Skip to content

Under the hood

Never load the whole file

Argonaut doesn't read a file into memory. It memory-maps it and shows a small window onto it, backed by an index built in the background - so the file is on screen as soon as there's enough index to draw the first page, and scrolling anywhere reads only the bytes it needs.

The indexes are small next to the file. Roughly, per gigabyte of file:

View Index
CSV and NDJSON about 0.3 MB
JSON under 1 MB
Text about 2.5 MB

They're small enough to keep, so switching views on the same file - JSON to text and back - reuses the index instead of building it again, unless the file has changed.

Your operating system may show a higher figure for Argonaut: pages of the file it has read count towards its memory. That's the system's file cache, and it's handed back when something else needs it.

A paste is the exception: with no file to map, its bytes stay in memory while it's open. That's why pasting has a 64 MB limit.

Tech stack

Argonaut is a .NET application built on Avalonia, a cross-platform UI framework, so one C# codebase runs natively on Windows, macOS and Linux. The views draw their own rows rather than using stock list controls, which is what lets them scroll through billions of rows smoothly. Updates are delivered by Velopack.

... And Claude! Let's not forget the agent that did the work. I'll take the credit for telling it what to do, and how to do it, and knowing what good looks like. But I'm not going to lie, I didn't type a single line of this app myself. We really are living in the future.

Open source

Argonaut is released under the MIT License. The code, issues and releases are on GitHub.