// PYTHON LIBRARY

search()

`df.omna.search(query, on, k)` finds the rows that best match your query by meaning — not just keywords — and returns them ranked by relevance. It runs on a local Rust kernel: ~9ms over 50,000 rows, no API key, no network.

2 min readHybrid semantic + keyword search

Signature

python

df.omna.search(query, on, k, hybrid=True)

Parameter	Description
`query`	The search string, in plain language
`on`	The column to search
`k`	Number of top results to return, ranked by relevance
`hybrid`	Hybrid (semantic + keyword) when `True` (default); pure semantic when `False`

Note: Run [df.omna.embed("column")](/docs/embed) once before searching. search() loads the saved index automatically on every call after that.

Returns a Polars DataFrame of the matched rows, with a _score column (cosine similarity, 0–1), ordered from most to least relevant.

Basic example

python

import polars as pl
import omna

df = pl.read_csv("documents.csv")
df.omna.embed("text")                                       # once

results = df.omna.search("insurance claim denied", on="text", k=5)
print(results)

code

 uid            document_type         domain      text                               _score
 67fccc1e207…   ClaimSummary          insurance   **Claim ID: 285-14-1755, Policy…   0.762
 b8ae088cd21…   ClaimSummary          insurance   **Claim Summary**…                 0.749
 de5bba0a2cc…   Insurance Claim Form  healthcare  **Insurance Claim Form**…          0.748

None of these rows contain the literal phrase "insurance claim denied" — Omna finds them by meaning.

Hybrid by default

Search fuses semantic (embedding) similarity with BM25 keyword matching using Reciprocal Rank Fusion. Semantics catch meaning; BM25 catches rare exact tokens the embeddings blur — part codes, IDs, surnames, acronyms.

python

results = df.omna.search("XJ9000", on="parts", k=5)               # exact code → BM25 nails it
results = df.omna.search("claim denied", on="text", hybrid=False)  # meaning only

vs. keyword search and regex

Regex and str.contains match only exact strings. A query for "chest pain" misses "cardiac pressure", "tightness in the chest", and "angina". Semantic search encodes meaning, so synonyms, paraphrases, and related concepts all match — with no synonym list to maintain.

All search computation happens inside your Python process. Patient records, trade secrets, and other sensitive text never leave the host machine.

What's next

DocsFilter by threshold — filter() →DocsMask PII first — mask_pii() →