LatestFree desktop app for Windows, macOS and Linux
Open huge data files in seconds.
DataLiq opens multi-gigabyte CSV, Parquet, JSON and Excel files in about a second,
then lets you filter, query with SQL, chart, clean and export them. Nothing is
uploaded. No account. It works offline.
Real screenshot · a 1.0 GB CSV of 9,221,261 rows · timings from the app's own status bar
CSV
TSV
Parquet
JSON
NDJSON
Excel .xlsx · .xls
.gz · .zst
SQLite
DuckDB
MySQL
PostgreSQL
Folders as one table
HTTPS · S3
A2SpeedMeasured on a 4-core Linux machine
Built for the files that freeze spreadsheets.
DataLiq reads only what is on screen, runs on the DuckDB engine, and keeps a cache on
disk so the second open is instant. Queries that outgrow memory spill to disk instead
of crashing.
B2fxfirst_rows("orders_5gb.csv")
First rows of a 5 GB CSV
0.28s
Reopening a file you opened before
2ms
Opening a 50-million-row Parquet file
0.05s
Full import of that 5 GB CSV, 46 million rows
40s
Filter 50M rows, first rows
0.1–0.9s
Sort 50M rows, first and last rows
0.8–1.0s
Chart of 50M rows
0.1–1.9s
Column profile of 50M rows
0.7–2.7s
A3FeaturesThe real app on real files
From a raw file to an answer, without leaving the window.
SQL
Every open file is a table.
Query, join and aggregate your files with SQL. Drag a column from the sidebar into
the editor, or Ctrl+click it, and its name goes in where you want it.
current_view is always the rows on screen, with your filters applied.
Explain plans, history and saved queries
Open any result as a new tab
Joins across files, folders and databases
48 ms for this query over 9.2M rows
SQL
JSON
See the shape of JSON before you read a line.
The structure view reads every record and draws one map: each key with its type, how
often it appears and how long its arrays are. Here, two million log records show
that user_id is in 80% of them.
One record as connected cards, or the shape of all
Filter by a value, add it as a column, unnest arrays
Also for nested Parquet columns
2,072,884 records in one map
web_logs.ndjson.gz — Structure
Charts
Charts from the whole file, not a sample.
Bar, line, pie, histogram, scatter and box plots, computed by the engine over every
row. Group by a date, split by a column, stack it. Export as PNG or SVG, or copy the
chart's SQL.
Dates bucketed by day, month or year
Sum, average, count and more
Follows your filters as you change them
543 ms for 60 bars over 9.2M rows
orders.csv — Chart
And everything else you reach for.
#A · FeatureB · What it does
1Filter, sort, searchStack filters as chips, pick values with counts, search every column.
2Column profileNulls, distinct values, a histogram and the most common values of any
column.
3CleanTrim, change types, find and replace, fill empty cells; save the steps as a
recipe.
4Data qualityEmpty columns, mixed types, stray spaces, outliers and duplicates, with their
rows.
5Compare filesRows added, removed or changed between two versions, cell by cell.
6Group by and pivotCounts, sums, medians and percentiles as a new tab. No SQL needed.
7Join, append, sampleJoin two files, stack several into one, or open a random sample.
8Export and convertCSV, Parquet, JSON or Excel: one file, a folder per value, or a whole folder at
once.
9Databases and URLsMySQL and PostgreSQL tables, files over HTTPS and from S3.
10Never lose your
placeFiles, filters and layouts come back after a restart. Undo and redo every
change.
11Command paletteEvery action by name with Ctrl+K. Workspaces save tabs, filters and charts.
12Light, dark,
keyboardBoth themes meet WCAG 2.2 AA contrast; everything works from the keyboard.
A4AutomationCommand line and MCP
Run it without the window.
Save your cleaning steps as a recipe and apply them from the command line: reports on
a schedule, with no window. Or turn on the built-in MCP server and let Claude Desktop,
Claude Code or any MCP client list, describe and query the folders you choose. It is
read-only, limited to those folders, every call is logged, and it is off until you
turn it on.
list_filesThe data files in the folders you share
describe_fileRow count, columns and types, first rows
profile_columnNulls, distinct values, ranges, common values
queryOne read-only SQL query at a time
open_in_dataliqShow a file in the app, as a new tab
Terminal
# Apply a saved recipe and write a report$dataliq run monthly.json --in orders.csv.gz --out report.xlsx# Open files in the running app$dataliq orders.csv web_logs.ndjson.gz# Start the MCP server for AI assistants$dataliq mcp
A5PrivacyOffline by design
0 bytes uploaded.
Your data stays on your computer.
No accountOpen the app and work. Nothing to sign up for.
No telemetryNo tracking, no analytics, no upload.
Files untouchedEverything happens in a view. You write a new file when you export.
The only network requests are the URLs and databases you open yourself, DuckDB
extensions the first time one is needed, and an optional daily check for a new version.
Open source under the MIT licence.
A6DownloadFree, open source, no account
Get DataLiq Latest
Installers for Windows, macOS and Linux are published on GitHub Releases.
Yes. DataLiq is open source under the MIT licence, free for personal and commercial
use, with no paid tier.
How big a file can it open?
Files larger than your memory are fine: a 5 GB CSV with 46 million rows and a
50-million-row Parquet file are part of its benchmarks. Parquet is read in place;
other formats are imported once into a cache on disk.
Does it change my files?
Never. Filters, cleaning steps and layout changes apply to a view of the data. You
write a new file when you export.
Will it take over my CSV and Excel files?
No. DataLiq is the default program only for its own files (.parquet, .duckdb,
.dataliq). For CSV, Excel, JSON and SQLite your current program stays the default,
and DataLiq is under Open with.
macOS or Windows warns me about the app. Why?
The installers are built on GitHub from the public source, and each release lists
SHA-256 checksums. If your system says the app can't be verified, open it once with
right-click › Open on macOS, or More info › Run anyway on Windows.