What is Hermes

#

Hermes is a personal AI agent with one agent core shared across many surfaces:

• CLI — interactive terminal assistant
• Messaging gateway — Telegram, Discord, Slack, and many other platforms
• TUI — Ink terminal UI (hermes --tui)
• Desktop — Electron app talking to the same backend over JSON-RPC

It learns across sessions (memory + skills), can delegate to subagents, run scheduled jobs, and drive a real terminal and browser. New capability should arrive primarily as plugins and skills, not by growing the core forever.

Mental model

#

  CLI / Gateway / TUI / Desktop / Cron / ACP
                    │
                    ▼
              AIAgent (one core)
                    │
        ┌───────────┼───────────┐
        ▼           ▼           ▼
     Prompt      Providers    Tools
     + cache                  + backends

Everything above the core is an entry point. Platform differences belong there — not inside AIAgent.

Two ideas you will see everywhere

#

1. Prompt caching is sacred. A long conversation reuses a cached system-prompt prefix every turn. Mutating past context, swapping toolsets, or rebuilding the system prompt mid-conversation multiplies cost. Compression is the rare, sanctioned exception.
2. Narrow waist, rich edges. Every core model tool is sent on every API call. Prefer skills, CLI commands, plugins, or MCP before adding a new core tool.

You will study both in depth under Architecture.

What you will build toward

#

By the end of this series you should be able to:

• Run Hermes and verify a clean chat
• Explain the architecture in one diagram
• Find code by topic (stem_*.py siblings, not god-file reading)
• Choose the right extension rung (skill → plugin → tool)
• Complete the Labs without touching core source

Next

#

Install and first chat

Further reading

#

• Repository root AGENTS.md — project invariants
• Official docs: website/docs/getting-started/