Pinecone.
Managed vector database — the default agent memory layer in enterprise.
What it is.
The category-defining managed vector database. Pinecone Serverless decoupled storage from compute and dropped retrieval cost by an order of magnitude in 2024. Hosted on AWS, GCP, and Azure with documented enterprise governance. The default vector layer for production RAG.
Where it fits.
Anywhere the buyer needs millions to billions of vectors served sub-100ms with no infrastructure team. Replaces self-hosted Postgres-pgvector or Weaviate at the point where operational overhead exceeds the cost differential.
- Managed — no infrastructure team required
- Multi-cloud with documented governance
- Strongest enterprise procurement story in the category
- Lock-in to managed service for hosted workloads
- Cost climbs faster than self-hosted pgvector at certain scales
Frequently asked.
Pinecone or pgvector?
pgvector for under 10M vectors and an existing Postgres estate. Pinecone above that scale or when operational overhead matters more than per-vector cost.
Does Pinecone support hybrid search?
Yes. Dense plus sparse retrieval with metadata filters. The standard RAG pattern at production scale.
Is Pinecone Serverless the default tier?
Yes for new deployments since late 2024. Pod-based legacy tier remains supported for existing customers.
