Hermes Agent

Hermes Agent is an open source agent from Nous Research with a genuine learning loop. It keeps memory across sessions, writes itself new skills when it solves something hard, and reaches you through a messaging gateway. It runs self hosted or in the cloud, on whichever model provider you choose.

Both are open, self hostable, messaging first agents. OpenClaw’s strength is its enormous ecosystem and channel reach; Hermes is built around the learning loop of persistent memory and self written skills. We deploy both and recommend per use case.

With governance, yes. Memory can hold unverified inferences and skills are code the agent wrote, so we review both on a rhythm. Unsupervised self improvement is not a feature we ship.

Your choice. It runs against major cloud providers or local models through Ollama or LM Studio, so sovereignty and cost can be designed rather than accepted.

Recurring knowledge work: daily briefings, research sweeps, monitoring and any workflow where remembered context and repeatable method beat starting fresh.

When it works out how to do something awkward, it can save that method and reuse it later instead of starting again. That is useful, and it is also why it needs review. You want to know what it has taught itself.

By treating its memory and its skills as things that get reviewed, not things that just pile up. We agree where they are stored, who looks at them and how often, before it goes near real work.

Teams with a repeating task that changes slightly each time, and someone technical who can own it. If you need something that behaves identically on every run, a plain automation is the better tool.