Meet Hermes Agent

Hermes Agent is a self-improving AI agent built by Nous Research. Unlike a chatbot that forgets you when the tab closes, Hermes is designed to get more capable the longer it runs.

What makes it different

Chatbot	Hermes Agent
Answers questions	Takes actions (files, shell, web, delegation)
Forgets between sessions	Persistent memory across sessions
Static capabilities	Creates and improves skills from experience
Tied to one interface	CLI, Telegram, Discord, Slack, 20+ platforms
Cloud API required	Runs with local models via Ollama, vLLM, llama.cpp

The core idea: a learning loop

Hermes has a built-in loop that compounds over time:

1. Memory: stores facts, preferences, and context across sessions.
2. Skills: packages procedures it learns and reuses automatically.
3. Self-improvement: refines skills during use and nudges itself to persist knowledge.
4. User modeling: builds a deepening model of who you are and how you work.

The longer you use it, the less you repeat yourself.

What Hermes can do

With the right local model (one that supports tool calling), Hermes can:

• Edit files and run shell commands on your machine.
• Search and browse the web (when online; optional for offline setups).
• Delegate work to subagents running in parallel.
• Schedule recurring tasks via built-in cron.
• Talk to you from Telegram, Discord, or the CLI while working on a remote server.
• Connect to MCP servers for extended tool capabilities.

What this handbook focuses on

Official docs cover everything Hermes can do. This handbook focuses on one setup I actually use:

Hermes Agent + Ollama + open-source model, running fully offline on my laptop.

No API keys. No subscriptions. All data stays on my machine.

Next: Install & first run.