Pinecone

Pinecone is a closed-source, managed vector database that its docs call “a search engine for AI agents and applications”. One index can serve full-text, semantic, sparse-vector and hybrid search.

Capabilities (docs.pinecone.io)

  • Index types: a schema-defined document index, and a legacy vector index. The overview says that from API version 2026-07, creating an index is schema-only.
  • Search: sparse and hybrid search from the same index; BM25 full-text with Lucene query syntax; metadata filters; reranking; integrated embedding (server-side text to vector) and a hosted sparse model, or bring your own embeddings.
  • Multitenancy: namespaces; dedicated read nodes for high query volume.
  • Stated limits (sparse-index docs): 2,048 non-zero values per sparse vector; top_k up to 10,000; 4 MB maximum query result; 40 KB metadata per record. The distance metrics shown are cosine in the example; the full list and dimension limits were not on the pages read.

Deployment and billing model

Managed serverless, or Bring Your Own Cloud; backups and restores. No self-hosted option appears in the docs read. Plans are named Starter, Builder, Standard and Enterprise. Usage is metered as read units, write units, storage and egress, with imports, backups, embedding (per token) and reranking billed separately; the larger plans carry minimum commitments, and HIPAA is an add-on for Standard and included in Enterprise. Rates are on Pinecone’s pricing page and are not recorded here. Serverless versus pod index types were not on the pages read.

Where it fits

Pinecone is the no-operations choice: you trade control and portability for a managed service. Compare the open options qdrant, weaviate, milvus, chroma and pgvector in vector-databases-compared.

Sources