hermes-agent-packer/scripts/cloud-init.yaml

#cloud-config
# ------------------------------------------------------------------
# Deploy-time cloud-init for Vultr Marketplace
#
# The snapshot already has Hermes Agent, Docker, Caddy, ttyd, code-server
# installed.  This script:
#   1. Fetches per-instance metadata from the Vultr metadata service
#   2. Configures Hermes Agent with Vultr Inference provider
#   3. Configures code-server with password
#   4. Starts ttyd and code-server services
#   5. Configures Caddy as a TLS reverse proxy
#      - /              → ttyd (Hermes terminal)
#      - /code/*        → code-server
# ------------------------------------------------------------------

runcmd:
  # ── Fetch Vultr metadata and persist ──
  - |
    APP_PASSWORD=$(curl -sf -H "METADATA-TOKEN: vultr" http://169.254.169.254/v1/internal/app-password)
    APP_INF_API_KEY=$(curl -sf -H "METADATA-TOKEN: vultr" http://169.254.169.254/v1/internal/app-inf_api_key)
    APP_DOMAIN=$(curl -sf -H "METADATA-TOKEN: vultr" http://169.254.169.254/v1/internal/app-domain)
    printf 'APP_PASSWORD="%s"\nAPP_INF_API_KEY="%s"\nAPP_DOMAIN="%s"\n' \
      "$APP_PASSWORD" "$APP_INF_API_KEY" "$APP_DOMAIN" \
      > /etc/hermes-deploy.env
    chmod 600 /etc/hermes-deploy.env

  # ── Configure Hermes Agent with Vultr Inference provider ──
  - |
    . /etc/hermes-deploy.env

    # Create .env with API keys
    cat > /home/hermes/.hermes/.env << ENV_EOF
# =============================================================================
# VULTR INFERENCE PROVIDER (Pre-configured)
# =============================================================================
VULTR_API_KEY=${APP_INF_API_KEY}
VULTR_BASE_URL=https://api.vultrinference.com/v1

# For OpenAI-compatible custom endpoint mode:
OPENAI_API_KEY=${APP_INF_API_KEY}
OPENAI_BASE_URL=https://api.vultrinference.com/v1
ENV_EOF
    chown hermes:hermes /home/hermes/.hermes/.env
    chmod 600 /home/hermes/.hermes/.env

    # Create config.yaml
    cat > /home/hermes/.hermes/config.yaml << 'CONFIG_EOF'
# Hermes Agent Configuration - Vultr Inference
# =============================================================================
# Pre-configured for Vultr Inference API (OpenAI-compatible)

model:
  # Default model (GLM-5-FP8 - 200K context, reasoning-capable)
  default: "zai-org/GLM-5-FP8"
  # Use custom provider for Vultr Inference (OpenAI-compatible)
  provider: "custom"
  base_url: "https://api.vultrinference.com/v1"
  # api_key will be substituted from .env

# Terminal backend (local execution)
terminal:
  backend: "local"
  cwd: "."
  timeout: 180
  lifetime_seconds: 300

# Enable persistent memory
memory:
  memory_enabled: true
  user_profile_enabled: true
  memory_char_limit: 2200
  user_char_limit: 1375
  nudge_interval: 10

# Context compression for long conversations
compression:
  enabled: true
  threshold: 0.50
  target_ratio: 0.20
  protect_last_n: 20
  summary_model: "zai-org/GLM-5-FP8"

# Agent settings
agent:
  max_turns: 60
  verbose: false
  reasoning_effort: "medium"

# Toolsets for CLI
platform_toolsets:
  cli: [hermes-cli]

# Display settings
display:
  compact: false
  tool_progress: all
  streaming: true
CONFIG_EOF
    chown hermes:hermes /home/hermes/.hermes/config.yaml
    chmod 600 /home/hermes/.hermes/config.yaml

  # ── Configure and start code-server ──
  - |
    . /etc/hermes-deploy.env
    cat > /home/hermes/.config/code-server/config.yaml << CSCFG
bind-addr: 127.0.0.1:8080
auth: password
password: ${APP_PASSWORD}
cert: false
CSCFG
    chown hermes:hermes /home/hermes/.config/code-server/config.yaml
    chmod 600 /home/hermes/.config/code-server/config.yaml

  # ── Start ttyd and code-server services ──
  - systemctl enable ttyd-hermes code-server-hermes
  - systemctl start ttyd-hermes code-server-hermes

  # ── Write Caddyfile ──
  # ttyd at root, code-server at /code/
  # ZeroSSL primary (avoids LE rate limits), Let's Encrypt fallback
  - |
    . /etc/hermes-deploy.env
    HASHED=$(caddy hash-password --plaintext "$APP_PASSWORD")
    cat > /etc/caddy/Caddyfile << CADDY
${APP_DOMAIN} {
    tls {
      issuer acme {
        dir https://acme.zerossl.com/v2/DV90
        eab ur6CkPWFhPtp8af1SASQEA dRKaIDk9u4tn7MYrHNczzl_tnKSJ8Hipb2IFVui3oxEHCj_3wkGBA-zG93X7NXaROh-pzDJOugrtn1nxxGasxw
      }
      issuer acme {
        dir https://acme-v02.api.letsencrypt.org/directory
      }
    }

    basic_auth {
        hermes ${HASHED}
    }

    # code-server at /code/
    handle_path /code/* {
        reverse_proxy 127.0.0.1:8080
    }

    # Default: Hermes terminal (ttyd) at root
    handle {
        reverse_proxy 127.0.0.1:7681
    }
}
CADDY

  # ── Start Caddy and wait for cert issuance ──
  # ZeroSSL can take 30-60s; wait up to 90s for cert before proceeding
  - systemctl enable caddy
  - systemctl restart caddy
  - |
    echo "Waiting for TLS certificate issuance (up to 90s)..."
    for i in $(seq 1 90); do
      if find /var/lib/caddy -name '*.crt' 2>/dev/null | grep -q .; then
        echo "Certificate obtained!"
        break
      fi
      sleep 1
    done

  # ── Clean up secrets from disk ──
  - rm -f /etc/hermes-deploy.env