datasight — a fast terminal viewer for CSV, Parquet, and JSON

A vim-keybinded TUI for exploring tabular data from the command line. Browse, filter, sort, group, and plot CSV, TSV, Parquet, JSON, and NDJSON files directly in your terminal — no spreadsheet, no notebook, no web UI required. Built in Rust on ratatui, with 9 built-in Base16 color themes.

Features

• Vim-style navigation (hjkl, g/G, PageUp/PageDown)
• Search within a column (/, n/N)
• Multi-column filtering with comparison operators — > 30, = Engineering, != 0 (f, F)
• Unique values popup — browse and filter by distinct values instantly (u)
• Hierarchical multi-column sort — s cycles ascending → descending → off, additional columns append as secondary priorities (s, S to clear)
• Group-by with per-column aggregations; filters and sorts survive the aggregation switch (b, a, B)
• Multi-series plots — select multiple Y columns with Space for side-by-side comparison, or a single Y for line/bar/histogram (p, t)
• Column Inspector — schema and stats for every column at a glance (i)
• Column stats popup (e)
• Export the current view to CSV — filters, sorts and group-by included (w)
• Automatic date detection for ISO (YYYY-MM-DD) and common non-ISO formats (MM/DD/YYYY, DD-Mon-YYYY), with an ambiguity guard for slash-dates
• In-app help popup (?)
• 9 built-in Base16 themes — --theme nord, gruvbox-dark, dracula, etc., switchable live with T and persisted across runs
• Supports CSV, TSV, Parquet, JSON ([{...}]), and NDJSON/JSON Lines (.ndjson, .jsonl) files
• Custom delimiter support via -d flag — works with pipe-separated, semicolon-separated, and any single-character delimiter
• Pipe-friendly — reads from stdin with automatic format detection (CSV, JSON, NDJSON)
• Viewport-windowed rendering — stays fast on large files
• File browser (browse) — navigate local directories and Azure Blob / S3 buckets, fuzzy-find in the listing (/), open files directly into the viewer, download cloud objects locally (d)

Install

Pre-built binaries (recommended)

Download the latest binary for your platform from the GitHub Releases page.

Build from source

Requires Rust 1.75 or higher.

cargo install datasight

Or clone and run locally:

cargo run -- tests/fixtures/orders.csv
cargo run -- tests/fixtures/orders.tsv
cargo run -- tests/fixtures/orders.parquet
cargo run -- tests/fixtures/orders.json
cargo run -- tests/fixtures/orders.ndjson
cargo run -- -d '|' <path-to-file.psv>   # pipe-separated
cargo run -- -d ';' <path-to-file.csv>   # semicolon-separated

Sample fixtures for all supported formats are in tests/fixtures/.

Pipe from stdin

datasight reads from stdin when no file is given. Format is detected automatically from content:

# CSV
cat data.csv | datasight

# TSV
cat data.tsv | datasight

# Pipe-separated via flag
cat data.psv | datasight -d '|'

# JSON array of objects
curl https://api.example.com/records | datasight

# Newline-delimited JSON
kubectl get pods -o jsonlines | datasight

# From a database
psql -c "\copy (SELECT * FROM orders) TO STDOUT CSV HEADER" | datasight

File Browser

Browse local directories or cloud storage and open files directly into the viewer:

datasight browse .                          # current directory
datasight browse /data/exports              # local path
datasight browse az://my-container         # Azure Blob Storage
datasight browse s3://my-bucket            # AWS S3

Cloud backends require building with the relevant feature:

cargo install --path . --features azure     # Azure support
cargo install --path . --features aws       # S3 support

Azure reads AZURE_STORAGE_CONNECTION_STRING or individual AZURE_STORAGE_ACCOUNT_NAME / AZURE_STORAGE_ACCOUNT_KEY env vars.

Browse keybindings

Key	Action
j / k	Navigate list
. / Enter	Open file or enter directory
Esc	Go up to parent
Tab	Switch focus browser ↔ viewer
ctrl-e	Toggle browser sidebar
d	Download the selected cloud file locally (az:// / s3:// only)
/	Fuzzy-find in the listing
T	Open theme picker
q	Quit

/ filters the listing as you type: the query matches anywhere in a name, in order, so ordcsv finds orders.csv. Matches are ranked — hits on word boundaries and runs of consecutive characters score highest — and the matched characters are highlighted. A lowercase query is case-insensitive; one uppercase character makes the whole query case-sensitive. Enter opens the highlighted match, Esc cancels and restores the full listing, arrows (or ctrl-n / ctrl-p) move through the matches, and ctrl-u clears the query.

d prompts for a destination, pre-filled with the remote file name. ~ expands to your home directory, and a destination ending in / (or an existing directory) keeps the remote name — so ~/Downloads/ saves sales.csv as ~/Downloads/sales.csv. The prompt shows the resolved path and warns before overwriting an existing file; Enter saves, Esc cancels. Missing directories in the destination are created for you.

Themes

datasight ships with 9 Base16 color themes:

Name	Display
mocha (default)	Catppuccin Mocha
latte	Catppuccin Latte
frappe	Catppuccin Frappé
macchiato	Catppuccin Macchiato
gruvbox-dark	Gruvbox Dark
nord	Nord
dracula	Dracula
solarized-dark	Solarized Dark
tokyo-night	Tokyo Night

Pick one at startup:

datasight --theme nord file.csv
DATASIGHT_THEME=dracula datasight file.csv

Or switch live: press T to open the theme picker, j/k to browse with live preview, Enter to confirm (saves to ~/.config/datasight/state.toml), Esc to revert.

Precedence: --theme > DATASIGHT_THEME > saved state file > mocha (default).

Keybindings

Navigation

Key	Action
j / Down	Move down
k / Up	Move up
h / Left	Move left
l / Right	Move right
g / Home	Jump to first row
G / End	Jump to last row
PageDown	Scroll down 20 rows
PageUp	Scroll up 20 rows

Search

Key	Action
/	Enter search mode (searches in current column)
Enter	Confirm search and jump to first match
n	Next match
N	Previous match
Esc	Exit search and clear results

Filter

Key	Action
f	Enter filter mode (filters rows by current column)
Enter	Confirm filter and return to normal mode
F	Clear all filters
Esc	Discard input

Supports comparison operators for numeric columns: > 30, < 100, >= 0, <= 50, = 42, != 0. Use = text or != text for exact string matching. Plain text falls back to substring search.

Unique Values

Key	Action
u	Open unique values popup for current column (sorted by frequency)
type	Search / filter the list live
j / k	Navigate the list
Enter	Apply selected value as a filter and close
Esc	Close without filtering

Sort

Key	Action
s	Cycle current column: ascending → descending → off
s on a new column	Append as next sort priority (primary stays in place)
S	Clear all sorts

Active sorts are shown in the header with ①▲ / ②▼ glyphs and summarised in the status bar.

Group By

Key	Action
b	Toggle group-by key for current column
a	Cycle aggregation for current column (Σ μ # ↓ ↑)
B	Execute group-by / clear and return to full view

Plot

Key	Context	Action
p	Normal	Enter pick-Y mode
h / ← / l / →	Pick-Y / Pick-X	Move between columns
Space	Pick-Y	Toggle the current column as a Y series (select one or many)
Enter	Pick-Y	Confirm Y selection and advance to pick-X
i	Pick-Y	Use row index as X and render the chart immediately (skips pick-X)
Enter	Pick-X	Confirm X column and render the chart
Esc	Pick-Y / Pick-X	Cancel (pick-X goes back to pick-Y; pick-Y returns to normal)
t	Plot	Cycle chart type (line → bar → histogram for single-Y; line ↔ bar for multi-Y)
Esc / p	Plot	Close the chart and return to normal mode
q	Plot	Quit

Selecting multiple Y columns draws them as overlaid series with a shared X axis — great for comparing actual vs fit vs residual over a date column.

Numeric X columns are plotted directly. String or date X columns use row indices as data points and render the actual values as rotated (vertical) labels below the chart — all labels are shown when they fit, otherwise they are sampled evenly.

For histogram, the Y column is binned automatically — no X column selection needed.

Column Inspector

Key	Action
i	Open Column Inspector (type, count, nulls, unique, min, max, mean, median)
j / k	Navigate rows
Enter	Jump to the selected column and return to data view
Esc / i	Close and return to data view

Column Stats

Key	Action
e	Toggle stats popup for current column (count, min, max, mean, median)

Copy

Key	Action
y	Copy the selected cell to the clipboard
Y	Copy the current row to the clipboard, tab-separated

Copying uses the OSC 52 escape sequence, so the terminal performs the copy — which means it also works over SSH. Supported by Windows Terminal, WezTerm, kitty, iTerm2, Alacritty, Ghostty and Konsole. GNOME Terminal disables OSC 52 by default, and inside tmux you need set-clipboard on. Where it is unsupported the sequence is ignored silently.

Export

Key	Action
w	Write the current view to a CSV file
Enter	Confirm and write
Esc	Cancel without writing

The prompt opens prefilled with a name derived from the source (orders.csv → orders.export.csv; piped input becomes export.csv) and writes to the current directory unless you type a path. Absolute paths, relative paths and a leading ~ all work, and .csv is appended when it is missing so the extension always matches the contents. The status line reports the resolved path, so you can see where the file actually went.

The file contains the rows exactly as filtered, sorted and grouped on screen — not the original file. Null values are written as empty fields: the ∅ shown on screen and the (null) filter sentinel are display conventions and never reach the file.

A warning appears in the prompt before an existing file is overwritten, and the write goes to a temporary file that is renamed into place only on success — so a failed export (a column CSV cannot represent, a full disk) leaves the existing file untouched. Missing directories in the destination are created for you, the same as d in the file browser.

Exports are always local. Viewing a file from browse az://… or s3://… suggests a local name from the basename (az://container/data/orders.parquet → orders.export.csv) and writes to the current directory — the rows are already in memory, so nothing goes back over the network. Typing a remote destination is flagged in the prompt and refused before any file is touched; writing back to object storage is not supported.

Other

Key	Action
i	Open Column Inspector
_	Autofit current column width (press again to reset)
=	Autofit all columns
-	Narrow current column
+	Widen current column
e	Toggle column stats popup
?	Toggle help popup
T	Open theme picker
q	Quit

Troubleshooting

The display looks garbled or misaligned

Your terminal may not support 256 colors or Unicode box-drawing characters. Try a modern terminal emulator (kitty, alacritty, iTerm2, Windows Terminal) and make sure TERM is set to xterm-256color.

Columns are too narrow or too wide

Press = to autofit all columns to their content, or _ to autofit only the current column; press _ again to return that column to the default width.

Autofit grows a column to fit its longest value, capped at the width of the table pane — a column wider than the pane cannot be scrolled into, so fitting beyond that point would only push other columns off-screen. For values longer than your terminal is wide, widen the terminal and autofit again, or use - and + to size the column by hand.

Filtering with >, <, >=, <= shows an error

These operators only work on numeric columns. Use = value or != value for exact string matching, or plain text for substring search.

The unique values popup shows fewer results than expected

The popup is capped at the 500 most frequent values. If your column has more than 500 distinct values, the title will say [top 500].

Large files are slow to open

datasight reads the entire file into memory on startup using Polars. In practice, load times are fast: a 42 MB / 1M-row CSV loads in under 0.1s using ~115 MB RAM. Multi-GB files will use proportionally more memory. Once loaded, navigation and filtering are fast regardless of row count.

Parquet file fails to open

Make sure the file is a valid Parquet file. Compressed or encrypted Parquet variants are not supported.