Titan is Frank's long-term knowledge graph and memory system (tools/titan/). It's a tri-hybrid storage engine:
Three Storage Layers
| Layer | Technology | What It Stores |
|---|---|---|
| SQLite | titan.db (WAL mode) |
9,072 nodes, 18,725 edges, claims, events, access logs |
| Vector Store | titan_vectors.npz |
608 vectors, 384-dim MiniLM-L6 embeddings |
| Knowledge Graph | Entity-relation triples | Relationships between concepts, people, events |
Retrieval
Queries go through Reciprocal Rank Fusion (RRF): vector cosine similarity + full-text search scores are fused, then modulated by time decay, confidence, recency, and graph connectivity.
The Hippocampus
The new hippocampus neural retrieval system (226K params) sits in front of Titan's legacy retrieval. It cuts latency from 300-400ms to 5-17ms and adds mood-congruent recall, temporal intent understanding, and Hopfield pattern completion.
Neural Cortex
The Titan Neural Cortex (77K params, 6 micro-networks) manages memory lifecycle: importance scoring, emotion tagging, retrieval weight learning, associative strengthening, consolidation gating, and interference detection.