 
-    ports:
-      - "127.0.0.1:11434:11434"
-    networks:
-      - ai_backend
-    volumes:
-      - /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
-    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"
-
-# ── Paperclip Worker Hermes Agents ──────────────────────────
-# Each worker is an isolated Hermes agent for a Paperclip employee.
-# Add new workers with: ./scripts/provision-hermes-worker.sh <name> <discord-token>
-# The API server key and port are generated automatically.
-# Workers are CPU-only — they use OpenCode Go or remote providers.
-
-  # ── Worker Template (commented — uncomment + configure to activate) ──
-  # hermes-worker-1:
-  #   <<: *hermes-worker
-  #   container_name: hermes-worker-1
-  #   environment:
-  #     API_SERVER_PORT: "8651"
-  #     API_SERVER_KEY: "generated-by-provision-script"
-  #     DISCORD_BOT_TOKEN: ${WORKER_1_DISCORD_BOT_TOKEN}
-  #   volumes:
-  #     - /mnt/HoardingCow_docker_data/Hermes/worker-1:/opt/data
-# ─────────────────────────────────────────────────────────────
-
-networks:
-  ai_net:
-    external: true
-    name: ai_net
-  ai_backend:
-    driver: bridge
-    name: ai_backend
-    
-  # 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
-  #   # Required for multi-GPU communication (NCCL)
-  #   ipc: host 
-  #   init: true
-  #   shm_size: '2g' 
-  #   networks:
-  #     - ai_backend
-  #   ports:
-  #     - "8300:8000"
-  #   devices:
-  #     - "/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
-  #     NCCL_P2P_DISABLE: 1
-  #     VLLM_WORKER_MULTIPROC_METHOD: spawn
-  #     VLLM_USE_TRITON_FLASH_ATTN: 0
-  #     VLLM_USE_ROCM_CUSTOM_PAGED_ATTN: 0
-  #     VLLM_ATTENTION_BACKEND: ROPE_NAIVE
-  #     VLLM_SKIP_WARMUP: 1
-  #     VLLM_USE_V1: 0
-  #     HF_TOKEN: ${HF_TOKEN}
-  #   command: >
-  #     vllm serve "mistralai/Devstral-Small-2-24B-Instruct-2512" 
-  #     --tensor-parallel-size 2
-  #     --max-model-len 8192
-  #     --gpu-memory-utilization 0.90
-  #     --tokenizer_mode mistral
-  #     --config_format auto
-  #     --load-format auto
-  #     --enforce-eager
-  #     --disable-custom-all-reduce
-  #     --trust-remote-code
-  #     --task generate
-  #     --block-size 16
-  #   volumes:
-  #     - /mnt/HoardingCow_docker_data/vllm/models:/root/.cache/huggingface
-  #   restart: unless-stopped
-
-  # n8n:
-  #   image: n8nio/n8n:latest
-  #   container_name: n8n
-  #   restart: unless-stopped
-  #   networks:
-  #     - ai_net
-  #   environment:
-  #     - N8N_HOST=n8n.lazyworkhorse.net
-  #     - N8N_PORT=5678
-  #     - N8N_PROTOCOL=https
-  #     - NODE_ENV=production
-  #     - N8N_ENCRYPTION_KEY=${N8N_ENCRYPTION_KEY}
-  #     - WEBHOOK_URL=https://n8n.lazyworkhorse.net/
-  #     - GENERIC_TIMEZONE=America/New_York # Adjust to your timezone
-  #     - N8N_BLOCK_EXTERNAL_STORAGE_ACCESS=false
-  #     - N8N_NODES_PYTHON_CAN_IMPORT_MODULES=true 
-  #     - N8N_NATIVE_PYTHON_RUNNER=true
-  #     - N8N_PYTHON_ALLOW_STDLIB=uuid,re,os,json
-  #     - N8N_PYTHON_ALLOW_EXTERNAL=requests,pandas
-  #     - NODE_FUNCTION_ALLOW_EXTERNAL=uuid,requests
-  #   volumes:
-  #     - /mnt/HoardingCow_docker_data/n8n:/home/node/.n8n
-  #   labels:
-  #     - "traefik.enable=true"
-
-  #     # Router for HTTP + redirection to HTTPS
-  #     - "traefik.http.routers.n8n-http.rule=Host(`n8n.lazyworkhorse.net`)"
-  #     - "traefik.http.routers.n8n-http.entrypoints=web"
-  #     - "traefik.http.routers.n8n-http.middlewares=redirect-to-https"
-
-  #     # Router for HTTPS with TLS
-  #     - "traefik.http.routers.n8n-https.rule=Host(`n8n.lazyworkhorse.net`)"
-  #     - "traefik.http.routers.n8n-https.entrypoints=websecure"
-  #     - "traefik.http.routers.n8n-https.tls=true"
-  #     - "traefik.http.routers.n8n-https.tls.certresolver=njalla"
-
-  #     # Service Loadbalancer (n8n default port)
-  #     - "traefik.http.services.n8n.loadbalancer.server.port=5678"
-
-  # openclaw:
-  #   image: coollabsio/openclaw:latest
-  #   container_name: openclaw
-  #   restart: unless-stopped
-  #   expose:
-  #     - "8080"  # WebUI
-  #     - "18789" # Gateway/WebSocket
-  #     - "8788"  # Nextcloud Webhook
-  #   networks:
-  #     - ai_net
-  #     - ai_backend
-  #   volumes:
-  #     - /mnt/HoardingCow_docker_data/openclaw/data:/data
-  #     - /home/gortium/infra:/data/workspace/infra
-  #   environment:
-  #     - TZ=America/Toronto
-  #     - OPENCLAW_GATEWAY_TOKEN=${OPENCLAW_GATEWAY_TOKEN}
-  #     - OPENROUTER_API_KEY=${OPENROUTER_API_KEY}
-  #     # Point to the sidecar browser
-  #     - BROWSER_CDP_URL=http://openclaw-browser:9222
-  #     - BROWSER_EVALUATE_ENABLED=true
-  #     - OPENCLAW_GATEWAY_HOST=0.0.0.0
-  #     - OPENCLAW_ALLOWED_ORIGINS=https://claw.lazyworkhorse.net
-  #   labels:
-  #     - "traefik.enable=true"
-
-  #     - "traefik.http.routers.openclaw-http.rule=Host(`claw.lazyworkhorse.net`)"
-  #     - "traefik.http.routers.openclaw-http.entrypoints=web"
-  #     - "traefik.http.routers.openclaw-http.middlewares=redirect-to-https"
-
-  #     - "traefik.http.routers.openclaw-https.rule=Host(`claw.lazyworkhorse.net`)"
-  #     - "traefik.http.routers.openclaw-https.priority=50"
-  #     - "traefik.http.routers.openclaw-https.entrypoints=websecure"
-  #     - "traefik.http.routers.openclaw-https.tls=true"
-  #     - "traefik.http.routers.openclaw-https.tls.certresolver=njalla"
-  #     - "traefik.http.services.openclaw.loadbalancer.server.port=8080"
-  #   depends_on:
-  #     - openclaw-browser
-
-  # openclaw-browser:
-  #   image: ghcr.io/browserless/chromium:latest
-  #   restart: always
-  #   expose:
-  #     - "3000"
-  #   environment:
-  #     - MAX_CONCURRENT_SESSIONS=10
-  #     - CONNECTION_TIMEOUT=300000
-  #     - PREBOOT_CHROME=true
-  #     - DEMO_MODE=false
-  #   networks:
-  #     ai_backend:
-  #       aliases:
-  #         - browser
-
-  # openclaw-ssh:
-  #   image: linuxserver/openssh-server:latest
-  #   container_name: openclaw-ssh
-  #   environment:
-  #     - PUID=1000
-  #     - PGID=1000
-  #     - PUBLIC_KEY_FILE=/config/ssh/authorized_keys
-  #     - SUDO_ACCESS=false
-  #     - PASSWORD_ACCESS=false
-  #   volumes:
-  #     - /mnt/HoardingCow_docker_data/openclaw/ssh-config:/config
-  #     - /home/gortium/infra:/data/workspace/infra:ro
-  #   restart: unless-stopped
-  #   networks:
-  #     - ai_backend
-  #   labels:
-  #     - "traefik.enable=true"
-  #     - "traefik.tcp.routers.openclaw-ssh.rule=HostSNI(*)"
-  #     - "traefik.tcp.routers.openclaw-ssh.entrypoints=sshnode"
-  #     - "traefik.tcp.routers.openclaw-ssh.tls.passthrough=false"
-  #     - "traefik.tcp.services.openclaw-ssh.loadbalancer.server.port=2222"
--- a/ai/hermes/Dockerfile
+++ b/ai/hermes/Dockerfile
@@ -1,93 +0,0 @@
-# syntax=docker/dockerfile:1
-# Hermes Agent -- custom fork build
-# Builds on top of official image + overlays our forked source from Gitea.
-# Requires Docker BuildKit. Pass SSH agent for git clone:
-#   docker compose build hermes
-# Or manually:
-#   DOCKER_BUILDKIT=1 docker build --ssh default -t hermes-agent:custom .
-
-# ---------- Base: official Hermes image (system deps, npm, uv, Playwright) ----------
-FROM nousresearch/hermes-agent:latest
-
-# ---------- Overlay our forked source ----------
-# Uses SSH agent forwarding from the build host (no key baked into image).
-# --exclude node_modules/.venv keeps the base image's pre-built layers intact.
-# Only the Python source, web UI source, and config change.
-RUN --mount=type=ssh \
-    mkdir -p /root/.ssh && \
-    ssh-keyscan -p 2222 code.lazyworkhorse.net >> /root/.ssh/known_hosts 2>/dev/null && \
-    cd /tmp && \
-    GIT_SSH_COMMAND='ssh -p 2222 -o StrictHostKeyChecking=no' \
-    git clone --depth 1 --branch main \
-    git@code.lazyworkhorse.net:gortium/hermes-agent.git fork && \
-    rsync -a --delete fork/ /opt/hermes/ \
-      --exclude node_modules \
-      --exclude .venv \
-      --exclude .git && \
-    rm -rf /tmp/fork /root/.ssh/
-
-# ---------- Rebuild web UI ----------
-# Source files changed; node_modules (from base image) reused.
-RUN cd /opt/hermes && npm run build
-
-# ---------- Reinstall Python package (editable) ----------
-# Picks up source changes from our fork.
-RUN . /opt/hermes/.venv/bin/activate && \
-    uv pip install --no-cache-dir --no-deps -e /opt/hermes
-
-# ---------- Extra system deps ----------
-USER root
-RUN apt-get update && \
-    apt-get install -y --no-install-recommends \
-        libportaudio2 ca-certificates poppler-utils imagemagick \
-        texlive-latex-base texlive-latex-extra texlive-fonts-recommended \
-        texlive-xetex texlive-science \
-        qemu-user-static binfmt-support emacs-nox && \
-    rm -rf /var/lib/apt/lists/*
-
-# ---------- UV ----------
-COPY --chmod=0755 --from=ghcr.io/astral-sh/uv:latest /uv /usr/local/bin/
-
-# ---------- Piper TTS ----------
-RUN . /opt/hermes/.venv/bin/activate && \
-    uv pip install --no-cache-dir piper-tts sounddevice numpy && \
-    mkdir -p /opt/hermes/.venv/share/piper/voices
-
-RUN /opt/hermes/.venv/bin/python3 /dev/stdin << 'PYEOF'
-import urllib.request
-base = '/opt/hermes/.venv/share/piper/voices'
-url = 'https://huggingface.co/rhasspy/piper-voices/resolve/main/en/en_US/ryan/high/en_US-ryan-high.onnx'
-urllib.request.urlretrieve(url, base + '/en_US-ryan-high.onnx')
-urllib.request.urlretrieve(url + '.json', base + '/en_US-ryan-high.onnx.json')
-PYEOF
-
-# ---------- Install Himalaya email CLI ----------
-RUN /opt/hermes/.venv/bin/python3 /dev/stdin << 'PYEOF'
-import urllib.request, tarfile, os, shutil
-url = 'https://github.com/pimalaya/himalaya/releases/download/v1.2.0/himalaya.x86_64-linux.tgz'
-tgz = '/tmp/himalaya.tgz'
-urllib.request.urlretrieve(url, tgz)
-with tarfile.open(tgz) as t:
-    t.extractall('/tmp')
-shutil.move('/tmp/himalaya', '/usr/local/bin/himalaya')
-os.chmod('/usr/local/bin/himalaya', 0o755)
-os.remove(tgz)
-print('himalaya v1.2.0 installed')
-PYEOF
-
-# ---------- Install himalaya-ro wrapper ----------
-COPY --chmod=0755 himalaya-ro.sh /usr/local/bin/himalaya-ro
-
-
-# ---------- Runtime ----------
-USER hermes
-ENV HERMES_HOME=/opt/data
-ENV PATH="/opt/data/.local/bin:${PATH}"
-# Point browser tool to Playwright's Chromium (already in base image)
-ENV CHROME_EXECUTABLE=/opt/hermes/.playwright/chromium/chrome-linux/chrome
-
-# Ensure tools directory and toolsets.py are writable by the hermes runtime user
-# so custom tools can be injected from the persistent volume at startup.
-RUN chown -R hermes:hermes /opt/hermes/tools /opt/hermes/toolsets.py
-
-VOLUME [ "/opt/data" ]
--- a/ai/hermes/himalaya-ro.sh
+++ b/ai/hermes/himalaya-ro.sh
@@ -1,73 +0,0 @@
-#!/usr/bin/env bash
-# ─────────────────────────────────────────────────────────────
-# himalaya-ro — Read-only wrapper for himalaya
-#
-# Blocks destructive commands and logs audit trail.
-# Pass-through for read-only commands (list, read, search).
-#
-# Usage:  himalaya-ro [options] <command> [args...]
-#
-# Install: place in PATH before the real himalaya, or use
-#          `ln -sf himalaya-ro /usr/local/bin/himalaya`
-# ─────────────────────────────────────────────────────────────
-set -o pipefail
-
-# ── Configuration ───────────────────────────────────────────
-HIMALAYA_BIN="${HIMALAYA_BIN:-/usr/local/bin/himalaya}"
-AUDIT_LOG="${HIMALAYA_AUDIT_LOG:-/var/log/himalaya-audit.log}"
-
-# ── Destructive commands we block ──────────────────────────
-BLOCKED_CMDS=(
-  "message move"
-  "message delete"
-  "message copy"
-  "flag add"
-  "flag remove"
-  "folder create"
-  "folder delete"
-  "folder rename"
-  "template send"
-  "account configure"
-  "account delete"
-)
-
-# ── Determine the subcommand being invoked ─────────────────
-# Strip leading options (--account, --output, etc.) to find the verb
-ARGS=()
-SKIP_NEXT=false
-for arg in "$@"; do
-  if $SKIP_NEXT; then
-    SKIP_NEXT=false
-    continue
-  fi
-  if [[ "$arg" == --* ]]; then
-    case "$arg" in
-      --account|--output|--page|--page-size|--folder|--color|--format)
-        SKIP_NEXT=true ;;
-    esac
-    continue
-  fi
-  ARGS+=("$arg")
-done
-
-# Build subcommand string and check against blocklist
-CMD_STR=""
-for ((i=0; i<${#ARGS[@]}; i++)); do
-  if [ -z "$CMD_STR" ]; then
-    CMD_STR="${ARGS[$i]}"
-  else
-    CMD_STR="$CMD_STR ${ARGS[$i]}"
-  fi
-  for blocked in "${BLOCKED_CMDS[@]}"; do
-    if [[ "$CMD_STR" == "$blocked" ]]; then
-      TS=$(date '+%Y-%m-%d %H:%M:%S')
-      echo "[AUDIT] $TS BLOCKED: himalaya $*" >> "$AUDIT_LOG"
-      echo "ERROR: Command 'himalaya $CMD_STR ...' is blocked by read-only policy." >&2
-      echo "       Audit log: $AUDIT_LOG" >&2
-      exit 100
-    fi
-  done
-done
-
-# ── Allow pass-through ─────────────────────────────────────
-exec "$HIMALAYA_BIN" "$@"
--- a/ai/ollama/Dockerfile
+++ b/ai/ollama/Dockerfile
@@ -1,106 +0,0 @@
-# ollama-gfx906/Dockerfile
-#
-# Custom ollama image with ROCm 6.1 + gfx906 (MI50) support.
-# The official ollama/rocm image ships ROCm 7.2 which dropped gfx906.
-# This uses v0.23.2's native CMake build system with AMDGPU_TARGETS including gfx906.
-#
-# Build: docker build -t ollama/ollama:rocm-gfx906 ai/ollama
-
-FROM rocm/dev-ubuntu-22.04:6.1.2-complete AS builder
-
-# Build dependencies (CMake, Ninja, Go)
-ARG CMAKEVERSION=3.31.2
-ARG NINJAVERSION=1.12.1
-ARG GOLANG_VERSION=1.22.0
-
-RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y \
-    curl git ccache build-essential pkg-config unzip \
-    && rm -rf /var/lib/apt/lists/*
-
-# Install CMake from official binaries
-RUN curl -fsSL https://github.com/Kitware/CMake/releases/download/v${CMAKEVERSION}/cmake-${CMAKEVERSION}-linux-x86_64.tar.gz \
-    | tar xz -C /usr/local --strip-components 1
-
-# Install Ninja
-RUN curl -fsSL -o /tmp/ninja.zip \
-    https://github.com/ninja-build/ninja/releases/download/v${NINJAVERSION}/ninja-linux.zip \
-    && unzip /tmp/ninja.zip -d /usr/local/bin && rm /tmp/ninja.zip
-
-# Install Go
-RUN curl -fsSL https://go.dev/dl/go${GOLANG_VERSION}.linux-amd64.tar.gz \
-    | tar xz -C /usr/local
-ENV PATH=/usr/local/go/bin:$PATH
-
-ARG OLLAMA_VERSION=v0.23.2
-RUN git clone --depth 1 --branch ${OLLAMA_VERSION} https://github.com/ollama/ollama.git /build
-WORKDIR /build
-
-# ROCm paths
-ENV HIP_PATH=/opt/rocm
-ENV ROCM_PATH=/opt/rocm
-ENV CMAKE_GENERATOR=Ninja
-ENV LDFLAGS=-s
-
-# Step 1: Build CPU backends with GCC (no ROCm preset)
-# Pre-set CMAKE_HIP_COMPILER="" to prevent check_language(HIP) from
-# finding a HIP compiler (it searches /opt/rocm even without PATH).
-# Remove /opt/rocm from PATH to prevent find_program from finding hipcc.
-RUN mkdir -p build-cpu && \
-    PATH=/usr/local/go/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin \
-    cmake -B build-cpu -DCMAKE_BUILD_TYPE=Release \
-      -DCMAKE_HIP_COMPILER="" \
-      -DCMAKE_INSTALL_PREFIX=/build/dist && \
-    cmake --build build-cpu --target ggml-cpu -- -l $(nproc) && \
-    cmake --install build-cpu --component CPU --strip && \
-    echo "=== CPU install ===" && \
-    (find /build/dist/lib/ollama -type f -o -type l 2>&1 | head -20 || echo "empty")
-
-# Step 2: Build HIP backend with ROCm preset + gfx906 target only
-# The ROCm 6 preset enables HIP language detection (enable_language(HIP))
-# which ensures GPU kernels are properly compiled for gfx906.
-# OLLAMA_RUNNER_DIR=rocm from the preset, so HIP goes to lib/ollama/rocm/
-# Need CMAKE_PREFIX_PATH so find_package(hip) finds hip-config.cmake
-# at /opt/rocm/lib/cmake/hip/hip-config.cmake.
-RUN mkdir -p build-hip && \
-    cmake -B build-hip \
-      --preset 'ROCm 6' \
-      -DAMDGPU_TARGETS="gfx906:xnack-" \
-      -DCMAKE_BUILD_TYPE=Release \
-      -DCMAKE_PREFIX_PATH="/opt/rocm" && \
-    cmake --build build-hip --target ggml-hip -- -l $(nproc) && \
-    cmake --install build-hip --component HIP --strip && \
-    echo "=== HIP install ===" && \
-    find /build/dist/lib/ollama -type f -o -type l | head -20
-
-# Step 3: Build Go binary (GCC for CGo linking)
-ENV CGO_ENABLED=1
-RUN go build -trimpath -ldflags="-X=github.com/ollama/ollama/version.Version=${OLLAMA_VERSION}" -o /build/dist/ollama .
-
-# ---------- Runtime image ----------
-FROM ubuntu:24.04
-
-RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y \
-    ca-certificates curl libstdc++6 libgomp1 libvulkan1 libopenblas0 \
-    && rm -rf /var/lib/apt/lists/*
-
-# Copy ROCm 6.1 runtime libraries
-# These are needed at runtime by ggml-hip via LD_LIBRARY_PATH
-COPY --from=builder /opt/rocm/lib/ /opt/rocm/lib/
-COPY --from=builder /opt/rocm/share/ /opt/rocm/share/
-
-# Copy ollama binary + all backends (CPU + HIP)
-# CPU install:  /build/dist/lib/ollama/libggml-*.so
-# HIP install:  /build/dist/lib/ollama/rocm/libggml-hip.so
-COPY --from=builder /build/dist/ollama /usr/bin/ollama
-COPY --from=builder /build/dist/lib/ollama/ /usr/lib/ollama/
-
-RUN ldconfig
-
-ENV LD_LIBRARY_PATH=/opt/rocm/lib:/usr/lib/ollama/rocm:/usr/lib/ollama
-ENV HSA_OVERRIDE_GFX_VERSION=9.0.6
-ENV HCC_AMDGPU_TARGET=gfx906
-ENV HSA_ENABLE_SDMA=0
-
-EXPOSE 11434
-ENTRYPOINT ["/bin/ollama"]
-CMD ["serve"]
--- a/ai/scripts/provision-hermes-worker.sh
+++ b/ai/scripts/provision-hermes-worker.sh
@@ -1,135 +0,0 @@
-#!/usr/bin/env bash
-set -euo pipefail
-
-# ── Hermes Worker Provisioner ──────────────────────────────
-# Adds a new Paperclip Hermes worker to the ai compose stack.
-#
-# Usage:
-#   ./provision-hermes-worker.sh <name> <discord_bot_token_var>
-#
-# Example:
-#   ./provision-hermes-worker.sh worker-1 WORKER_1_DISCORD_BOT_TOKEN
-#
-# The script APPENDS only — never modifies or removes existing
-# content, even commented lines.
-#
-# Post-provision steps (manual):
-#   1. Add secrets to agenix .env file
-#   2. systemctl restart ai_stack.service
-#   3. Configure Paperclip agent
-# ─────────────────────────────────────────────────────────────
-
-NAME="${1:?Usage: $0 <name> <discord_bot_token_var>}"
-TOKEN_VAR="${2:?Usage: $0 <name> <discord_bot_token_var>}"
-
-# ── Paths ───────────────────────────────────────────────────
-SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
-COMPOSE_DIR="$(cd "$SCRIPT_DIR/.." && pwd)"
-COMPOSE_FILE="${COMPOSE_DIR}/compose.yml"
-
-# Each Hermes worker gets its own volume on the NFS HoardingCow
-VOLUME_BASE="/mnt/HoardingCow_docker_data/Hermes"
-VOLUME_DIR="${VOLUME_BASE}/${NAME}"
-
-# The Hermes container runs as UID 10000 (hermes user from Dockerfile)
-HERMES_UID=10000
-
-# ── Validation ──────────────────────────────────────────────
-if ! [ -f "$COMPOSE_FILE" ]; then
-  echo "❌ compose.yml not found at $COMPOSE_FILE"
-  exit 1
-fi
-
-if grep -q "^  ${NAME}:" "$COMPOSE_FILE"; then
-  echo "❌ Service '${NAME}' already exists in ${COMPOSE_FILE}"
-  exit 1
-fi
-
-# ── Generate unique API key ─────────────────────────────────
-# Used by Paperclip to authenticate against this worker's
-# Hermes API server (/v1/chat/completions)
-API_KEY="pc_worker_$(openssl rand -hex 16)"
-
-# ── Find next available API port ────────────────────────────
-# Workers get sequential ports starting at 8650.
-# Scans compose.yml for existing API_SERVER_PORT values and
-# picks the next one.
-BASE_PORT=8650
-MAX_PORT=0
-while IFS= read -r line; do
-  port="${line#*API_SERVER_PORT: \"}"
-  port="${port%%\"*}"
-  if [ -n "$port" ] && [ "$port" -gt "$MAX_PORT" ]; then
-    MAX_PORT="$port"
-  fi
-done < <(grep -oP 'API_SERVER_PORT:\s*"\d+"' "$COMPOSE_FILE" 2>/dev/null)
-
-NEW_PORT=$((MAX_PORT + 1))
-if [ "$NEW_PORT" -lt "$BASE_PORT" ]; then
-  NEW_PORT=$BASE_PORT
-fi
-
-# ── Create volume directory (on NFS) ────────────────────────
-echo "📁 Creating volume directory: ${VOLUME_DIR}"
-mkdir -p "$VOLUME_DIR"
-
-# Hermes container runs as UID 10000 — set ownership so the
-# container can write its config, sessions, skills
-if command -v chown &>/dev/null; then
-  chown -R "${HERMES_UID}:${HERMES_UID}" "$VOLUME_DIR" 2>/dev/null || \
-    echo "⚠ Could not chown ${VOLUME_DIR} — run with sudo if needed"
-fi
-
-# Make it group-readable for debugging
-chmod 755 "$VOLUME_DIR" 2>/dev/null || true
-
-# ── Append service to compose.yml ───────────────────────────
-echo "📝 Appending service '${NAME}' to compose.yml ..."
-
-TMPFILE=$(mktemp)
-
-awk -v name="$NAME" \
-    -v port="$NEW_PORT" \
-    -v api_key="$API_KEY" \
-    -v token_var="$TOKEN_VAR" \
-    '
-    # Insert new worker service block just before the networks: section
-    /^networks:/ {
-      print ""
-      print "  " name ":"
-      print "    <<: *hermes-worker"
-      print "    container_name: " name
-      print "    environment:"
-      print "      API_SERVER_PORT: \"" port "\""
-      print "      API_SERVER_KEY: \"" api_key "\""
-      print "      DISCORD_BOT_TOKEN: ${" token_var "}"
-      print "    volumes:"
-      print "      - /mnt/HoardingCow_docker_data/Hermes/" name ":/opt/data"
-      print ""
-    }
-    { print }
-' "$COMPOSE_FILE" > "$TMPFILE" && mv "$TMPFILE" "$COMPOSE_FILE"
-
-# ── Done ────────────────────────────────────────────────────
-echo ""
-echo "✅ Worker '${NAME}' provisioned successfully"
-echo ""
-echo "────────────────────────────────────────────"
-echo "  NEXT STEPS"
-echo "────────────────────────────────────────────"
-echo ""
-echo "1. Add secrets to the agenix .env stack file:"
-echo ""
-echo "   # ${NAME}"
-echo "   ${TOKEN_VAR}=<paste-discord-bot-token-here>"
-echo ""
-echo "2. Restart the AI stack:"
-echo ""
-echo "   systemctl restart ai_stack.service"
-echo ""
-echo "3. In Paperclip, create an agent with HTTP adapter:"
-echo ""
-echo "   Endpoint: http://${NAME}:${NEW_PORT}/v1/chat/completions"
-echo "   API Key:  ${API_KEY}"
-echo ""
-echo "────────────────────────────────────────────"
--- a/hermes/Dockerfile
+++ b/hermes/Dockerfile
@@ -0,0 +1,73 @@
+# 1. On récupère la version la plus récente d'UV
+FROM ghcr.io/astral-sh/uv:latest AS uv_source
+
+# 2. Image officielle Hermes Agent de NousResearch
+# Contient déjà: Python, Node.js, npm, Playwright/Chromium, venv, tts_tool.py, etc.
+FROM nousresearch/hermes-agent:latest
+
+# ---------- System dependencies ----------
+# The official hermes-agent image already has: git, curl, ffmpeg, python3,
+# gcc, build-essential, openssh-client, procps, tini, ripgrep, docker-cli,
+# libportaudio2, ca-certificates, etc.
+#
+# These extras we need to add back:
+#   - poppler-utils, imagemagick  (PDF/image processing)
+#   - texlive-*                   (LaTeX typesetting for reports)
+#   - qemu-user-static, binfmt-support (QEMU cross-compilation)
+#   - emacs-nox                   (text editing in container)
+USER root
+RUN apt-get update && \
+    apt-get install -y --no-install-recommends \
+        libportaudio2 \
+        ca-certificates \
+        poppler-utils \
+        imagemagick \
+        texlive-latex-base \
+        texlive-latex-extra \
+        texlive-fonts-recommended \
+        texlive-xetex \
+        texlive-science \
+        qemu-user-static \
+        binfmt-support \
+        emacs-nox && \
+    rm -rf /var/lib/apt/lists/*
+
+# ---------- UV (hyperfast pip alternative) ----------
+COPY --chmod=0755 --from=uv_source /uv /usr/local/bin/
+
+WORKDIR /opt/hermes
+
+# ---------- Piper TTS dans le venv existant ----------
+# Le venv de l'image de base est root-owned, on doit installer en root aussi
+RUN . /opt/hermes/.venv/bin/activate && \
+    uv pip install --no-cache-dir piper-tts sounddevice numpy
+
+# ---------- Télécharger la voix Piper Ryan (high quality) ----------
+RUN mkdir -p /opt/hermes/.venv/share/piper/voices && \
+    /opt/hermes/.venv/bin/python3 /dev/stdin << 'PYEOF'
+import urllib.request
+base = '/opt/hermes/.venv/share/piper/voices'
+url = 'https://huggingface.co/rhasspy/piper-voices/resolve/main/en/en_US/ryan/high/en_US-ryan-high.onnx'
+urllib.request.urlretrieve(url, base + '/en_US-ryan-high.onnx')
+urllib.request.urlretrieve(url + '.json', base + '/en_US-ryan-high.onnx.json')
+PYEOF
+
+# ---------- Patch tts_tool.py: remplacer Edge TTS par Piper ----------
+# Edge TTS appelle les serveurs Microsoft — on ne veut jamais ça.
+# Piper roule localement sur CPU, aucun cloud, aucune donnée qui sort.
+COPY patch_tts_tool.py /tmp/patch_tts_tool.py
+RUN /opt/hermes/.venv/bin/python3 /tmp/patch_tts_tool.py && rm /tmp/patch_tts_tool.py
+
+# ---------- Runtime ----------
+# Retour à l'utilisateur non-root pour la sécurité
+USER hermes
+
+ENV HERMES_HOME=/opt/data
+ENV PATH="/opt/data/.local/bin:${PATH}"
+
+VOLUME [ "/opt/data" ]
+
+# Script de réparation des permissions + patch TTS au démarrage
+COPY --chmod=0755 fix-permissions.sh /opt/hermes/fix-permissions.sh
+
+ENTRYPOINT [ "/usr/bin/tini", "-g", "--", "/opt/hermes/fix-permissions.sh" ]
--- a/hermes/compose.yml
+++ b/hermes/compose.yml
@@ -0,0 +1,38 @@
+version: "3.8"
+services:
+
+  hermes:
+    build: ./
+    container_name: hermes
+    restart: always
+    # Gateway run enables the internal API server on port 8642
+    command: gateway run
+    environment:
+      - OLLAMA_HOST=http://ollama:11434
+      - API_SERVER_ENABLED=true
+      - API_SERVER_PORT=8642
+      - API_SERVER_HOST=0.0.0.0
+      - API_SERVER_KEY=hermes_local_key
+      - GATEWAY_ALLOW_ALL_USERS=true
+      - OPENROUTER_API_KEY=${OPEN...KEY}
+      # ROCm for GPU-accelerated faster-whisper STT
+      - HSA_OVERRIDE_GFX_VERSION=9.0.6
+      - HCC_AMDGPU_TARGET=gfx906
+      - HIP_VISIBLE_DEVICES=0,1
+      - ROCR_VISIBLE_DEVICES=0,1
+      - HSA_ENABLE_SDMA=0
+    volumes:
+      - /mnt/HoardingCow_docker_data/Hermes/data:/opt/data
+    devices:
+      - /dev/kfd:/dev/kfd
+      - /dev/dri:/dev/dri
+    group_add:
+      - "303"
+      - "26"
+    networks:
+      - ai_backend
+
+networks:
+  ai_backend:
+    external: true
+    name: ai_backend
--- a/hermes/fix-permissions.sh
+++ b/hermes/fix-permissions.sh
@@ -0,0 +1,38 @@
+#!/bin/bash
+# Startup permission fix + TTS patch.
+# Runs as root before the entrypoint drops to the hermes user.
+set -e
+
+HERMES_HOME="${HERMES_HOME:-/opt/data}"
+
+# Fix ownership on critical writable directories
+chown -R hermes:hermes \
+  "$HERMES_HOME/sessions" \
+  "$HERMES_HOME/checkpoints" \
+  "$HERMES_HOME/skills" \
+  "$HERMES_HOME/memories" \
+  "$HERMES_HOME/workspace" \
+  "$HERMES_HOME/pastes" \
+  "$HERMES_HOME/logs" \
+  "$HERMES_HOME/cron" \
+  "$HERMES_HOME/plans" \
+  "$HERMES_HOME/hooks" \
+  "$HERMES_HOME/cache" \
+  2>/dev/null || true
+
+# Fix data volume root ownership
+if [ "$(stat -c %u "$HERMES_HOME" 2>/dev/null)" != "$(id -u hermes)" ]; then
+  chown hermes:hermes "$HERMES_HOME" 2>/dev/null || true
+fi
+
+# ---------- Patch tts_tool.py: replace Edge TTS with Piper ----------
+# Fallback runtime patch in case the volume's site-packages differ from the image.
+# Idempotent: if already patched, the script does nothing.
+PATCH_SCRIPT="/opt/hermes/patch_tts_tool.py"
+if [ -f "$PATCH_SCRIPT" ]; then
+  echo "Applying TTS patch (Piper only, no Edge fallback)..."
+  /opt/hermes/.venv/bin/python3 "$PATCH_SCRIPT" 2>&1 || true
+fi
+
+# Chain to the official Hermes entrypoint
+exec /opt/hermes/docker/entrypoint.sh "$@"
--- a/ai/hermes/patch_tts_tool.py
+++ b/ai/hermes/patch_tts_tool.py
--- a/ollama/Dockerfile
+++ b/ollama/Dockerfile
@@ -0,0 +1,61 @@
+# ollama/Dockerfile
+#
+# Custom ollama image with ROCm + gfx906 (MI50) support.
+# The default ollama/rocm image ships ROCm 7.2 which drops gfx906 support.
+# This builds ollama and its llama.cpp runner from source, targeting gfx906.
+#
+# Build:
+#   docker build -t ollama/ollama:rocm-gfx906 ./ollama
+
+FROM rocm/dev-ubuntu-22.04:6.1.2-complete AS builder
+
+RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y \
+    git golang-go cmake build-essential pkg-config \
+    && rm -rf /var/lib/apt/lists/*
+
+ARG OLLAMA_VERSION=v0.23.2
+RUN git clone --depth 1 --branch ${OLLAMA_VERSION} https://github.com/ollama/ollama.git /build
+WORKDIR /build
+
+ENV HIP_PATH=/opt/rocm
+ENV ROCM_PATH=/opt/rocm
+ENV PATH=/opt/rocm/bin:/opt/rocm/hip/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
+RUN cd llama.cpp && \
+    mkdir -p build && cd build && \
+    cmake .. \
+      -DLLAMA_HIPBLAS=ON \
+      -DCMAKE_C_COMPILER=clang \
+      -DCMAKE_CXX_COMPILER=clang++ \
+      -DAMDGPU_TARGETS=gfx906 \
+      -DCMAKE_BUILD_TYPE=Release \
+      -DLLAMA_NATIVE=OFF \
+      -DLLAMA_BUILD_TESTS=OFF \
+      -DLLAMA_BUILD_EXAMPLES=OFF \
+      -DLLAMA_BUILD_SERVER=OFF && \
+    cmake --build . --config Release -j$(nproc) && \
+    cmake --install . --prefix /build/dist
+
+ENV CGO_ENABLED=0
+RUN go build -trimpath -o dist/ollama .
+
+FROM ubuntu:22.04
+
+RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y \
+    ca-certificates curl libstdc++6 libgomp1 \
+    && 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/dist/ollama /usr/bin/ollama
+COPY --from=builder /build/dist/lib/ /usr/lib/ollama/
+
+RUN ldconfig
+
+ENV LD_LIBRARY_PATH=/opt/rocm/lib:/usr/lib/ollama
+ENV HSA_OVERRIDE_GFX_VERSION=9.0.6
+ENV HCC_AMDGPU_TARGET=gfx906
+ENV HSA_ENABLE_SDMA=0
+
+EXPOSE 11434
+ENTRYPOINT ["/bin/ollama"]
+CMD ["serve"]
--- a/ollama/compose.yml
+++ b/ollama/compose.yml
@@ -0,0 +1,41 @@
+version: "3.8"
+
+services:
+  ollama:
+    build:
+      context: ./
+      dockerfile: Dockerfile
+    image: ollama/ollama:rocm-gfx906
+    container_name: ollama
+    privileged: true
+    tty: true
+    restart: always
+    ports:
+      - "127.0.0.1:11434:11434"
+    networks:
+      - ai_backend
+    volumes:
+      - /mnt/HoardingCow_docker_data/Ollama/ollama:/root/.ollama
+    environment:
+      - OLLAMA_VULKAN=0
+      - OLLAMA_HOST=0.0.0.0
+      - OLLAMA_DEBUG=1
+      - OLLAMA_FLASH_ATTENTION=0
+      - OLLAMA_NUM_PARALLEL=2
+      # ROCm / gfx906 configuration
+      - 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
+    devices:
+      - /dev/kfd:/dev/kfd
+      - /dev/dri:/dev/dri
+    group_add:
+      - "303"
+      - "26"
+
+networks:
+  ai_backend:
+    external: true
+    name: ai_backend
--- a/versioncontrol/compose.yml
+++ b/versioncontrol/compose.yml
@@ -7,14 +7,8 @@ services:
      - USER_UID=1000
      - USER_GID=1000
      - GITEA__server__ROOT_URL=https://code.lazyworkhorse.net
-      - GITEA__actions__ENABLED=true
-      - GITEA__actions__DEFAULT_ACTIONS_URL=off
      - SSH_PORT=2222
      - SSH_LISTEN_PORT=2222
-      # Enable Gitea Actions (act_runner required on host)
-      - GITEA__actions__ENABLED=true
-      # Don't fetch actions from GitHub (offline mode + local only)
-      - GITEA__actions__DEFAULT_ACTIONS_URL=off
    volumes:
      - /mnt/HoardingCow_docker_data/Gitea:/data
    networks:
@@ -46,22 +40,6 @@ services:
      # Internal Routing
      - "traefik.http.services.gitea.loadbalancer.server.port=3000"

-  act_runner:
-    image: gitea/act_runner:latest
-    container_name: act_runner
-    environment:
-      - GITEA_INSTANCE_URL=https://code.lazyworkhorse.net
-      - GITEA_RUNNER_REGISTRATION_TOKEN=${GITEA_RUNNER_TOKEN}
-      - GITEA_RUNNER_NAME=ai-host-runner
-      - GITEA_RUNNER_LABELS=ubuntu-latest:docker://catthehacker/ubuntu:full-22.04,nixos-builder:docker://nixos/nix
-    volumes:
-      - /var/run/docker.sock:/var/run/docker.sock
-    networks:
-      - vc_net
-    restart: always
-    depends_on:
-      - gitea
-
 networks:
  vc_net:
    external: true
--- a/vpn/Dockerfile
+++ b/vpn/Dockerfile
@@ -1,9 +0,0 @@
-# Custom wg-easy with iptables-nft (nftables-backed iptables)
-# Fixes crash-loop when host kernel lacks legacy iptable_nat module.
-FROM ghcr.io/wg-easy/wg-easy:latest
-
-# The upstream image registers only iptables-legacy with update-alternatives.
-# iptables-nft binary exists but isn't registered as an alternative key.
-# Override the alternatives-managed symlinks directly.
-RUN ln -sf /usr/sbin/iptables-nft /usr/sbin/iptables && \
-    ln -sf /usr/sbin/ip6tables-nft /usr/sbin/ip6tables
--- a/vpn/compose.yml
+++ b/vpn/compose.yml
@@ -2,10 +2,7 @@ version: "3.8"

 services:
  wireguard:
-    build:
-      context: .
-      dockerfile: Dockerfile
-    image: wg-easy-iptables-nft:latest
+    image: weejewel/wg-easy:latest
    container_name: wireguard
    cap_add:
      - NET_ADMIN