Not Surfie’s rack, but the spirit is right.
I have been using the Hermes agent for just over two weeks now.
Installed it twice, and moved it once. It now lives on a virtual machine within a Proxmox server that lives on a machine on the network. I call it Surfie, because long ago — months ago — I started using agents with Openclaw, and the first install was on a Microsoft Surface Pro 4 machine.
Surfie has access to most of the machines on the network. Hermes is improving almost daily, as did Openclaw. Eventually, it will take over the projects that I’ve been working on.
For now, I know it can read and write files between computers. It can run shell commands on those computers. It can do everything it needed to do to learn how to do those things — so it is capable of much more.
This took days. Just shy of two weeks, working off and on, because I have a life.
Now I pause to consider what has been done, and what is possible. Yesterday, I asked Surfie to summarize what we have done so far. It did. It was very technical. I took that output and gave it to the model in the Open WebUI interface and told it to clean up the output. It was still too technical. So, I modified the ask:
“Revise your last output. I need a fluid narrative — entertaining, slightly colorful. Remove information unique to my network.”
The result is on the other blog for AI content.
易 What makes Hermes different
The feature that sets Hermes apart is the way it learns. Everything it does is saved to .json and .md files. I asked it to combine the session .json files into a single markdown file. There are two files — one for each installation. I’m thinking I might load up the most capable model I have, give it the files, and tell it to write its (short) life history.
No comments:
Post a Comment