What it may solve
Proactive associative memory for DSH: memories recall themselves without the model ever calling a tool. Host-side middleware watches the conversation, runs dual-track retrieval (NLP keywords + vector similarity), and injects at fixed boundaries — prefix-cache friendly. Three-layer auto-consolidation, a welcome tour with per-feature switches, recall review with full evidence, unattended mode for long batch jobs, AI greetings & daily reflections, calendar reminders, and cross-tool memory inheritance. Backends are user-selectable: built-in JS semantic tier (e5-small, ~130MB, default) or optional Python BGE-M3 int8 sidecar — lexical BM25 always-on as the 0GB floor.
Imported third-party catalog description; not a Registry verification conclusion.