Nous Research Ships Hermes Agent: An Open-Source Agent That Writes Its Own Skills After Every Complex Task

Nous Research — the team behind the Hermes model series — has shipped Hermes Agent, an open-source autonomous agent with an unusual architecture: it writes new skills to its own library after completing complex tasks, gradually expanding its own capabilities through use.

**The skill creation loop** works by having the agent reflect after complex tasks and encode reusable patterns as skills. Over time, the agent becomes more capable without requiring the developer to manually add tools or functions. The skills are stored locally and can be reviewed, edited, or deleted by the user.

**Model-agnostic by design:** Hermes Agent supports 200+ models via OpenRouter and accepts direct connections to Nous Portal, OpenAI, Anthropic, and custom endpoints. Switching models requires no code changes. The persistent memory system uses SQLite with FTS5 for fast cross-session recall.

**Multi-platform messaging** is handled via a single gateway that spans Telegram, Discord, Slack, WhatsApp, Signal, and email. A single agent can be the backbone of your entire messaging presence across platforms, with cron-based scheduling for recurring automations like daily briefings or competitor monitoring.

**Infrastructure accessibility** is a notable feature: Hermes Agent runs on a $5/month VPS, supports six terminal backends (local, Docker, SSH, Daytona, Singularity, Modal), and can hibernate serverlessly when idle. The one-line curl installer and five-question project scaffold make initial setup unusually fast for a project of this scope.

**Community traction:** 26.3k stars, 271 contributors, MIT license. The project is moving quickly and the Nous Research team has a track record of shipping functional open-source model infrastructure.