Vector Database

Retrieval by meaning, not by keyword.

A Pinecone index your agent reaches over one edge endpoint, with every tenant kept in its own namespace. Upsert embeddings, run a similarity search, and pull back the closest matches with their metadata — the retrieval layer behind RAG and semantic search, without a vendor account to manage.

search by meaning

Matches on intent, not wording

Embeddings turn text into points in space, so “reset my password” lands next to “I'm locked out” even with no shared words. A query returns the nearest neighbours ranked by similarity — candidates your agent can ground an answer on.

  • Similarity scores on every match
  • topK plus metadata filters in one query

one index, many namespaces

Shared infrastructure, isolated data

Every tenant gets a private namespace inside a single Pinecone index. You get the economics and warm performance of shared infrastructure, while the edge scopes each read and write to your partition — so no query can ever cross the line into another tenant's vectors.

  • Namespace-scoped upserts and queries
  • No shared collection, no cross-tenant reads

pricing

Fluid pricing

Vector usage draws from your plan's monthly bucket; past that, it auto-buys from fluid credit. Two numbers set the ceiling:

read units

1m / month on free

write units

500k / month on free

Cohesivity Vector Database is a Pinecone index your agent reaches over one edge endpoint, with each tenant pinned to its own namespace. You hand it vectors and metadata; it gives you nearest-neighbour matches back, ranked by similarity. There's no index to size, no pod type to choose, and no Pinecone account to wire up.

Related

Pinecone, Namespace, Edge API, Nearest-neighbour, Managed

For more information, ask your agent.