pgvector
Best if you use PostgresPostgreSQLSelf-host~13kPostgres extension for vector search. The default if you already use Supabase, Neon, or Postgres. Production-ready.
github.com/pgvector/pgvectorEvery RAG app needs a vector database. The question in 2026 is no longer "Pinecone or build your own" — it's which open-source option fits your stack. pgvector wins if you already use Postgres. Qdrant leads on pure-vector performance. Chroma is the easiest start. Weaviate has the best hybrid (keyword + vector) search. Below are the honest picks for production AI apps.
Postgres extension for vector search. The default if you already use Supabase, Neon, or Postgres. Production-ready.
github.com/pgvector/pgvectorPostgres extension for vector search. The default if you already use Supabase, Neon, or Postgres. Production-ready.
github.com/pgvector/pgvectorRust-based dedicated vector DB. Best raw performance + filtering. Hosted cloud or self-host with one Docker container.
github.com/qdrant/qdrantEasiest start. `pip install chromadb` and you have a vector DB. Best for prototyping + sub-100k embeddings.
github.com/chroma-core/chromaHybrid search (vector + keyword + filters). Best when search relevance matters more than pure semantic similarity.
github.com/weaviate/weaviateLargest-scale vector DB. Billion+ vectors. Pick this when you outgrow others. Heavy to operate but unmatched at scale.
github.com/milvus-io/milvusEmbedded vector DB (think SQLite for vectors). Zero infra. Best for on-device or edge-deployed AI.
github.com/lancedb/lancedb