# Casper

A high-performance, Rust-built vector search database aimed at high-load AI and RAG systems, reporting an order-of-magnitude higher throughput than Qdrant in the project's own benchmarks.

Casper is a vector search database built in Rust for high-load search and AI/RAG applications, with client libraries in Python, Go, and Rust. It stores, searches, and manages vectors using a single monolithic HNSW index per collection.

The project's own benchmarks, run on a two-socket AMD EPYC 9474F server with 256GB of RAM against a nearly 10-million-vector, 96-dimension dataset, report up to an order of magnitude higher requests-per-second than Qdrant at matched HNSW parameters, across both float32 and int8-quantized configurations, which the project attributes to lower CPU and memory use per query. It notes that Qdrant's segmented index gives it higher recall at the same settings, since Casper's single monolithic index trades some recall for its throughput advantage.

Website: https://casper-frontend.pages.dev/