VectorDBBench

Why we included this project

Choosing a vector database for production often means trusting vendor benchmarks, so it helps to run the same workloads on your own hardware and budget. VectorDBBench is a self-serve tool for that: it runs insert, search, and filtered-search tests across many open-source engines and managed services, and compares throughput and recall on identical datasets. The web UI makes it approachable for people who are not database specialists, while the CLI and YAML setup keep runs scriptable for CI or notebooks. It ships with production-derived datasets such as SIFT, GIST, Cohere, and C4-based embeddings, and the recent addition of BM25-style full-text search means you can evaluate vector and keyword retrieval in one pass. For managed clouds it also produces cost-effectiveness reports, which helps with right-sizing spend rather than just picking the fastest engine.