So I’ve recently started using both OpenClaw and Hermes Agent, and I’ve come to a quick conclusion. Hermes Agent is wayyy better. First of all, OpenClaw was hogging all of my CPU on my Raspberry Pi. Hermes Agent uses 5%. Hermes Agent is also much much faster, and has a better UX. It shows what tools calls the agent runs, it gets execution approvals right, its self learning is actually great, and the model patches it so often and everything ends up working that I’m not even sure why does OpenClaw.
I’m currently using Hermes Agent with my Codex subscription and Kimi K2.6 via OpenRouter, and routing it through Discord, and it just ends up working so well. I ask a question, it gets me a response before I can even process the fact that I sent a message. For complicated messages, it gets it right (mostly) without too much hassle. I’ve found that using GPT 5.4 works better than GPT 5.5 for Codex subscription limits, even though GPT 5.5 is more token efficient.
Nous, keep up the good work. I love it. It definitely exhibits the self learning well, where as OpenClaw would shove random things into MEMORY.md and call it a day.