Pinecone · Weaviate · Qdrant

What it is

Specialized databases that store embeddings and answer 'find me the 10 most similar vectors' in milliseconds across millions of items. Pinecone (managed), Weaviate (open-source, schema-rich), Qdrant (Rust-fast, self-host friendly).

How Vaaani uses it

• Semantic search across product catalogs or documentation
• Retrieval layer for every RAG pipeline
• Image / multimodal embedding search
• Metadata-filtered hybrid search (vector + keyword + filter)

Why it makes the cut

Mongo Atlas Vector or pgvector handle small-scale. Above 1M docs you need a real vector DB — Qdrant is my self-host default; Pinecone when the customer wants no ops.

Sample code

from qdrant_client import QdrantClient

client = QdrantClient(url="https://vaaani-qdrant.io")
hits = client.search(
    collection_name="docs",
    query_vector=embed("how do I cancel?"),
    limit=5)