Quick Start

This walkthrough takes you from an empty page to a downloaded CSV without leaving the browser. We'll use GLEIF — 3.36M legal entities — because it's big enough to show that the in-browser engine is real, not a toy.

No install, no signup. The only prerequisite is a modern browser.

Step 1: Open a dataset

Go to /datasets and open GLEIF — Legal Entity Identifiers, or jump straight to /datasets/gleif.

The page renders the dataset's name, license, row count, and source immediately (that part is server-rendered). Below it, the explorer shows:

Loading DuckDB in your browser…

A DuckDB-Wasm engine is booting in your tab and connecting to the Parquet file over HTTP. Once it's ready, the first page of rows appears in the Explore grid. Nothing you do from here leaves your machine — every query runs locally against byte ranges of the remote file.

Step 2: Search

Type NVIDIA into the Search box (top right of the grid). Results narrow as you type. Search matches across the dataset's text columns through a precomputed index, ranked best-match first.

You'll notice the results are broader than "the chip company":

That's expected, and worth internalizing: search is a text match over legal names, not entity resolution. A raw registry contains every fund, subsidiary, and look-alike whose name mentions "NVIDIA." Narrowing to the operating company is your job — which is exactly what the next step is for.

Step 3: Filter

Click a column header to open its menu. You get three tools on one column:

• Sort — Asc / Desc.
• Filter — pick an operator (=, contains, is one of, …) and a value.
• Show values… — a live facet: the distinct values under your current filters, each with a count. Check the ones you want.

Filter country to US, or use Show values… to see the country breakdown first. Each filter you add becomes a chip under the toolbar; click a chip to remove it, or Clear all to reset. For nested AND/OR logic, use the Advanced button.

Every click mutates one underlying query. The grid, the SQL tab, and the copy-paste snippet all reflect it — the filters you build are the SQL you get.

Step 4: Run SQL

Switch to the SQL tab. It mirrors the query you built in Explore, and you can edit it freely. Replace it with a simple aggregate and press Run:

SELECT country, count(*) AS entities
FROM gleif
GROUP BY country
ORDER BY entities DESC
LIMIT 20;

┌─────────┬──────────┐
│ country │ entities │
├─────────┼──────────┤
│ IN      │   373791 │
│ US      │   354980 │
│ IT      │   254319 │
│ DE      │   251466 │
│ GB      │   228583 │
│ …       │      …   │
└─────────┴──────────┘

In the browser the table is named for its slug — gleif, edgar, naics. This full GROUP BY over 3.36M rows runs client-side in a fraction of a second, because DuckDB only reads the country column's byte ranges, not the whole file.

Step 5: Profile the columns

Switch to the Profile tab. It runs SUMMARIZE and shows, per column: type, row count, null percentage, approximate distinct count, and min/max.

Columns that Meridian could type confidently carry a small Finetype badge — lei is tagged LEI, country is a Country Code, initial_registration is an ISO Date. Those aren't guesses printed at page load; they were computed offline by Finetype and travel with the dataset. See Data & provenance.

Step 6: Export CSV

Whenever a result grid has rows, a CSV button appears in the toolbar. Click it to download exactly what's on screen — your current filters, columns, and row limit — as gleif-export.csv. Adjust the Rows input first if you want more than the default 50.

That's the whole loop

Browse → search → filter → SQL → profile → export, and never a server round-trip for the query itself.

When you're ready to run the same data in a notebook, a script, or a scheduled job, it's the same file — see Use anywhere.