WunDeeDB.jl: An easy to use, zero config, WAL, SQLite backend vector database

WunDeeDB.jl (v1.0.0) is a package written in Julia Lang that provides a SQLite disk-backed system for storing, searching, and managing embedding vectors at scale, influenced by disk oriented graph-based ANN techniques and the broader insights from hierarchical small-world graphs. By maintaining embeddings in an SQLite database, WunDeeDB.jl reduces in-memory overhead while supporting efficient similarity searches on commodity hardware. Its design also facilitates integration with common vector-database or ML pipelines that rely on embedding retrieval. In contrast to fully in-memory approaches, WunDeeDB.jl leverages disk-based storage and user-configurable adjacency (e.g., HNSW, LM-DiskANN, or fallback linear search), allowing large-scale data to be handled without saturating RAM. It supports incremental insertions and deletions, ensuring the index remains up-to-date as datasets evolve. By combining these disk-native strategies with tunable BFS expansions and adjacency pruning, WunDeeDB.jl enables robust nearest neighbor searches for high-dimensional embeddings