Hermes Agents in 30 Minutes: The OpenClaw Killer

What are Hermes Agents?

Hermes Agents are AI-powered messaging assistants that run continuously on a server, giving you persistent, always-on automation through the communication channels your team already uses: WhatsApp, Telegram, and Slack.

Unlike OpenClaw and similar tools, Hermes is designed from the ground up for persistence and delegation. It doesn't spin up on demand and shut down. It lives on your server, remembers context across conversations, and can hand tasks off to specialised sub-agents automatically.

The result: a real AI teammate that works in the background, responds to messages, checks your tools, and updates information without anyone needing to prompt it manually.

Why Hermes over OpenClaw?

OpenClaw works well for one-shot tasks, but it falls short the moment you need continuity. Hermes solves three things OpenClaw doesn't:

1. Persistent memory: Hermes remembers what happened in previous conversations and carries that context forward. OpenClaw starts fresh every time.
2. Always-on availability: Hermes runs as a server process. It's reachable through WhatsApp or Slack at any hour without you needing to trigger it.
3. Sub-agent delegation: Complex tasks get broken down and handed to specialised sub-agents. Hermes orchestrates the whole workflow, not just a single response.

How the architecture works

Hermes Agents are built around three core components that work together:

Memory storage

Hermes maintains a persistent memory layer that stores facts, preferences, past interactions, and task history. Every conversation builds on what came before. This is what makes it feel like a real assistant rather than a stateless chatbot.

Skills system

Skills are pre-built capabilities you attach to the agent: checking a Trello board, reading a Slack channel, querying a database, sending a report. You define the skills once and the agent calls them automatically based on what it's asked to do.

Sub-agents for delegation

When a task is too complex for a single response, Hermes breaks it into subtasks and assigns each one to a specialised sub-agent. The main agent orchestrates the results and delivers a unified output. This mirrors how a real team operates: a coordinator delegating to specialists.

Get started in 30 minutes

Step 1: Choose your hosting

Hermes needs to run continuously, so you need a server. Two options:

• VPS (recommended): Services like DigitalOcean, Hetzner, or Contabo give you a lightweight Linux server for a few dollars a month. Always on, low maintenance, easy to scale.
• Mac Mini: Works well if you already have one sitting idle. The downside is that it depends on your home network staying up and your machine staying awake.

For most teams, a VPS is the right call. A basic 2GB RAM instance is plenty to start.

Step 2: Install Hermes on your server

SSH into your VPS and clone the Hermes repository, then install dependencies:

git clone https://github.com/hermes-agents/hermes
cd hermes
npm install

Step 3: Connect your communication channel

Add your channel credentials to the config file. Hermes supports WhatsApp (via Twilio or WhatsApp Business API), Telegram (via Bot API), and Slack (via Bolt). Pick the one your team uses and paste in the token.

# .env
TELEGRAM_BOT_TOKEN=your_token_here
OPENAI_API_KEY=your_key_here

Step 4: Add your first skill

Define a skill by describing what it does and what tool it calls. The Trello skill from the session looks like this:

// skills/trello.js
export const trelloSkill = {
  name: "check_trello",
  description: "Fetches cards from a Trello board",
  run: async () => { /* Trello API call */ }
};

Step 5: Start the agent

Run Hermes as a background process so it stays alive after you close your SSH session:

pm2 start npm --name hermes -- start
pm2 save

Your agent is now live. Message it on WhatsApp or Telegram and it will respond.

Cutting costs with open-source models

One of Aki's key recommendations: you don't need to run every task through a premium model. For high-volume, lower-complexity tasks, Chinese open-source language models available through OpenCode offer strong performance at a fraction of the cost.

The pattern that works well: route complex reasoning tasks to a capable model like Claude, and handle simpler retrieval and formatting tasks with a cost-efficient open-source alternative. Hermes supports multi-model routing out of the box.

What you can automate with Hermes

• Daily standup summaries pulled from Trello or Linear and sent to Slack
• Customer support triage via WhatsApp, with escalation to a human when needed
• Automated follow-up messages triggered by CRM events
• Internal Q and A bot that searches your Notion or Confluence docs
• Meeting prep briefs sent to your Telegram 30 minutes before each call