Persistent dual memory: time-agnostic facts + timestamped episodes, with automatic contradiction detection and vector-based recall.
Retrieve stored facts and episodes ranked by semantic similarity, recency, and importance.
Time-agnostic spatial knowledge ("Eiffel Tower is at 48.86 N, 2.29 E").
Timestamped observation ("PM2.5 was 120 in Delhi today"). Higher importance = longer retention.
Finds pairs of facts that are highly similar but claim different things — likely conflicting information about the same entity.
Current state of the spatial intelligence store. FAISS accelerates vector search when installed.
Deactivates old low-importance episodes to keep the store lean. Facts are always preserved.