docs: add AGENTS.md with dev workflow and YAML conventions

- Add x-hermes-worker YAML anchor template in compose.yml
  (CPU-only workers, no GPU passthrough, OpenCode Go provider)
- Add commented worker example with env vars placeholder
- Create scripts/provision-hermes-worker.sh for automated worker
  provisioning (generates port, API key, volume dir, appends service)
- Workers connect to Discord only, isolated per container
- Volumes under /mnt/HoardingCow_docker_data/Hermes/<name>/

.gitea/workflows/build-hermes.yml

@@ -0,0 +1,31 @@
+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

.gitea/workflows/build-ollama.yml

@@ -0,0 +1,31 @@
+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

AGENTS.md

@@ -0,0 +1,31 @@
+# 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

ai/Dockerfile

@@ -1,66 +0,0 @@
-# 1. On récupère la version la plus récente d'UV
-FROM ghcr.io/astral-sh/uv:latest AS uv_source
-
-# 2. Image de base stable
-FROM debian:stable-slim
-
-# Disable Python stdout buffering to ensure logs are printed immediately
-ENV PYTHONUNBUFFERED=1
-
-# Install system dependencies in one layer, clear APT cache
-# tini reaps orphaned zombie processes (MCP stdio subprocesses, git, bun, etc.)
-RUN apt-get update && \
-    apt-get install -y --no-install-recommends \
-        build-essential python3 ripgrep ffmpeg gcc python3-dev libffi-dev procps git openssh-client docker-cli tini \
-        curl poppler-utils imagemagick \
-        texlive-latex-base texlive-latex-extra texlive-fonts-recommended texlive-xetex texlive-science \
-        qemu-user-static binfmt-support qemu-user-binfmt \
-        emacs-nox \
-        libportaudio2 && \
-    rm -rf /var/lib/apt/lists/*
-
-# Création de l'utilisateur 'hermes' directement avec les bons accès
-RUN useradd -u 10000 -m -d /opt/data hermes
-
-# Copie d'uv (dernière version)
-COPY --chmod=0755 --from=uv_source /uv /usr/local/bin/
-
-WORKDIR /opt/hermes
-
-# On donne la propriété du dossier de travail à l'utilisateur hermes
-RUN chown hermes:hermes /opt/hermes
-
-# ---------- Hermes venv ----------
-# Passer immédiatement sous l'utilisateur hermes pour tout le reste du build
-USER hermes
-
-# ---------- Source code ----------
-# On copie tout le projet d'un coup sans assumer la présence de fichiers de lock spécifiques
-COPY --chown=hermes:hermes . .
-
-# ---------- Python virtualenv avec Piper TTS ----------
-RUN uv venv && \
-    uv pip install --no-cache-dir piper-tts sounddevice numpy faster-whisper
-
-# ---------- Télécharger la voix Piper Ryan ----------
-RUN mkdir -p /opt/hermes/.venv/share/piper/voices && \
-    /opt/hermes/.venv/bin/python3 -c "
-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 + '?download=true', base + '/en_US-ryan-high.onnx')
-urllib.request.urlretrieve(url + '.onnx.json?download=true', base + '/en_US-ryan-high.onnx.json')
-"
-
-# ---------- Patch tts_tool.py: replace Edge TTS fallback with Piper ----------
-COPY ai/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 ----------
-ENV HERMES_HOME=/opt/data
-ENV PATH="/opt/data/.local/bin:${PATH}"
-
-VOLUME [ "/opt/data" ]
-
-# Le conteneur tourne de manière ultra-sécurisée sous l'utilisateur hermes dès le départ
-ENTRYPOINT [ "/usr/bin/tini", "-g", "--", "/opt/hermes/docker/entrypoint.sh" ]

ai/compose.yml

@@ -1,4 +1,35 @@
 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:
@@ -26,8 +57,14 @@ services:
   #     - "traefik.http.routers.webui-https.tls.certresolver=njalla"
 
   hermes:
-    build: ./
+    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
@@ -45,6 +82,7 @@ services:
       - 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:
@@ -57,9 +95,11 @@ services:
       - ai_backend
 
   ollama:
-    image: ollama/ollama:latest
+    build:
+      context: ./ollama
+      dockerfile: Dockerfile
+    image: ollama/ollama:rocm-gfx906
     container_name: ollama
-    privileged: true
     tty: true
     restart: always 
     ports:
@@ -77,7 +117,7 @@ services:
       - HSA_ENABLE_SDMA=0 
       - OLLAMA_HOST=0.0.0.0
       - OLLAMA_DEBUG=1
-      - OLLAMA_FLASH_ATTENTION=0
+      - OLLAMA_FLASH_ATTENTION=1
       - OLLAMA_NUM_PARALLEL=2
     devices:
       # Map the render nodes and KFD for ROCm to work inside the container
@@ -87,6 +127,24 @@ services:
       - "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

ai/hermes/Dockerfile

@@ -0,0 +1,93 @@
+# 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" ]

ai/hermes/himalaya-ro.sh

@@ -0,0 +1,73 @@
+#!/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" "$@"

ai/hermes/patch_tts_tool.py

@@ -0,0 +1,181 @@
+#!/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/ollama/Dockerfile

@@ -0,0 +1,106 @@
+# 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"]

ai/patch_tts_tool.py

@@ -1,96 +0,0 @@
-#!/usr/bin/env python3
-"""Patch Hermes TTS tool: remove Edge TTS, replace with Piper as default/fallback."""
-import sys
-
-tts_path = '/opt/hermes/.venv/lib/python3.13/site-packages/tools/tts_tool.py'
-
-with open(tts_path) as f:
-    code = f.read()
-
-# Replace the Edge fallback with Piper fallback
-old_edge = '''        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_piper = '''        else:
-            # Default: Piper TTS (local, CPU, no cloud, no Microsoft)
-            piper_available = False
-            try:
-                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):
-                    piper_available = True
-            except Exception:
-                pass
-
-            if piper_available:
-                logger.info("Generating speech with Piper TTS (local, CPU)...")
-                import subprocess
-                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")
-                cmd = [piper_binary, "--model", model_path, "--output-raw"]
-                proc = subprocess.Popen(cmd, stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.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]
-                subprocess.run(ffmpeg_cmd, input=raw_audio, capture_output=True, timeout=30)
-                logger.info("Piper TTS audio saved: %s", file_str)
-            else:
-                return json.dumps({
-                    "success": False,
-                    "error": "No TTS provider available. Install Piper TTS (pip install piper-tts) "
-                             "and download a voice model."
-                }, ensure_ascii=False)'''
-
-if old_edge in code:
-    code = code.replace(old_edge, new_piper)
-    print("Edge fallback replaced with Piper")
-else:
-    if 'Default: Piper TTS' in code:
-        print("Piper fallback already present")
-    else:
-        print("ERROR: Could not find Edge fallback in tts_tool.py")
-        # Debug output
-        import re
-        for m in re.finditer(r'        else:\n            # Default:', code):
-            start = max(0, m.start() - 100)
-            end = min(len(code), m.end() + 200)
-            print(f"Found else/default at position {m.start()}:")
-            print(code[start:end])
-        sys.exit(1)
-
-with open(tts_path, 'w') as f:
-    f.write(code)
-print("tts_tool.py patched successfully")

ai/scripts/provision-hermes-worker.sh

@@ -0,0 +1,135 @@
+#!/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 "────────────────────────────────────────────"

versioncontrol/compose.yml

@@ -7,8 +7,14 @@ 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:
@@ -40,6 +46,22 @@ 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

vpn/Dockerfile

@@ -0,0 +1,9 @@
+# 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

vpn/compose.yml

@@ -2,7 +2,10 @@ version: "3.8"
 
 services:
   wireguard:
-    image: weejewel/wg-easy:latest
+    build:
+      context: .
+      dockerfile: Dockerfile
+    image: wg-easy-iptables-nft:latest
     container_name: wireguard
     cap_add:
       - NET_ADMIN