PGVector

pgvector is an open-source PostgreSQL extension for vector similarity search, not a Faiss port and not a hosted vector cloud. Rechecked 2026-08-18: github.com/pgvector/pgvector had 22,651 stars, description Open-source vector similarity search for Postgres, and a PostgreSQL-style license (GitHub SPDX NOASSERTION, LICENSE text matches the Postgres / UC wording). README lists L2 / inner product / cosine / L1 / Hamming / Jaccard, HNSW and IVFFlat, plus halfvec, bit, and sparsevec types. Drop leftover copy that said the implementation is "based on Faiss."

Compare PostgreSQL if you wanted the database itself, Supabase if you wanted a hosted Postgres that already ships vectors, and Faiss if you wanted Meta's in-process ANN library instead of SQL.

Key Features

• Extension, not a service: You install CREATE EXTENSION vector in Postgres. There is no pgvector SaaS dollar table on the repo.
• Not Faiss: First-party README and LICENSE do not say the code is based on Faiss. Faiss is a separate MIT library. Do not keep that leftover claim.
• Indexes you actually get: HNSW and IVFFlat, plus iterative index scans. Scale still follows your Postgres cluster, not a dedicated vector fleet.
• Often bundled by hosts: Supabase and many Postgres hosts enable the extension. Their seats are not pgvector prices.

Use Cases

• People who already run Postgres and want embeddings in the same transactions as rows.
• Teams that will accept HNSW/IVFFlat instead of a dedicated vector engine.
• People comparing SQL vs Qdrant / Pinecone after they see 22,651 stars.

Limitation: We did not invent a pgvector Cloud seat. Hosted Postgres bills belong to the host.

Pricing

GitHub on 2026-08-18.

If a leftover blog still says pgvector is based on Faiss, treat this table as the source.

Getting Started

1. Open github.com/pgvector/pgvector and install the extension in PostgreSQL.
2. Recheck Supabase only if you want a hosted Postgres that already includes vectors.
3. Recheck Faiss if you wanted an in-process library, not SQL.
4. Do not paste "based on Faiss" into a deck.

Frequently Asked Questions

Is it free?

Yes. The extension is open source under a PostgreSQL-style license.

Is it based on Faiss?

No. That leftover claim is dropped. Faiss is a separate Meta library.

Same as Pinecone?

No. Pinecone is a managed vector DB with Free / $20 / $50 min / $500 min. pgvector is a Postgres extension.

Alternatives

• PostgreSQL: The database you attach this to.
• Supabase: Hosted Postgres with vectors as one feature.
• Faiss: MIT similarity-search library, not a SQL type.
• Qdrant: Dedicated OSS vector engine plus Cloud.

Tips

1. Quote 22,651 stars and the Postgres license from the GitHub check.
2. Drop the Faiss origin story.
3. Recheck the host bill before you invent a pgvector seat.

Conclusion

pgvector is a PostgreSQL-licensed vector extension with 22,651 stars, not a Faiss wrapper and not a hosted dollar seat. Start at github.com/pgvector/pgvector, then decide whether Supabase or Qdrant already covers the product shape you need.