shape - see the shape of any data file

shape - events.ndjson illustrative mock-up

search any value… amount > 0 amount ▼ ~200,000 rows · 63,204 match

#	id	user	event	amount▼	props
88	4824	turing	purchase	149.00	{ plan: "pro" }
351	4835	ritchie	purchase	99.00	{ plan: "pro" }
12	4821	ada.byron	purchase	49.00	{ plan: "pro" }
210	4830	mccarthy	purchase	24.00	{ plan: "team" }
47	4823	hopper	purchase	12.50	{ plan: "team" }
103	4826	knuth	purchase	8.00	{ plan: "pro" }

jq SQL

.[] | select(.amount > 0) | {id, user, event, amount, props}

Illustrative mock-up - a stylized rendering of the interface, not a screen capture. Sample data is fictional.

Bigger than RAM

Streams, never loads

A multi-gigabyte file opens in one pass with bounded memory - nothing is slurped into RAM.

Honest numbers

Counts are exact - or say ~

Match counts are computed exactly. When a number is an estimate, it wears a leading tilde. No quiet guessing.

One engine

Same rows, every source

In-memory, streaming, SQLite and Parquet all run one filter model - byte-identical results whatever the file.

Everything, by clicking

Poke at a data file the way you always wanted to.

Most tools assume you already know jq or SQL. shape is for everyone who bounces off them - the whole workflow is visual, and the query is generated for you.

Explore the real rows

A virtualized table over the actual data - millions of rows scroll smoothly - beside a structure map of every field with its type, presence and distinct count.

Filter & search

Build type-aware AND/OR conditions by clicking, and search any value across every field. Applied live, with a cancellable exact match count.

Sort at any size

Click a header to sort - exact over the whole result on every tier, even a multi-gigabyte streaming file. Row numbers stay the true source ordinals.

Column statistics

Expand a field for its full profile in place: a distribution histogram, top values, quantiles and health flags - from the single open-time pass, no rescan.

Edit & save a copy

Double-click a scalar cell to change it; number literals stay exact, so a 19-digit id never loses a digit. Save back to a new file - the original is never touched.

Saved views

Name a filter + search + sort + reshape and reapply it anytime - across restarts. Plus export the full result to JSON, NDJSON, CSV, TSV or Parquet.

Take the query with you

Learn jq and SQL by watching them get written.

jqcopy

.[]
  | select(.amount > 0)
  | {id, user, event, amount, props}

SQLcopy

select id, user, event, amount, props
from data
where amount > 0
order by amount desc

Everything you build has both. The filter, search, sort and reshape you assembled by clicking appear as an exact jq expression and SQL query, ready to copy.
Differences are called out, not glossed. Where the three engines genuinely disagree, shape says so instead of pretending they match.
On SQLite, it actually runs. A filter it can translate is pushed into the database - measured ~12× faster on a 200k-row count - and anything it can't vouch for falls back to the same Go predicate, so the answer never changes.

Same core, on the command line

A streaming profiler for your terminal and CI.

The cgo-free engine also ships as a CLI: profile a file in one pass, infer a JSON Schema, and fail a pull request on breaking data-shape changes.

bash

$ shape profile events.ndjson

field type null distinct range

id number 0% 200,000 4821 … 61240

user string 0% 48,210 "aaronson" … "zuse"

event string 0% 6 purchase · signup · …

amount number 41% 312 -9.99 … 149.00

props object 0% - { plan, ref }

$ shape diff baseline.ndjson events.ndjson --fail-on breaking

✓ amount: nullability widened (added null) - non-breaking

✓ country: new field (string) - non-breaking

✓ no breaking changes - exit 0

Gate a PR with the hoijunkim/shape@v1 GitHub Action, or install with go install github.com/hoijunkim/shape@latest.

Open a file. See its shape.

Stop guessing at your data. Start reading it.

★ Star on GitHub Download

# desktop app - grab a build for your platform
→ github.com/hoijunkim/shape/releases

# or the CLI, anywhere Go runs
$ go install github.com/hoijunkim/shape@latest