Skip to content

Comment on Building AI Memory at 10M+ Nodes: Architecture, Failures, and Lessonsparent

Comments

Hey we already had PostgreSQL so no new infrastructure to manage, it was easy way to see if vector database change has any value. It also has good enough performance - handles 10M vectors with HNSW indexes adequately open source - leverages existing infrastructure for future migration. we've created a vector service, easy to swap later if needed

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.