Problem

The Memory module currently uses SQLite FTS5 (full-text search) for memory recall. Phase 2 — vector embeddings for semantic search — is not yet implemented.

File: Packages/Sources/Memory/MemoryStore.swift:10

/// Phase 1: SQLite FTS (full-text search). Phase 2: vector embeddings.

What's needed

• Generate embeddings via LLM provider (or on-device via CoreML)
• Store embedding vectors in SQLite or a separate vector store
• Implement cosine similarity search for semantic recall
• Fall back to FTS when embeddings are unavailable

Current state

• ✅ Phase 1: FTS5 search with tagging, importance ranking, session-scoped retrieval
• ❌ Phase 2: Vector embeddings and semantic search