15.2 Installation & Migration (from OpenClaw)

☤ "From 'run it' to 'migrate from OpenClaw' — both in one command."

5 Deployment Scenarios

Mode A/B: One-Liner Install

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
# installs uv + Python 3.11 (isolated), Node.js, ripgrep, ffmpeg
hermes chat "你好，今天有什么任务？"
hermes update

VPS daemon: pm2 start "hermes gateway" --name hermes-gateway && pm2 save.

Mode C: Docker

services:
  hermes:
    image: hermes/agent:latest
    restart: unless-stopped
    volumes:
      - ~/.hermes:/home/hermes/.hermes
      - ~/workspace:/workspace
    environment:
      - HERMES_LLM=anthropic
      - ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY}

Mode D: hermes claw migrate (from OpenClaw)

$ hermes claw migrate --dry-run   # preview first
$ hermes claw migrate             # real migration
✓ config      4 KB copied
✓ memory      38 MB copied, FTS5 index rebuilt
✓ skills      28 skills ported (3 had incompatibilities)
✓ api-keys    2 keys installed to Keyring
✓ persona     persona ported

Compatibility boundary

Shared-config mode (keep both running): shared.openclaw.enabled: true, mode: read_only.

Mode E: Serverless (Modal / Daytona)

hermes deploy modal — the environment hibernates when idle and wakes on demand; long tasks keep running even when your laptop is off.

Onboarding Wizard

🦤 Welcome to Hermes Agent!
? LLM provider: Nous Portal / OpenAI / Anthropic / OpenRouter / Local
? Channel: Telegram / Discord / WhatsApp / Slack / Signal / CLI
? Persona / reply language / timezone
? Self-evolving: [✓] auto skill creation [✓] offline improvement [✓] cross-session blending

Enable cross-session persona blending — it turns on Honcho user modeling (see 15.6).

Section Summary

Next section: 15.3 Three-Layer Architecture