Heidelberg AI News

Video · NeuralNine ·

Hermes Agent Crash Course: From Basics to Advanced

In short

This crash course explains how Hermes Agent works as an always-on, self-hosted AI agent that runs on a VPS, connects through multiple messaging channels, and retains context through memory and skills. It covers setup, everyday operation, and advanced extensibility so technical builders can deploy safer, more capable agents that improve through repeated use.

Key points

  • Hermes runs 24/7 on an isolated VPS and can connect through Telegram, WhatsApp, Discord, Slack, and voice interfaces.
  • The `.hermes` directory uses `SOUL.md`, `USER.md`, and `MEMORY.md` to define behavior and persistent context.
  • Skills are selectively loaded workflows that can be created, improved, installed from the Hermes hub, or managed by the skill curator.
  • The CLI includes commands such as `hermes`, `hermes setup model`, `hermes doctor`, `hermes update`, and `hermes gateway restart`.
  • Approval modes are `smart`, `manual`, and `off`, controlling when Hermes requires permission to run risky commands.
  • Tools, MCP servers, connectors, and plugins extend Hermes with services such as Notion, browser automation, memory systems, and image generation.
  • The course also introduces scheduled jobs, profiles, bot mode, agent delegation, gateway and shell hooks, and voice integration.

Written by an AI model from the video’s captions and description.

Watch on YouTube The opening of the article at youtube.com.

In this story

A tool or model opens in the Heidelberg AI Curriculum: what it does, what it costs, how to start. A repo opens on GitHub.

Related stories

All stories

VideoBijan Bowen

Liquid LFM3 Is INSANE – Tiny Local AI That ANYONE Can Run!

The video tests Liquid AI’s LFM3 family of small local models across text, agent workflows, vision, voice, video, and edge devices. The 1.7B models can run offline on typical computers and inexpensive hardware, making practical local AI more accessible despite modest accuracy and a proprietary license.

VideoNeuralNine

Claude Fable 5.2 is Absolutely Insane...

The video evaluates Claude Fable 5.2 against Opus 5.5 using pricing, public benchmarks, a concision benchmark, and coding experiments. It argues that Fable 5.2 delivers stronger, cleaner results on complex technical work and may be more cost-efficient per completed task despite higher token prices.

VideoNeuralNine

LLMs Can Now Directly Write Model Weights?!

The video examines weight space programming, where frontier LLMs analyze, modify, and generate neural-network weights without conventional training loops. Experiments suggest this can identify model behavior, fix specific parameter bugs, and outperform some traditionally trained small models, but current capabilities remain narrow and do not enable general knowledge injection.