Skip to content
MLabelMLabel

What MLabel is

MLabel is a desktop app for labeling tabular data by hand. You give it two things — a config and a data file — and it shows you one record at a time with a form beside it. When you are done it writes your labels out as a new file.

The one idea

Section titled “The one idea”

Nothing about the data is hard-coded. There is no built-in notion of a “sentiment column” or a “quality rating”. A single .jsonc config file declares:

  • which columns to read and how to display them,
  • what to capture and with which widgets,
  • how it is all laid out,
  • what counts as a finished record,
  • which keys are shortcuts.

Change the config and you have a different labeling tool. No rebuild, no code.

What it is not

Section titled “What it is not”

MLabel does not manage a labeling workforce. There is no server, no queue, no accounts, no inter-rater agreement dashboard. It labels a file. Distributing work across people is done by splitting a file into parts and joining the results back together.

It is also not a spreadsheet. You see one record at a time, in a layout designed for reading, because the thing it is optimized for is the quality of a human judgement rather than throughput of bulk edits.

Local by construction

Section titled “Local by construction”

Everything runs on your machine. The only remote request MLabel is capable of making is its own update check against GitHub Releases, and:

  • it is refused outright at the network layer unless a loaded config permits it,
  • "network": { "updateChecks": false } in your config disables even that,
  • the window itself can never make a network request, in any configuration.

Nothing you load, label, or export is transmitted anywhere. See Network policy for exactly what is allowed and what is denied.

What comes out

Section titled “What comes out”

When you press Done, MLabel writes two files next to your input:

File Contents
*-output.* Every record whose required fields are filled and valid, in output schema.
*-remaining.* Every record that is unfinished, re-emitted in the input schema.

The remaining file is deliberately loadable again as input, so a long job can be picked up later — or handed to someone else. See Exporting.

Next

Section titled “Next”