live1,247 agents deployedbuilt by a solo devpowered by hermes
← Blog
2026-05-20 · 6 min read

How to host your own Hermes AI agent (free)

Hermes is an open-source agentic AI that can browse the web, write code, remember facts, and run tasks autonomously. Hosting it yourself used to require Docker knowledge and a VPS. With Karp, you can have a running Hermes agent in under 60 seconds — no DevOps required.

What is Hermes?

Hermes is an agentic LLM runtime built by Nous Research. Unlike ChatGPT or Claude, Hermes is designed to run in a loop — it can call tools, browse the web, write and execute code, and self-improve by consolidating what it learns into long-term memory.

The key difference: Hermes is stateful. Every conversation adds to its memory. Over time it learns your preferences, your projects, your style.

Option 1: Use Karp (recommended)

Karp handles all the infrastructure. You get a managed Hermes agent with Telegram integration, web chat, and persistent memory — hosted on Karp with automatic scaling and health checks.

  1. Sign up at karpdo.com
  2. Choose a template (Personal Assistant, Developer Agent, or blank)
  3. Connect your LLM key (DeepSeek is cheapest at ~$0.001/1k tokens)
  4. Click Deploy — agent is live in under 60 seconds
  5. Chat via the web UI or scan the Telegram QR code

Free tier includes 5 demo messages without an API key. No credit card required.

Option 2: Self-hosted on a VPS

If you want full control, you can run Hermes yourself on any Linux server.

# Install Hermes
pip install hermes-agent

# Run with your DeepSeek key
OPENAI_API_KEY=sk-your-deepseek-key \
OPENAI_BASE_URL=https://api.deepseek.com \
hermes serve --port 8000

You'll also need to set up Telegram, persistent storage, and process management (e.g. systemd or PM2). The Karp platform handles all of this automatically.

Which LLM should I use?

Hermes works with any OpenAI-compatible API. Our recommendations:

  • DeepSeek V3 — best value, great reasoning, $0.001/1k tokens
  • OpenRouter + Llama 3.3 70B — free tier available, strong for coding
  • Claude 3.5 Sonnet — best instruction following, $3/M tokens
  • GPT-4o — strong multi-modal, good for vision tasks

Next steps

Once your agent is running, try connecting it to Telegram for mobile access. See our Telegram setup guide for step-by-step instructions.

Deploy your agent now — free

No credit card. No DevOps. 60-second setup.

Start for free →