Show HN: A pgvector fork with the performance of Pinecone
medium.com
Excited to share our latest project, pgvector-remote, with you all!
pgvector-remote is an extension designed to effortlessly incorporate remote vector stores, like Pinecone, into pgvector. With pgvector-remote, harnessing the power of a remote (Pinecone) index becomes as easy as utilizing ivfflat or hnsw.
Why pgvector-remote?
1. Simplifies integration of memory-intensive vector indexes like Pinecone into standard database servers.
2. Addresses critical feature gaps in pgvector, such as metadata filtering for AI applications.
3. Hides the complexity of maintaining and synchronizing vector stores from the user.
Key Features:
+ Achieve <10ms latency and 200QPS throughput, similar to Pinecone.
+ Enjoy metadata filtering capabilities, just like Pinecone.
+ Benefit from SQL compatibility and transaction support akin to PostgreSQL.
Usage: CREATE INDEX ON items USING PINECONE (embedding);
We're continuously enhancing pgvector-remote and would love to hear your feedback!