Local DataWorkbench

Private data tools · Edition 01

Your data never
leaves the desk.

Open a multi-gigabyte CSV, JSON, or Parquet file. Profile it, reshape it, and publish a repeatable result—without an upload or a one-off script.

Download for your computer Read the method

Detecting your platform…

A strip of gridded archival paper passes through a black mechanical typesetter, connected by a thin oxblood thread.
Fig. 01 Raw rows in. A documented edition out.
01 · Files stay local02 · Memory stays bounded03 · Recipes stay portable04 · Exports stay deterministic

The working method

A paper trail for data work.

Not a spreadsheet pretending to be a database. Not a SQL editor expecting you to remember what you typed last Tuesday.

  1. 01

    Open, never upload

    Read CSV, JSON/JSONL, and Parquet directly from disk. Only a bounded preview is held in memory.

  2. 02

    Inspect the facts

    See inferred types, missing values, distinct counts, ranges, total rows, and a source fingerprint.

  3. 03

    Name each change

    Filter, clean, rename, select, and join through explicit steps that read like an editorial docket.

  4. 04

    Publish the same result

    Save a human-readable .ldw.json recipe. Reopen it and export the full file in deterministic order.

The durable artifact

The recipe is the work.

Every step has a name, inputs, and a fixed position. The portable JSON recipe carries a source fingerprint and schema version, so the transformation can be inspected outside the app and replayed without code.

  • Readable, diffable JSON
  • Source identity and format
  • Ordered transformation docket
  • No proprietary cloud project
Recipe / monthly-orders.ldw.json
{
  "schema": "recipe@1",
  "source": "orders.parquet",
  "steps": [
    { "name": "Keep shipped",
      "filter": "status = shipped" },
    { "name": "Clean region",
      "derive": "trim(region)" },
    { "name": "Add territory",
      "join": "region_code" }
  ]
}

A boring privacy policy, by design

Nothing to breach because nothing was sent.

The app has no analytics SDK, account system, content logging, cloud sync, or update pings. It can do its real work offline. A license check sends the token—and never a filename or file content—to Sociobot at most once per day.

Read the plain-language privacy notice →

One tool. One purchase.

A desk license, not a data tax.

The free desk can inspect, transform, save three recipes, reopen them, and export CSV. Pay once for unlimited saved recipes, joins, and JSON Lines export.

$29one-time Buy the full desk Sociobot/Dodo is merchant of record. No subscription.

Current edition · v0.1

Install it on your desk.

Unsigned preview builds are published from this public repository. Checksums are provided for every asset.

macOS

Apple silicon and Intel disk images.

Apple siliconIntelUnsigned: right-click the app, then choose Open.

Windows

Windows 10/11 installer.

Download installerirm https://local-data-workbench.sociobot.in/install.ps1 | iexUnsigned preview; Windows may show a SmartScreen notice.

Linux

AppImage and Debian package.

Download AppImagecurl -fsSL https://local-data-workbench.sociobot.in/install.sh | shThe installer verifies SHA-256 before placing the AppImage in ~/.local/bin.

Checking the latest published checksums…

Verify a download: latest.json · SHA256SUMS

Desk notes

Can it really open multi-gigabyte files?

Yes. CSV and JSON Lines are scanned as streams, Parquet uses its native row iterator, and only a 100-row preview plus bounded profile counters are retained. Regular JSON arrays are capped at 256 MB; use JSON Lines for larger sources.

Does “local” include license verification?

Your file work is fully local. If you add a paid license, the app verifies only that token with Sociobot at most once per day. The free desk never needs a network request.

Which native engine does it use?

The core is written in Rust and uses the Apache Arrow Parquet reader plus the Rust CSV and Serde projects. Notices and licenses ship with the app.