Vectorwrap hides those differences behind one tiny Python adaptor: create a collection, upsert, query — then change only the connection string when you switch databases.
Here, I have tried to solve my direct pain point, so curious if this has application beyond my usecase.
What’s inside
Backends: Postgres 16 + pgvector, MySQL 8.2 (Vector Store), SQLite VSS, DuckDB VSS
Same filter dict syntax on every engine (SQLite uses adaptive oversampling internally)
Repo https://github.com/mihirahuja1/vectorwrap
I’d love some feedback on API design, missing edge cases, or reasons you wouldn’t adopt something like this. Thanks for taking a look!