Hermes Agent: NousResearch's Self-Improving Open-Source Agent Framework

Nous Research has released Hermes Agent, an open-source AI agent framework that learns and improves itself over time. Trending on GitHub with 874 stars per day and 11,000 total stars, Hermes Agent represents a new approach to persistent, self-evolving AI assistants.

Hermes Agent features a closed learning loop: it autonomously creates skills from complex tasks, refines them during use, and maintains persistent memory across sessions. It ships with 40+ bundled skills covering MLOps, GitHub workflows, and research, and creates new skills on the fly using the open agentskills.io format.

The agent runs across Telegram, Discord, Slack, WhatsApp, Signal, and CLI through a unified gateway process. It supports 200+ models via OpenRouter, Nous Portal, OpenAI, Anthropic, and custom endpoints. Deployment is flexible — from a $5 VPS to GPU clusters to serverless infrastructure that costs nearly nothing when idle.

Version 0.3.0 was released on March 17, 2026, featuring unified real-time token delivery, a first-class plugin architecture, and a rebuilt provider system. The project also includes Tinker-Atropos for batch trajectory generation and RL environment support.

GitHub: https://github.com/NousResearch/hermes-agent
License: MIT