Category · 3 options
Choosing vector search
Stores embeddings of documents so an AI feature can find the passages to answer from. Start inside your existing Postgres (pgvector) until you have millions of chunks. Filter every query by tenant so one customer never sees another's documents.
Free plans and prices
In Cuely's default order for the general case. Your needs change the order; see below.
| Tool | Free plan | Paid from | At 1,000 users |
|---|---|---|---|
| pgvector (in Postgres) No extra vendor: vectors live in the Postgres you already have, so tenant filters are a plain WHERE clause. | Free: on Neon it is "available on every Neon plan with no add-on or paid tier required"; you pay only for database storage and compute. | Not stated | $0/month |
| Pinecone Managed vector search with a free plan that covers a small app. | Starter (free): up to 2 GB storage, 2M write units and 1M read units per month, 5 indexes, us-east-1 only. | Builder: $20/month flat. Standard: $50/month minimum usage; storage $0.33/GB-month. | $0/month |
| turbopuffer Low storage cost at very large scale. | None | Launch: $16/month minimum usage; storage up to $0.33 per GB. | $16/month |
Best pick by need
From the published scores Cuely ranks with. Only needs where one tool clearly leads.
- a database per customer: turbopuffer. Many small namespaces (one per customer) are a common pattern.
- fewer vendors: pgvector (in Postgres). Vectors stay in your existing Postgres; no new vendor or sync job.
- high volume: turbopuffer. Object-storage based, so very large indexes stay cheap.
- keep costs as low as possible: pgvector (in Postgres). No separate bill; it uses the database you already pay for.
- prefers open source or self-hosting: pgvector (in Postgres). Open-source extension that runs on any Postgres.