The strongest general-purpose memory combines two ways of remembering. Vector memory recalls by similarity (what is this like); graph memory recalls by relationship (what is connected to this). GBrain is our hybrid approach that uses both, so an agent can retrieve a relevant past fact and traverse how it connects to others, which is closer to how useful memory actually works.

This lesson teaches the pattern and when the hybrid earns its complexity over either alone.