Omna · Python

Build RAG Without a Vector Database

For most internal RAG pilots, a vector DB is overkill. If your corpus already fits in a DataFrame, Omna gives you retrieval as a method call.

The API

import polars as pl, omna
docs = pl.read_parquet("kb.parquet")
hits = docs.omna.search(user_question, top_k=5)
context = "\n".join(hits["text"].to_list())

Try it now

pip install omna