Vector Databases & Semantic Search

Allows developers to store relational application state, user authentication (Row Level Security), and high-dimensional vector embeddings inside a single ACID-compliant Postgres database.

Graduated as a top-tier open-source project under the Linux Foundation (LF AI & Data); powers similarity search across collections exceeding tens of billions of embeddings in Fortune 500 banks.

Engineered from scratch in Rust with SIMD hardware optimizations to prevent unpredictable garbage collection pauses that plague Java- and C++-based vector engines.

Began as an open-source weekend project for local AI experimentation and rapidly became the default vector store in more than 80% of LangChain and LlamaIndex tutorials.

Changed the vector landscape by demonstrating that 90% of RAG architectures do not need a standalone vector database; they can simply execute vector queries on their existing Postgres tables.

Pioneered native Hybrid Search in vector databases, allowing developers to mathematically merge sparse BM25 keyword matching with dense HNSW vector similarity in one call.

Originally engineered by Yahoo to serve billions of real-time queries per day; evaluates machine-learned ranking models (ONNX/XGBoost) directly inside the storage nodes.

Runs directly on Cloudflare's 300+ global edge data centers alongside Workers AI, eliminating cross-cloud network roundtrips during RAG retrieval.

Founded by former Head of Amazon AI Lab Edo Liberty; the serverless storage engine completely decouples compute from index indexing using cloud object stores.

Powered by JVector, a pure Java vector search engine developed by DataStax that outperforms Lucene HNSW by using DiskANN-inspired graph indexing.