Nous Research has released Hermes Agent, an open-source AI agent that does something no other agent framework does well: it learns and improves from every interaction. With 70+ bundled skills across 15+ categories and a built-in learning loop, Hermes Agent creates new skills from experience and deepens its understanding of each user over time.
How the Skills System Works
Hermes Agent ships with skills covering MLOps, GitHub workflows, research, and more. But the real power is in self-evolution: the agent creates new skills on the fly based on recurring patterns it observes. These skills follow the open agentskills.io format and can be shared with the community or installed from ClawHub, LobeHub, and GitHub.
Self-Evolution Research
The companion repository hermes-agent-self-evolution implements evolutionary self-improvement using DSPy and GEPA (Genetic Evolutionary Prompt Architecture). This research was featured as an ICLR 2026 Oral presentation — a significant recognition. The system optimizes skills, prompts, and code automatically, making the agent genuinely better over time.
Multi-Platform and MCP Native
Hermes Agent works across Telegram, Discord, Slack, WhatsApp, Signal, and Email. It supports native MCP (Model Context Protocol) for standardized tool integration, meaning it can connect to any MCP server for extended capabilities. Skills activate conditionally based on which tools are available in the current environment.
Latest Release: v0.2.0
The v0.2.0 release (March 12, 2026) was a major milestone with 63 contributors, 216 merged pull requests, and 119 resolved issues. The Skills Hub for community discovery was a key addition in this release. Get started at github.com/NousResearch/hermes-agent.
Leave a Reply