// PYTHON LIBRARY
Python library overview
Omna adds hybrid search, enterprise-grade PII detection and masking, and schema understanding directly to your Polars DataFrames — through a `.omna` namespace, with no vector database, no API key for core features, and no data leaving your machine.
The .omna namespace
Importing omna registers a .omna accessor on every Polars DataFrame. There is nothing else to wire up.
import polars as pl
import omna
df = pl.read_csv("documents.csv")
df.omna.search("insurance claim denied", on="text", k=5)The one exception is [understand_df](/docs/understand-df), which is a top-level function (omna.understand_df(df)) rather than a namespace method.
What Omna does
| Method | What it does |
|---|---|
[omna.understand_df(df)](/docs/understand-df) | Schema inference — labels, null rates, samples. No LLM. |
[df.omna.embed(column)](/docs/embed) | Vectorize a text column once; reuse across sessions |
[df.omna.search(query, on, k)](/docs/search) | Top-k results by hybrid relevance (semantic + keyword) |
[df.omna.filter(query, on, threshold)](/docs/filter) | Every row above a similarity threshold |
[df.omna.pii_report()](/docs/pii-report) | Audit every string column for PII |
[df.omna.mask_pii()](/docs/mask-pii) | Redact PII, with an automatic audit log |
[df.omna.ask(question)](/docs/ask) | Natural-language queries over your DataFrame |
Install
pip install "omna[all]"See the Quickstart for the per-feature extras (omna[embed], omna[pii], omna[ask]) and platform support. Omna is tested on Polars 1.0+ and Python 3.10–3.12 (macOS or Linux).
Performance
Omna inherits Polars' Apache Arrow columnar memory. The Rust similarity kernel operates on that same memory — no copy into NumPy, no copy into a C buffer.
| 50k rows | 500k rows | |
|---|---|---|
| Omna search | 9ms | 27ms |
| Omna filter | 9ms | 27ms |
Benchmarked on a MacBook Air M5, 10-query median, warm index.
License
| Layer | License |
|---|---|
Python package (omna/) | MIT |
| Rust engine | Proprietary — ships as a compiled binary in the pip wheel |