Qdrant, the Vector Search Database, raised $28M in a Series A round

Original Article ↗ Hacker News Discussion ↗

Qdrant, an open-source Rust-based vector search database, has raised a $28M Series A round, prompting debate over its technical maturity and long-term prospects in an increasingly crowded and commoditizing vector database market. Commenters share mixed production experiences—ranging from praise for performance and ease of use to concerns about instability and “rough edges”—and frequently compare Qdrant to alternatives like pgvector, Milvus, Weaviate, and general-purpose databases adding vector support. The thread also surfaces skepticism about the defensibility of specialized vector DB businesses versus in-house or cloud-provider solutions, alongside criticism and defense of Qdrant’s hiring practices, which lean heavily on unpaid open-source contributions as a screening mechanism.

Real-world use of Qdrant

  • Several commenters report production or near-production use.
  • Positive reports: handles tens of millions of vectors, frequent inserts/deletes, predictable performance; core functionality (speed, accuracy, filters) seen as solid.
  • Negatives: described by some as “rough around the edges,” with breaking changes between releases and less robustness than mature data stores.
  • Proof-of-concept users like the simplicity for RAG pipelines and ease of spinning up an in-memory instance.

Vector DB vs Existing Databases/Search

  • Many use Postgres with pgvector (including HNSW) and prefer “the devil you know.”
  • Some argue vector DBs are really search engines (BASE) not full databases (ACID).
  • Multiple commenters say most “normal” DBs and search engines now support vectors, questioning the need for a separate product except at large scale.
  • Others value dedicated services for robustness, hybrid search features, and ease of use.

Market, Moat, and Business Viability

  • Disagreement on whether “vector databases” are a real, durable category.
  • Skeptics call it low-tech compared to modern AI, expect big AI providers or existing DB vendors to absorb this functionality, and doubt long-term standalone businesses.
  • Supporters point to active research in ANN search, performance variance among products, and enterprise needs (security, segmentation, reproducibility) as justification.
  • Some note Qdrant is open source, which both helps adoption and weakens moat.

Benchmarks and Comparisons

  • Multiple OSS and cloud options cited: Milvus/Zilliz, Weaviate, pgvector, Elasticsearch/OpenSearch, LanceDB, Chroma, others.
  • Vendor-run benchmarks (e.g., VectorDBBench, Qdrant’s own) are discussed; some see them as useful, others as marketing and not vendor-neutral.

Architecture & Implementation Notes

  • Qdrant’s Rust codebase praised as readable and permissively licensed.
  • One thread criticizes heavy use of Arc (shared pointers) as “Java-style”; others respond this is natural in async Rust and not necessarily a performance problem.

Hiring Practices Controversy

  • Significant debate over Qdrant’s hiring approach requiring applicants to build open-source connectors as a screening task.
  • Some see this as clever ecosystem seeding and self-selection for motivated contributors.
  • Others view it as unpaid labor and another burden on candidates already investing heavily in interview prep.