Hermes Atlas
Guides, templates & lists

Learn Hermes Agent

Ericcc-Ma/learn-hermes-agent

Lesson series for building a self-evolving agent harness modeled on Hermes Agent's learning system

In short

Learn Hermes Agent is a Python tutorial that rebuilds the self-learning ideas behind Hermes Agent step by step. Its lessons cover memory, skill creation and upkeep, context compression, gateway, cron, profiles, delegation and MCP. The README is mainly in Chinese, with an English version linked.

What Learn Hermes Agent does

Learn Hermes Agent teaches how to build a harness that improves from use rather than starting from zero each conversation. The README lists progressive lessons from s01 to s24, each adding one mechanism with a one-line motto: a loop with a memory file, background memory and skill review, a memory system with FTS5 search, a skill lifecycle of active, stale and archived, rule-based archiving after 30 and 90 days, and LLM-assisted merging of skills. The repository description says 12 lessons, so counts differ between the two.

Later lessons add context compression, token and cost tracking, error recovery, cron scheduling, a multi-platform gateway, profiles, delegate_task, MCP, a permission pipeline, hooks, git worktree isolation, planning, a kanban dispatcher and prompt assembly. A quick demo runs python s12_comprehensive/comprehensive.py, and you can try typing a correction and then running /insights and /curator. The description notes support for Anthropic, DeepSeek, Qwen and GLM models. It is an independent teaching project that reimplements ideas, not Hermes Agent itself.

Key features

  • Progressive lessons, each adding one self-evolution mechanism
  • Background memory and skill review that turn corrections into saved knowledge
  • Skill lifecycle with active, stale and archived states and a curator
  • Lessons on gateway, cron, profiles, delegation, MCP and permissions
  • Runnable demos such as s12_comprehensive and s18_full_hermes
  • Works with Anthropic, DeepSeek, Qwen and GLM models according to the repository description

When to use it

  • Learning how a self-improving agent stores memory and skills
  • Following a code-first walkthrough of Hermes-style features in Python
  • Prototyping your own agent harness with memory review and a skill library

Who it is for: Developers who read Chinese or English and want to understand how a Hermes-style learning loop is built.

How it fits with Hermes Agent

An educational reimplementation of Hermes Agent's learning, gateway, cron, profile and kanban ideas. It teaches the concepts and does not run on Hermes Agent.

FAQ

What is Learn Hermes Agent?

Learn Hermes Agent is a Python lesson series for building a self-evolving agent harness inspired by Hermes Agent. Each lesson adds one mechanism, from memory files to a kanban dispatcher.

Does Learn Hermes Agent work with Hermes Agent?

It is a separate teaching codebase that reimplements Hermes-style features in small examples. It does not install into or control a Hermes Agent instance.

Is Learn Hermes Agent free and open source?

Yes. The repository is released under the MIT license.

Similar resources for Hermes Agent

All resources

Related guides: SOUL.md for Hermes Agent: what it is and how to write one · What is Hermes Agent?