Skip to main content
Back

Nous Research's Hermes Agent Teaches Itself New Skills and Never Forgets

Nous Research's open-source Hermes Agent ships a self-improving loop, persistent memory, and 40+ tools — free, MIT-licensed, model-agnostic

Repo, Full Access
  • What it is: Nous Research's Hermes Agent is a free, MIT-licensed, self-improving AI agent with persistent memory and a built-in skill-creation loop.
  • Learning loop: After complex tasks, the agent autonomously writes reusable skill documents and refines them over time — the longer it runs, the more capable it becomes.
  • v0.19 Quicksilver release: First-token latency cut ~80% (4.3s → 0.9s), live reasoning streams by default, 14x faster desktop markdown rendering.
  • Platform reach: Runs on 23+ messaging platforms (Telegram, Discord, WhatsApp, iMessage, Teams, etc.) from a single gateway process, plus 7 terminal backends.
  • Model-agnostic: Works with OpenAI, Anthropic, OpenRouter, xAI, Google Vertex, local endpoints, and more — switch with hermes model, no code changes.
  • Research-ready: Built-in batch trajectory generation, Atropos RL integration, and ShareGPT export for fine-tuning tool-calling models.

Hermes Agent is an open-source, self-improving AI agent from Nous Research that has quietly become one of the most-starred agent frameworks on GitHub, now sitting at over 231,000 stars. Unlike session-based assistants, it is the only agent with a built-in learning loop , it creates skills from experience, improves them during use, nudges itself to persist knowledge, searches its own past conversations, and builds a deepening model of who you are across sessions. The pitch is simple: an agent that gets more useful the longer you run it.

What Hermes actually does

Hermes Agent is an open-source, self-improving AI agent by Nous Research that runs persistently with memory across sessions, and builds reusable skills from experience. The core idea is a closed learning loop that most agents skip entirely. When Hermes solves a hard problem, it writes a reusable skill document so it never forgets how. Skills are searchable, shareable, and compatible with the agentskills.io open standard.

Memory is not just a context window trick. Unlike session-based assistants, Hermes remembers everything across restarts. Your projects, preferences, and context are always there. It also builds a user model over time using Honcho dialectic modeling, so it learns how you think, not just what you've said.

The learning loop, under the hood

The self-improvement mechanism works in three layers:

  • Episodic memory , FTS5 full-text search across all past sessions, with LLM summarization for cross-session recall. Searching past sessions is now ~20ms and free after a rebuild that removed the aux-LLM dependency.