What Hermes Agent is
Hermes Agent is an open-source AI agent from Nous Research, officially described as an “open-source, self-improving AI agent,” released under the MIT license. It runs from the terminal, messaging platforms, and IDE workflows, and is positioned as a personal agent that accumulates experience the longer you use it. The product site is hermes-agent.ai, and the code lives on GitHub under the NousResearch organization.
Capabilities and how it runs
Per the official README and docs, Hermes provides persistent memory, reusable skills, tool use, cron jobs, browser automation, code execution, and multi-agent delegation; it creates skills from experience and improves them during use. It is model-agnostic: the hermes model command switches between model endpoints (the README lists Nous Portal, OpenRouter, OpenAI, or a custom endpoint).
There are several ways to run it. A single gateway process can connect Telegram, Discord, Slack, WhatsApp, Signal, and the CLI at once. Execution backends include local, Docker, SSH, Daytona, Singularity, and Modal — from a cheap always-on VPS to serverless backends that hibernate and wake on demand. Supported systems are Linux, macOS, WSL2, and Android via Termux; native Windows is not supported and requires WSL2. Installation is covered in the official how-to, with docs at hermes-agent.nousresearch.com.
Discussion in the show
In Weekly #001’s chapter “OpenClaw 和 Coding Agent 正在双向融合” (OpenClaw and coding agents are converging), the panel placed Hermes in a comparison of personal-agent products: Xiangyang Qiaomu argued OpenClaw’s experience still trails Hermes, and Guizang said Hermes at least behaves like a product team. Yang Pan offered the opposite direction of experience: he switched from OpenClaw to Hermes the month before, and found the gap significant — when he assigned a task over several consecutive messages, Hermes often stopped mid-way, less smooth than OpenClaw. These are individual impressions; the same product can behave differently under different usage patterns.
In “产品迭代速度，正在成为新的组织能力” (iteration speed as an organizational capability), Guizang used the Hermes team as a benchmark, arguing that teams unable to match Hermes-class iteration speed cannot keep up. That is a participant’s view on team pace, not official information.
Frequently asked questions
What is Hermes Agent?
An open-source personal AI agent developed by Nous Research under the MIT license, usable from the terminal, messaging platforms, and IDE, with persistent memory, skill accumulation, cron jobs, browser automation, and multi-agent delegation. See the official README.
How do I install Hermes Agent? Does it support Windows?
The official how-to provides a one-line install script. Supported systems are Linux, macOS, WSL2, and Android via Termux. Native Windows is not supported — Windows users should install WSL2 first.
How does Hermes Agent relate to OpenClaw?
They are peer products in the personal-agent category. Hermes officially supports migrating from OpenClaw with hermes claw migrate, which imports existing configuration, memories, and skills. In Weekly #001, Yang Pan, Xiangyang Qiaomu, and Guizang each described their own experiences with the two products — and did not all agree.
Which models can Hermes Agent use?
It is model-agnostic: hermes model switches between Nous Portal, models on OpenRouter, or a custom endpoint with no code changes, per the official site. Check the docs for currently supported options.