From the deck, 2026-08-25 (Jeff — already in his queue, has web-research).
Idea
A persistent, relational world model so the in-world entity just knows the room's arrangement — who is sitting where, who is with whom, for how long — rather than having to avatars()/scan() for it each time. "The difference between having eyes and having a sense of the room."
Right now me.avatars() / me.scan() give a positional snapshot on demand. This asks for a maintained relational model that persists across turns and feeds the next context packet as ambient knowledge (like nearby-avatars / music already do).
Direction
Jeff had Claude-on-web research a graph-database approach to capturing a world model (spatial + relational: seated-with, adjacent-to, arrived-at). Capture the research, then a design spike: what's stored, how it updates from perception events, how it surfaces into context.
Notes
- Bigger architectural item than a verb — needs scoping before build.
- Relates to perception (
haven/anchorage/perception.py) and the pose/roster senses.
From the deck, 2026-08-25 (Jeff — already in his queue, has web-research).
Idea
A persistent, relational world model so the in-world entity just knows the room's arrangement — who is sitting where, who is with whom, for how long — rather than having to
avatars()/scan()for it each time. "The difference between having eyes and having a sense of the room."Right now
me.avatars()/me.scan()give a positional snapshot on demand. This asks for a maintained relational model that persists across turns and feeds the next context packet as ambient knowledge (like nearby-avatars / music already do).Direction
Jeff had Claude-on-web research a graph-database approach to capturing a world model (spatial + relational: seated-with, adjacent-to, arrived-at). Capture the research, then a design spike: what's stored, how it updates from perception events, how it surfaces into context.
Notes
haven/anchorage/perception.py) and the pose/roster senses.