Practical AI Agents with Hermes: what we learned in 2 days

Published:

Practical AI Agents with Hermes was a two-day workshop at Mahidol University (19–20 Nov 2026). This page summarises the workshop labs and the hands-on setup done afterwards on a real Mac running Hermes Agent v0.23.2.

What it covers

  • The big idea: Claude Code builds (SOULs, skills, scripts, Office files); Hermes runs, keeping agents working on a schedule and messaging you in Telegram.
  • What an agent is made of: model provider, SOUL, memory, skills, MCP tools, gateway, cron jobs and profiles.
  • Seven labs: install and identity, photo → Excel, Use Case Canvas, prompt → skill, parallel meeting summaries, a daily website check, deploy + monitor, and a mini project.
  • Rules used every day: a preview and a typed confirmation (ยืนยัน) before any write, send or publish; ?? instead of guessing; outside text is data, not instructions; no secrets in chat.
  • 18 bot designs grouped by job, from a secretary and IT support to research, teaching, publishing and second-brain bots.
  • Using several AI accounts wisely: a front-desk bot that does most tasks on Codex and hands only Claude-worthy work to Claude Code.
  • An honest comparison of Hermes and Claude Code.

The page is one self-contained HTML file with inline SVG diagrams and light/dark themes.