Associative Memory Fields: Zero-Infrastructure Spreading Activation for LLM Multi-Agent Systems

Paper 2-2 in the Buddys Architecture Series. When DRM v2's retrieval scope was expanded from memory/ to all persistent text (session logs, handoff notes, inbox messages), keyword-based accuracy dropped from 100% to 72.7% due to keyword collisions. We introduce associative memory fields — free-form association words (3-5 per entry) that create implicit connections between memories, enabling spreading activation without embeddings or GPU. Results: v1=0%, v2=50%, v2+associations=100% on 12 probe queries. The architecture is structurally a neural network built from text files: neurons (entries), synapses (association words), activation (ACT-R), learning (session-end annotation). Zero infrastructure. One word per weight update.