What is Hermes?

Hermes (by Nous Research) is an open-source AI agent that grows with you. Unlike a plain chatbot, it keeps persistent memory, builds reusable skills, and runs as a long-lived agent you can talk to from a terminal, a gateway, or a native desktop app.

Three things make it different:

• Autonomous — it can plan and carry out multi-step tasks with tools.
• Persistent memory — it remembers across sessions, not just within one chat.
• Self-improving skills — it can write down a procedure once and reuse it.

What lives in ~/.hermes

Hermes keeps its “brain” as plain files on your machine — which is exactly what makes it teachable. Open them and look:

Why it matters: the agent’s behaviour isn’t hidden in a cloud account — it’s files you can read, edit, and version. That’s the whole point of Track 1.

Beyond SOUL.md, a lot of what you want an agent to know is project-specific — that’s what AGENTS.md is for (project architecture, conventions, key paths). It’s a whole concept of its own → AGENTS.md.

Try it: open the Hermes dashboard at localhost:9119 (it’s running in the kit — log in trainee / hermes), tour the Skills, Profiles, Memory and Tools screens, find those files on disk in ./hermes-home/, then drop an AGENTS.md into the latex-starter/ project and watch the agent pick up its conventions.

Next: Skills — the reusable procedures that show up across every agent tool, including Claude Code.