Hermes Agent
The agent that writes down what it learns
What it is
Nous Research's self-improving, model-agnostic agent runtime. After a session it evaluates what worked and distills it into reusable skill files, layered over a four-part memory system (session, persistent, skills, user modeling), six execution backends from local to Docker to Modal, and frontends spanning CLI, Telegram, Discord, Slack, WhatsApp, and Signal.
Why it's interesting
Launched February 2026 and near 100k stars within months — the most credible open attempt at an agent that genuinely improves with use. Plain MIT, no enterprise tier, no license asterisks, from a lab with a long open-weights track record.
Use cases
- A personal assistant that lives in your messaging apps
- Self-hosted automation that accumulates skills instead of resetting
- Research on skill compilation and agent self-improvement
Who it's for
Tinkerers, agent builders, self-hosters
Setup
Moderate. Python plus an LLM API key from any major provider; Docker recommended for sandboxed execution. Messaging integrations each need their own tokens
Limitations & cautions
Rapid v0.x releases with no API-stability promise, self-improvement works best on capable (usually paid) models, and an agent with shell access plus your messaging accounts deserves careful sandboxing.
Editorial takeaway
OpenClaw proved everyone wants a personal agent. Hermes is the bet that the one that remembers you wins.