Run SQL With the Network Disconnected
The strongest argument for a tool that does not upload your files is not a privacy policy. It is disconnecting the machine and carrying on working. This guide shows you how to do that yourself, in about two minutes.
The Test
- Open the app and let it finish loading. This page load is the only network activity in the whole exercise.
- Disconnect. Turn off Wi-Fi, unplug the cable, or on Linux run
sudo ip link set eth0 down. Confirm it is really down —ping -c 2 1.1.1.1should fail with Network is unreachable. - Now add your files. Drag them onto the File Select panel. They load and become tables.
- Query them. In English or in SQL, it makes no difference.
- Export the result. CSV, Excel, JSON or Parquet. The file is written in the tab and saved straight to your downloads folder.
All four of those steps work with no route to anywhere. Nothing was uploaded, because there was nowhere to upload it to.
Load the page, then pull the cable.
Open the tool →Why It Works
The parsers for every supported format, the SQL engine, the plain-English layer and the export writers are all part of the page. Once the browser has them, the work happens in the tab. There is no server component that a query is sent to, so there is nothing for a disconnection to break.
That is also why the file size limits are what they are — the ceiling is your own machine's memory, not a server's patience.
What Still Works, and What Does Not
| Works offline | Needs a connection |
|---|---|
| Adding files, in every supported format | The first load of the page |
| Running SQL, including joins across files | Reloading the page, or opening it in a new tab |
| The plain-English box | |
| Exporting to CSV, Excel, JSON and Parquet | |
| Reading Parquet and Avro, which use WebAssembly |
The one real caveat: reloading the tab while disconnected gives you the browser's offline page, because the page itself has to come from somewhere. Keep the tab open and you can work indefinitely. If you need it on a machine that has no connection at all, see the air-gapped guide.
A Worked Example
Three files — departments.csv, employees.csv and projects.csv — loaded after the network was taken down. They become D, E and P.
Asking total hours by department:
SELECT department, SUM(hours) AS total_hours FROM P GROUP BY department
Then how many employees by department:
SELECT department, COUNT(*) AS count_employees FROM E GROUP BY department
Both answered, both exported to Excel, with the interface still down and a request to this very site still failing.
Reasonable Questions
Does something get sent later, when I reconnect?
No. Nothing is queued. There is no storage of your file contents and no background sync — reconnecting changes nothing except that a page reload would now succeed. You can watch this too: reconnect with the network panel open and nothing goes out.
What about the Cloudflare requests I saw earlier?
When the page first loads you will see requests to /cdn-cgi/…. That is the host running its bot check, before any file is added, and it is part of serving the page rather than anything to do with your data. The verification guide covers this and two further checks.
Can I use this on a train, or on a plane?
Yes, and it is one of the more common uses. Load the page before you lose signal.
How long can the tab stay open?
As long as you like. The loaded files live in the tab's memory; closing or reloading it clears them, and you add them again. Nothing is cached to disk.
Is this the same as a progressive web app?
No. There is no service worker and nothing is installed. The tab simply does not need the network once it has the page — which is a weaker claim than a PWA makes, and a more honest one.