hermes-on-digitalocean

name	hermes-on-digitalocean
description	Step-by-step guide for deploying a Hermes Agent (by Nous Research) to a DigitalOcean droplet so it runs 24/7 with messaging-platform access (Telegram, Discord, Slack, etc.). Covers provisioning via doctl, Python 3.11 bootstrap on Ubuntu 24.04, non-interactive Hermes install, profile setup, credential migration from a laptop, OAuth caveats, and gateway-as-systemd-service. Load this skill when a user wants to move a Hermes agent off their laptop onto a cheap always-on cloud host.
version	1.0.0
author	cobibean
license	MIT

Deploy a Hermes Agent to a DigitalOcean droplet

Hermes Agent is a self-improving CLI/messaging AI agent by Nous Research. It belongs on a small always-on box — not your laptop. This skill walks through provisioning a $12/mo droplet, installing Hermes non-interactively, migrating credentials from a laptop install, and running the messaging gateway as a systemd service that survives logout.

Prereqs on your laptop:

• A working local Hermes install (hermes --version)
• A configured profile with the messaging platform you want (hermes -p <name> gateway setup)
• doctl installed and authenticated to a DigitalOcean account
• gh (only if your workspace repo is private and the droplet needs to clone it)

Pick specs

• s-1vcpu-2gb ($12/mo): the right floor. 1GB works only if you disable the browser toolset and never run anything heavy.
• Ubuntu 24.04 LTS: stable, but ships Python 3.12. Hermes needs 3.11 — uv bootstraps it cleanly (see below).
• Region: pick the one nearest your messaging users (Telegram users hit DO's Frankfurt or NYC POPs fine).

1. Provision the droplet

# Generate a project-scoped SSH key
ssh-keygen -t ed25519 -f ~/.ssh/hermes_droplet_ed25519 -N "" -C "hermes-droplet"

# Upload to DO
KEY_ID=$(doctl compute ssh-key import hermes-droplet-key \
  --public-key-file ~/.ssh/hermes_droplet_ed25519.pub \
  --format ID --no-header)

# Create droplet
doctl compute droplet create hermes \
  --image ubuntu-24-04-x64 \
  --size s-1vcpu-2gb \
  --region nyc3 \
  --ssh-keys $KEY_ID \
  --tag-names hermes-agent \
  --enable-monitoring \
  --wait \
  --format ID,Name,PublicIPv4,Status --no-header

Add an SSH alias to ~/.ssh/config:

Host hermes
  HostName <droplet-ip>
  User root
  IdentityFile ~/.ssh/hermes_droplet_ed25519
  IdentitiesOnly yes
  StrictHostKeyChecking accept-new

2. Wait for cloud-init, install prereqs, install Hermes

Ubuntu 24.04 holds the apt lock for ~60s on first boot (cloud-init). Wait it out:

ssh hermes '
  cloud-init status --wait
  export DEBIAN_FRONTEND=noninteractive
  apt-get update -qq
  apt-get install -y -qq curl git build-essential ripgrep ffmpeg ca-certificates

  # uv bootstraps Python 3.11 without needing the deadsnakes PPA
  curl -LsSf https://astral.sh/uv/install.sh | sh
  export PATH="/root/.local/bin:$PATH"
  uv python install 3.11

  # Hermes installer (--skip-setup because there is no TTY)
  curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh \
    | bash -s -- --skip-setup
'

The installer pulls Playwright Chromium (~165MB) regardless. On 2GB you can keep it; on 1GB disable the browser toolset after install:

ssh hermes 'hermes tools disable browser'

3. Create the profile + (optionally) wire a workspace repo

If you're running multiple agents or want the agent's SOUL/AGENTS files version-controlled, use a profile + symlinks pattern:

ssh hermes '
  hermes profile create my-agent
  # If you have a workspace repo for the agent (recommended):
  mkdir -p ~/DEV && cd ~/DEV
  git clone [e-mail]:<you>/<repo>.git workspace
  rm -f ~/.hermes/profiles/my-agent/SOUL.md
  ln -s ~/DEV/workspace/SOUL.md ~/.hermes/profiles/my-agent/SOUL.md
  ln -s ~/DEV/workspace/AGENTS.md ~/.hermes/profiles/my-agent/AGENTS.md
'

For a private repo, generate a deploy key on the droplet and add it to the repo:

ssh hermes '
  ssh-keygen -t ed25519 -f ~/.ssh/github_deploy -N "" -C "hermes-deploy"
  cat ~/.ssh/github_deploy.pub
  printf "Host github.com\n  IdentityFile ~/.ssh/github_deploy\n  IdentitiesOnly yes\n  StrictHostKeyChecking accept-new\n" > ~/.ssh/config
  chmod 600 ~/.ssh/config
'
# Then add the printed pubkey as a deploy key (write access if the agent should commit back):
gh repo deploy-key add /dev/stdin --repo <you>/<repo> --title "hermes-droplet" --allow-write <<< "<pubkey>"

4. Migrate credentials from your laptop

Hermes splits credentials across two files. Both must move:

# Per-profile config + secrets (Telegram bot token, etc.)
scp ~/.hermes/profiles/<profile>/config.yaml hermes:/root/.hermes/profiles/<profile>/config.yaml
scp ~/.hermes/profiles/<profile>/.env        hermes:/root/.hermes/profiles/<profile>/.env

# Account-wide OAuth credentials (Nous Portal, Codex, Qwen) — outside the profile
scp -p ~/.hermes/auth.json                    hermes:/root/.hermes/auth.json

ssh hermes 'chmod 600 ~/.hermes/profiles/<profile>/.env ~/.hermes/auth.json'

Then fix any laptop-specific paths in the profile's config:

ssh hermes 'hermes -p <profile> config set messaging.working_directory /root/DEV/workspace'

OAuth provider caveat

Some providers (notably OpenAI Codex) use PKCE OAuth with a localhost callback, not device code. On a headless droplet, a fresh hermes auth add openai-codex --type oauth will hang waiting for a callback that can never reach you. Two workarounds:

• Copy the laptop's auth.json (above) — refresh tokens are usually portable.
• SSH port-forward the callback port (Codex defaults to ~1455):

  ssh -L 1455:localhost:1455 hermes 'hermes -p <profile> auth add openai-codex --type oauth --timeout 600'
  

  Open the printed URL on your laptop; the OAuth redirect to localhost:1455 tunnels back to the droplet.

API-key providers (Anthropic, OpenRouter, Gemini, etc.) just need their key in .env — no OAuth dance.

5. Install gateway as a systemd service

ssh hermes 'hermes -p <profile> gateway install && hermes -p <profile> gateway start'

The service is profile-namespaced — name is hermes-gateway-<profile>.service, not hermes-gateway.service. gateway install also enables systemd linger automatically (loginctl enable-linger root), so the service survives SSH logout.

Verify:

ssh hermes 'systemctl --user is-active hermes-gateway-<profile>'
ssh hermes 'journalctl --user -u hermes-gateway-<profile> --no-pager -n 30'

Then message your bot. Round-trip = success.

6. Scrub the laptop (if applicable)

If you don't want the laptop ever competing with the droplet for messages — only one process can hold a given Telegram bot token at a time — strip the platform credentials from the laptop's profile:

sed -i.bak '/^TELEGRAM_BOT_TOKEN=/d; /^TELEGRAM_ALLOWED_USERS=/d; /^TELEGRAM_HOME_CHANNEL=/d' \
  ~/.hermes/profiles/<profile>/.env

Keep the .bak (gitignored) for rollback.

Day-2 ops cheatsheet

Task	Command
Tail gateway logs	ssh hermes 'journalctl --user -u hermes-gateway-<profile> -f'
Restart after config change	ssh hermes 'systemctl --user restart hermes-gateway-<profile>'
Update Hermes	ssh hermes 'hermes update && systemctl --user restart hermes-gateway-<profile>'
Push workspace changes	git push (laptop) → ssh hermes 'cd ~/DEV/workspace && git pull'
Cost check	doctl compute droplet list --format Name,Size,PriceMonthly --no-header

Gotchas (real ones, hit during a real deploy)

• doctl auth init requires a TTY — fails in non-interactive shells with unknown terminal. Write ~/Library/Application Support/doctl/config.yaml directly to bypass.
• Ubuntu 24.04 has no python3.11 package. Use uv python install 3.11 — don't waste time on the deadsnakes PPA.
• Cloud-init holds apt for ~60s post-boot. Always cloud-init status --wait before apt-get.
• The setup wizard always needs a TTY — pass --skip-setup to the installer for unattended installs.
• hermes login is removed — replaced by hermes auth add <provider>.
• OAuth callback hang — see the Codex note above. Symptom: hermes auth add ... --no-browser prints nothing and hangs.
• pkill over SSH can drop your connection (exit 255). The kill usually succeeds anyway; reconnect and verify with ps.
• Service name is profile-namespaced. systemctl --user status hermes-gateway returns "Unit not found" — use hermes-gateway-<profile>.service.

Why this is worth doing

A Hermes agent on a $12/mo droplet costs ~40¢/day for hosting. It runs 24/7, accepts messages from Telegram/Discord/Slack while your laptop is closed, runs cron jobs unattended, and accumulates persistent memory across sessions. The same setup on serverless would idle near zero but adds cold-start latency to every message. For an always-responsive assistant, a small always-on droplet is the better default.

References

• Hermes docs: https://hermes-agent.nousresearch.com/docs/
• Hermes repo: https://github.com/NousResearch/hermes-agent
• DO doctl docs: https://docs.digitalocean.com/reference/doctl/
• Skills standard: https://agentskills.io