Hermes and Honcho for memory!

[mortal hollow]

Started a thread to continue this discussion!!

I was gonna point Hermes at the OpenClaw honcho self hosting docs and have it stand me up a local docker-compose stack to play with, is this the recommended way to get things up and running or is there a more turnkey recommendation 😁

[lean portal]

Oh sweet! The dreaming layer seems to be the current edge of the stack. I guess I was asking how you are thinking about approaching that.

[inner needle]

we recently updated our self-hosting docs shouldnt have to point at the openclaw ones 🫡 --- https://docs.honcho.dev/v3/contributing/self-hosting

[mortal hollow]

Perfect

[inner needle]

there are definitely many ways expand upon it. the team has been going back and forth internally but we plan to release a roadmap soon-- might look like deeper links / tagging .. but then that could be more difficult to unmesh and self heal if there needs to be an identity pivot. also having abductive conclusions / world building hypothetical memory banks could be fun-- predictive reasoning to help with user-agent alignment

my ideal would be to expose all of these knobs which now live in a black-box sort of state within honcho

[lean portal]

I wonder if there is a good graph algorithm for removing links/tags like that.

I need to better understand what you are referring to by self-heal I think. I’ll take a look at the repo later today or tomorrow.

Is the idea of an “identity pivot” something you see in the wild or something you are targeting as a behavior?

[inner needle]

this could be from a user decidely changing their behavior or perhaps from a fragmented representation -- editing the peer card also helps with the self-healing here. some facts are weighted more than others when honcho reasons over user conclusions/data and while deduplication could be valid or 'deleting' conclusions might help, the more natural approach is to have honcho continue reasoning to redirect its course. and we are back to ideas of plasticity!