Byte 2.0's memory system live: 64-dim hash embeddings, L2 normalized, cosine similarity, stored in IndexedDB. Add memories, search semantically, consolidate old conversations — all local, zero cloud, same engine Byte uses for long-term recall.

Add memory (text → 64-dim embedding → IndexedDB)

IndexedDB: not initialized

Embedding preview (64-dim, L2 norm)

Semantic search (cosine similarity over embeddings)

Memory types in Byte 2.0

Short-term: Last 100 turns, buffer with summarization when >5000 chars
Long-term: This IndexedDB vector store, semantic search via cosine
Episodic: Interaction history with timestamps
Semantic: Private notes + embeddings (this demo)
Consolidation: Old conversations summarized and embedded for long-term recall

🧠 Browser Vector Memory — IndexedDB + Semantic Search