proj47-GGBOND

华东师范大学-T2026102699910471-GGBOND-proj47 | LIM-HNSW Agent Memory VectorIO is a dynamic vector retrieval prototype for long-term Agent memory under constrained memory. It combines in-memory HNSW navigation, PQ distance estimation, SSD-based exact reranking, and WAL/memtable-based incremental writes to support million-scale vector search, mixed read-write workloads, persistence, and Agent memory recall under an online runtime memory budget of about 20% of the dataset size.