From the author

I don't have official docs yet- i'm still using it to see how well it works. I will post here , so far it's been behaving very well- instead of adding any memory hooks that manage memory, or wiki or .mds it does not use any of that- it only uses the mem palace, and saves all things in the mem palace "format"

A subagent watches for patterns and automatically builds training scenarios for how the agent is fetching memories from mempalace, and watches for ways it can improve it.

Something I've noticed that is a huge difference between this and other models like openclaw or hermes is the speed-

Using Qwen3.6 the responses are under 500ms and they include memory recall.

https://github.com/kenchambers/kent

Screenshot 2026年05月03日 at 12 30 30 AM > Visual representation after only a few days of use.