Back to reviews
Hermes Agent

Hermes Agent

Self-improving AI agent that learns new skills and runs on 200+ models

Hermes Agent is an open-source autonomous agent from Nous Research that actually gets better the more you use it. After completing complex tasks, it writes new skills to its own library — essentially bootstrapping its own capabilities over time. It's model-agnostic (200+ models via OpenRouter), self-hosts cleanly on a $5 VPS, and spans 6 terminal backends including SSH, Docker, and serverless Modal. The multi-platform messaging integration is genuinely useful: Telegram, Discord, Slack, WhatsApp, Signal, and email all pipe through a single gateway, so your agent can respond across every channel without separate bots. Persistent FTS5 memory means it remembers context across sessions. With 26k stars and 271 contributors already, this is moving fast. The one-line curl install and automatic project scaffolding make the onboarding friction unusually low for a project of this ambition.

Panel Reviews

The Builder

The Builder

Developer Perspective

Ship

Model-agnostic + multi-platform messaging + self-hosted for $5/month is the trifecta I've wanted from an agent framework. The skill-creation loop is genuinely novel — most agent frameworks require you to hardcode tools, but Hermes writes them from experience. The curl installer working out of the box sealed it for me.

The Skeptic

The Skeptic

Reality Check

Skip

An agent that writes its own skills is also an agent that can write broken or insecure skills, and Nous Research's security track record is thin. 271 contributors on a project with autonomous code execution is a supply-chain red flag. I'd audit extensively before giving this access to anything sensitive.

The Futurist

The Futurist

Big Picture

Ship

This is the closest thing to a general-purpose agent OS that exists in open source right now. The self-improving skill loop is a primitive form of recursive self-improvement — not AGI, but the architecture patterns being proven here will matter enormously in 2-3 years.

The Creator

The Creator

Content & Design

Ship

Having one agent respond across every messaging platform with persistent memory means I can actually run creative workflows — briefing docs, newsletter drafts, social scheduling — without babysitting separate bots per channel. The cron scheduling for recurring automations is the cherry on top.

Community Sentiment

Overall910 mentions
70% positive20% neutral10% negative
Hacker News310 mentions

Self-improving skill creation loop

Reddit220 mentions

Comparison to other open-source agents

Twitter/X380 mentions

Multi-platform messaging gateway