SatoriDB is an embedded vector database designed for approximate nearest neighbor (ANN) search workloads that exceed available memory. It employs a two-tier architecture where a quantized HNSW index in RAM routes queries to disk-based vector buckets. CPU-pinned workers scan these buckets in parallel using io_uring and SIMD instructions. The system provides persistent storage through Walrus and RocksDB, targeting self-hosted applications requiring scalable semantic search and AI memory management without external services.