Cross-Source Fusion Search for Heterogeneous AI Memory Systems

AI coding assistants maintain two distinct types of memory: semantic memory (code indexes) and episodic memory (learned facts and decisions). These memories typically reside in separate databases with different schemas, lifecycles, and write patterns. Existing hybrid search approaches (RRF, DBSF, linear combination) assume a single data source, while multi-memory systems (Mem0, MemGPT, Zep) either delegate source selection to the LLM or return unranked concatenations. We present a cross-source fusion algorithm that exploits a shared embedding space to produce unified rankings across heterogeneous databases. Our multiplicative scoring formula addresses a previously undocumented vote asymmetry problem in additive multi-source fusion, where independent per-source BM25 rankings inflate text-match influence relative to a unified vector signal. Key contributions: - Three-level search architecture separating per-source and cross-source fusion - Multiplicative fusion formula eliminating vote asymmetry in multi-source BM25 - Graceful degradation to BM25-only mode when vector search is unavailable - 7x latency reduction via parallel retrieval (620ms → 86ms) The approach requires no changes to existing per-source search pipelines. Project: https://github.com/gladego/index1