Vector database and retrieval stack for teams building search, recommendations, copilots, and retrieval-augmented generation systems.
This stack is for teams adding semantic retrieval to applications. It includes the main architectural choices: Postgres-native vectors, dedicated vector databases, embedded stores, and RAG orchestration.
pgvector is a PostgreSQL extension enabling vector similarity search directly within the database, eliminating need for separate vector database infrastructure. Free and open-source (PostgreSQL license).
Used here for: PostgreSQL extension path when embeddings should live next to relational application data.
Chroma is an open-source (Apache 2.0) embedding database designed for developer simplicity and rapid prototyping of AI applications. Core library is free; Chroma Cloud in preview with pricing TBD.
Used here for: Developer-friendly vector store for local prototypes and small RAG applications.
Qdrant is an open-source (Apache 2.0) vector database written in Rust providing high-performance similarity search with rich filtering capabilities. Open-source is free to self-host.
Used here for: Vector database option with strong filtering, payload support, and production deployment paths.
Pinecone is the leading purpose-built vector database providing managed similarity search for AI applications including text-to-SQL implementations. Free Starter tier includes 100K vectors; Standard and Enterprise tiers with usage-based pricing.
Used here for: Managed vector database for teams that want hosted scale and less operational overhead.
Weaviate is an open-source (BSD-3) vector database providing hybrid (vector + keyword) search with optional managed cloud service. Open-source is free to self-host.
Used here for: Vector database with schema, hybrid search, and application-oriented APIs.
LanceDB is an open-source (Apache 2.0) serverless vector database using the Lance columnar format, designed for zero-copy integration with ML frameworks. Free and self-hosted; LanceDB Cloud in development.
Used here for: Embedded and file-native vector database pattern for local or data-lake-adjacent workflows.
LlamaIndex provides NLSQLTableQueryEngine and SQL-focused components for building text-to-SQL applications with emphasis on data retrieval quality. Open-source (MIT license) Python library.
Used here for: Retrieval and data connector layer for building RAG over structured and semi-structured data.
LangChain provides SQLDatabaseToolkit and SQL Agent capabilities for building text-to-SQL applications with LLM orchestration. Open-source (MIT license) Python and JavaScript libraries.
Used here for: Agent and SQL integration layer for teams orchestrating retrieval, tools, and model calls.
No more localhost demos
Ship AI-built prototypes as a live link
Publish AI-generated tools and demos to a public URL in seconds — no setup.
Publish free →