← All questions
Retrieval & EmbeddingsMedium· Database Selection· System Requirements· Multi-tenancy

We have 20 million chunks and multi-tenant access contr...

Avoid naming a product until you have defined the requirements. Evaluate the system based on memory headroom, filtered-search semantics, write rates, and operational ownership. Perform a capacity calculation: for 20 million chunks with M=16, you need significant resident memory. Consider the impact of multi-tenancy on filtering; if 98% of queries return nothing due to filtering, the database must handle that selectivity efficiently. Only after defining these axes should you select a class of database.

Code playground (Python)