Hardened CSV I/O
Read whole files or chunks with delimiter sniffing, explicit dtypes, decimal separators, encoding error policy, skip rows, and configurable bad-line handling.
Production data preparation before pandas does the analysis
Arnio combines a native C++ CSV engine with Python APIs for cleaning, quality profiling, schema validation, and data-stack interop.
$pip install arnioclick to copy
Latest releasev1.18.0
RuntimePython 3.9+
CoreC++ + pybind11
Current mainUnreleased updates
What Arnio ships today
The website now reflects the release train through v1.18.0 and the merged work currently on main after that release.
Quality reports
Profile nulls, duplicates, uniqueness, string cleanliness, high-cardinality columns, near-constant columns, quality score, drift, and CI quality gates.
Schema contracts
Validate types, ranges, regexes, emails, URLs, dates, datetimes, country and currency codes, custom validators, and row-level failures with max-error limits.
Interop built in
Move between ArFrame, pandas, Arrow, DuckDB, JSONL, CSV, and Parquet without scattering glue code across notebooks and pipelines.
A current Arnio workflow
Python
import arnio as ar
frame = ar.read_csv(
"sales.csv",
dtype={"id": "string"},
on_bad_lines="warn",
encoding_errors="replace",
)
report = ar.profile(frame, exclude_columns=["internal_notes"])
clean, explanation = ar.auto_clean(frame, mode="safe", explain=True)
schema = ar.Schema({
"id": ar.String(nullable=False, unique=True),
"email": ar.Email(nullable=True),
"homepage": ar.URL(allowed_schemes=["https"]),
})
result = ar.validate(clean, schema, max_errors=100)
table = ar.to_arrow(clean)
ar.write_parquet(clean, "sales.parquet", compression="zstd")Current main / unreleased
These changes are already merged after v1.18.0 and should be treated as upcoming release notes.
Frame ergonomics
ArFrame.__getitem__ column selection, ArFrame.drop_columns, and stronger tuple-key replacement handling.
Quality exports
profile(exclude_columns=...), DataQualityReport.to_dict(exclude_columns=...), and DataQualityReport.to_json().
Docs and CI polish
Benchmark regression smoke checks, duplicate-header hardening, and extra dry-run coverage for the pandas accessor.