Skip to main content
Back

Nous Research's Hermes Agent Builds Skills and Gets Smarter With Every Session

NousResearch's Hermes Agent is a self-improving, open-source AI agent that builds skills from experience, remembers across sessions, and runs on any model or platform

Repo, Full Access
  • Self-improving agent: Hermes Agent by Nous Research creates reusable skill files from completed tasks and refines them with each use.
  • 231k GitHub stars: MIT-licensed, launched February 2026, now at v0.20.3 with 22,000+ commits and a fast-growing ecosystem.
  • Free to self-host: Nous Portal optional paid tiers start at $20/month (Plus) for 248 models and a bundled tool gateway.
  • Runs everywhere: CLI, native desktop app (macOS/Windows/Linux), Telegram, Discord, Slack, WhatsApp, and 6 execution backends including Docker, SSH, and Modal.
  • Model-agnostic: Switch providers with hermes model — supports OpenAI, Anthropic, OpenRouter, local endpoints, and Nous Portal.
  • Honest tradeoff: Better persistent learning and lower costs than OpenClaw, but fewer community skills and a large open issues backlog (5,700+).

Hermes Agent is Nous Research's open-source autonomous agent, and it has a pitch that most agent frameworks don't dare to make: it gets better the more you use it. It's 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. With 231k GitHub stars and growing, it has become one of the most-watched open-source agent projects of 2026.

Hermes Agent is an open-source, self-improving autonomous agent built by Nous Research and released under the MIT license. It launched in February 2026 with a built-in learning loop that creates skills from experience, refines them during use, persists memory across sessions, and runs on 20+ platforms , from a $5 VPS to a GPU cluster to serverless infrastructure. The project is now at v0.20.3 and has accumulated over 22,000 commits.

The Loop That Actually Matters

Most agents forget everything the moment a session ends. Hermes is built around the opposite assumption. Hermes has a practical agent-system learning loop: the agent context, memory, skills, files, and procedures all improve over time. For most day-to-day work, that second kind of learning is what you actually need.

It's worth being precise about what "learning" means here. Hermes does not retrain the underlying language model. It does not update model weights. It is not fine-tuning itself in the background. Instead, the learning happens at the agent layer: memory files, skill documents, and session search indexes that accumulate and compound over time.

Core features include a self-improving learning loop that observes outcomes and refines behavior, a three-layer memory system (working, episodic, semantic/skill) for robust context retention, and automatic skill creation and evolution to produce reusable tool modules. After a complex task, Hermes can autonomously write a Markdown skill file capturing the procedure , and refine it the next time it runs the same workflow.

This is Hermes' consent-aware learning loop: repeated corrections and durable workflow lessons become compact memory entries or procedural skills, while