Do you want to give Agentic AI real memory?

Last week, when I was digging in deep with Hermes Agent, and one of the things that intrigued me was that it came with Honcho baked in.

I had never heard of it, but after a little research, I could see this being a tool that helps make AI agents actually useful.

One of the biggest drawbacks with a lot of this agentic stuff is that each new chat they seem to be completely blank slates waiting for you to explain yourself.

The privacy advocate in me kinda likes that it doesn’t remember things, but then quickly remembers that every keystroke is still likely recorded and sold off by the big data companies.

You power users of AI out there are probably way ahead of me yelling, “Skills” and prompts will fix it.

Perhaps, but what if you wanted it to learn from your conversations with and remember what you talked about yesterday? Then you might be looking for a technology like Honcho.

The thing I find unique about Honcho is that it allows you to track ‘peers’.

This means you can give it context/message history for more than just you and the chatbot. I could feed in conversations I had in my Q/A sessions or podcasts, and it could learn about a wide range of people I interact with.

As I write this, I will note that as exciting as this might be, it also has a Skynet vibe.

It also has “Dreaming,” which is a fancy(and somewhat scary) word for asynchronous background processing.

They offer a hosted version, which they charge for, but also a self-hosted version. I am playing around with the self-hosted version.

My guess is that under the hood, we are looking at some fancy vector indexing with heavy pre/post processing.