xlsxturbo

High-performance Excel writer with automatic type detection. Written in Rust, usable from Python.

xlsxturbo exports pandas and polars DataFrames and CSV files to .xlsx. It uses Rust for the hot path and keeps the Python API small enough to drop straight into a script, a report job, or a batch pipeline. Roughly 7-9x faster than pandas + openpyxl on the reference benchmarks, with the Excel features those exports usually need — tables, conditional formatting, charts, data validation, images — available as focused keyword arguments rather than a workbook object model.

The .xlsx files themselves are written by rust_xlsxwriter — see Built on rust_xlsxwriter below.

Full documentation · Capability matrix · Changelog

Install

pip install xlsxturbo

Wheels are published for Python 3.10+ on Linux, Windows, and macOS. There are no runtime dependencies beyond the interpreter.

Export a DataFrame

import pandas as pd
from xlsxturbo import df_to_xlsx

df = pd.DataFrame({
    "product": ["Widget", "Gadget", "Gizmo"],
    "price": [19.99, 34.50, 8.75],
    "in_stock": [True, False, True],
    "restock": pd.to_datetime(["2024-03-01", "2024-03-15", "2024-04-01"]),
})

df_to_xlsx(df, "products.xlsx", table_style="Medium2", autofit=True)

Types carry across without configuration: numbers stay numbers, booleans become Excel booleans, and dates and datetimes become real Excel date values with a display format attached. polars DataFrames work the same way — neither library is a dependency.

Coming from to_excel, XlsxWriter, openpyxl or polars.write_excel? The migration guide has side-by-side examples and every place the output differs.

Convert a CSV

from xlsxturbo import csv_to_xlsx

csv_to_xlsx("sales.csv", "sales.xlsx")

Types are detected from the file's text. The same conversion is available from the shell, with no Python code at all:

pip install xlsxturbo
xlsxturbo sales.csv sales.xlsx --date-order us

See CSV conversion.

Write to memory

Every export function also writes to a binary file-like object, which is what a web download or an object-storage upload needs:

import io

buffer = io.BytesIO()
df_to_xlsx(df, buffer, autofit=True)
data = buffer.getvalue()  # the complete .xlsx file

The finished file is held in memory before it is written out, including with constant_memory=True. A FastAPI example and the full rules are under Writing to memory.

What it can do

• DataFrame and CSV export — pandas, polars, and CSV in, .xlsx out
• Excel tables with 61 built-in styles, autofilter, and banded rows
• Formatting — header and per-column styles, number formats, per-side borders, alignment, wrapping, auto-fit and explicit column widths, row heights, merged ranges, rich text
• Conditional formatting — colour scales, data bars, icon sets
• Formulas — calculated columns and workbook-level defined names
• Native Excel charts and in-cell sparklines, both editable in Excel
• Data validation — dropdowns, numeric ranges, text-length constraints
• Cell-level extras — arbitrary cell writes, hyperlinks, comments, checkboxes, images, textboxes
• Multi-sheet workbooks with per-sheet option overrides
• Constant memory mode for very large exports, and optional parallel CSV parsing
• Atomic writes — a failed export never truncates the file already at that path

The capability matrix is the authoritative list: it is generated from the source and shows which options each function accepts, which are overridable per sheet, and which survive constant-memory mode.

Performance

On 100,000 rows x 50 columns of mixed types, with every library writing the same Excel table, number formats and column widths, xlsxturbo is about 3.6x faster than polars, 6x faster than pandas + xlsxwriter, and 8.6x faster than pandas + openpyxl, and needs the least memory of the four. Measured on one MacBook Pro; an older Windows measurement with each library's default output gave 3.9x, 5.8x and 7.4x. Full tables, test systems, and methodology are on the performance page, and both benchmark suites live in benchmarks/ so you can measure your own hardware.

Known limitations

• Write-only. xlsxturbo creates workbooks; it cannot open or modify an existing one.
• Timezone-aware datetimes are written as their local wall-clock value — Excel has no timezone concept, so the UTC offset is not preserved.
• Integers above 2^53 are written as text to avoid silent precision loss.
• Durations (Timedelta / timedelta64) are written as text; Excel has no duration type.

The compatibility page has the complete list with the workaround for each.

Project status

Stable since 1.0.0. Everything reachable from import xlsxturbo without a leading underscore is covered by Semantic Versioning and will not break before 2.0.0; anything removed gets a DeprecationWarning naming its replacement and its removal version, for at least one minor release and at least six months.

The stability page is the full statement — the public surface named exhaustively, what does and does not count as breaking, and the supported Python and platform matrices.

• Tested in CI on Python 3.10 and 3.12 across Linux, Windows, and macOS, plus Python 3.15 on Linux. One abi3 wheel per platform serves 3.10 through 3.15. Python 3.9 was dropped in 1.1.0; a 3.9 interpreter resolves to 1.0.0, which stays on PyPI.
• Advanced Excel features are exposed through focused parameters rather than a full workbook object model. That is a deliberate scope boundary, not a gap to be filled.

Built on rust_xlsxwriter

Every byte of the .xlsx files xlsxturbo produces is written by rust_xlsxwriter, John McNamara's Rust Excel writer (MIT licensed). It is the one substantial dependency, and it is not an implementation detail you can ignore:

• What xlsxturbo can do is bounded by what rust_xlsxwriter can do. The features on the capability matrix are the ones it exposes; xlsxturbo's job is type detection, the DataFrame and CSV pipeline, option validation, and a Python API — not the XLSX format itself.
• Bugs in the generated file are usually upstream. When one is, it is reported to rust_xlsxwriter rather than papered over here — most recently #185, filed 2026-08-15 and fixed in 0.98.1 the next morning. A file Excel refuses to open is still worth reporting to us; we will trace it and take it upstream.
• The xlsxwriter in the benchmark table is a different project — that is XlsxWriter, the pure-Python library by the same author, and one of the things xlsxturbo is measured against.

If you write Excel files from Rust, use rust_xlsxwriter directly; xlsxturbo exists to put it behind a Python DataFrame API. rust_xlsxwriter's notice, and those of every other crate compiled into the wheel, are in THIRD-PARTY-LICENSES.md, which is generated from the dependency tree rather than maintained by hand.

Contributing

Setup takes about five minutes and is described in CONTRIBUTING.md, along with the exact lint, type, and test commands CI runs. Security reports go through the process in SECURITY.md.

License

MIT — see LICENSE.

The wheel contains compiled code from the Rust crates listed above, so their notices travel with it: THIRD-PARTY-LICENSES.md, also installed as xlsxturbo-<version>.dist-info/licenses/THIRD-PARTY-LICENSES.md. It is generated from the dependency tree by python scripts/gen_third_party_licenses.py --write; do not edit it by hand.