# Qdrant Qdrant is an open-source (Apache 2.0) **vector database** written in [[Rust]] — the full-service counterpart to embeddable indexes like [[FAISS]] or [[Turbovec]]. Where those give you a library and a file, Qdrant gives you a server: an API, replication, snapshots, multitenancy, and a managed cloud if you want it. Runs as a single Docker container locally, scales to clusters. ## What it does well - **Filtered vector search as a first-class citizen.** Every vector carries a JSON payload; filters are applied *during* HNSW graph traversal rather than pre/post-filtering — the feature that made its reputation, since naive filtering wrecks either recall or speed. - **Quantization menu**: scalar, product, and binary quantization to trade memory for recall — with [[TurboQuant]] integrated as well, which is why the [[Turbovec]] HN thread kept asking "why not just use Qdrant?" - **Hybrid search**: dense + sparse vectors (BM25-style) fused in one query, the practical default for [[Retrieval-Augmented Generation (RAG)]]. ## Choosing between the tiers Embedded index ([[Turbovec]], [[FAISS]], usearch) when the index is a file inside your app and you own the ops. Qdrant when you need filters, persistence, concurrent writers, or a service boundary. Postgres/pgvector when the vectors are a column, not the product. ## References - [Qdrant Website](https://qdrant.tech) - [Qdrant on GitHub](https://github.com/qdrant/qdrant) ## Related - [[Vector Store]] · [[Embeddings]] · [[Semantic Search]] - [[Turbovec]] · [[FAISS]] · [[TurboQuant]] — the embedded-index end of the spectrum - [[Retrieval-Augmented Generation (RAG)]] - [[Rust]]