EN
Back to the archive

The archive · Developer & Business Tools · Product decision · 2023

Compress: corpus-ranked typing shortcuts drew 815 HN points and saved their author time

eschluntz parsed six months of Slack messages and ranked shortcuts by characters-saved × frequency; the 2023-08-30 Show HN drew 815 points.

Compress (eschluntz/compress)

The betThat good shortcuts are a data problem: rank your corpus by characters saved × frequency, and tiny frequent words like 'the → t' beat the long phrases you type most.Live

What the business is

Compress is an open-source Linux utility that mines a corpus of your own writing — including a Slack data export — for phrases you repeat, ranks them by characters-saved × frequency, and generates Autokey config files that expand short abbreviations while you type.

How it started

Developer eschluntz had been using the tool for about a year before posting it to Hacker News on 2023-08-30: he parsed six months of his Slack messages, found the phrases he typed most, and generated keyboard shortcuts for them, mainly for email and Slack. Earlier options shaped the design — Autokey's simulated key input proved more reliable than browser extensions that clashed with sites' own JavaScript, and expansion was restricted to Chrome because one-letter abbreviations collided with shell commands and variable names in a terminal or editor.

What happened

The ranking produced the project's core surprise: very short, frequent words such as 'the → t' saved far more characters overall than longer phrases he used often, such as 'what do you think → wdytk', because characters-saved × frequency favors the tiny words. He refined the output into memorable abbreviation families — plurals gain an 's', phrases inherit their words' abbreviations ('the robot → tr'), similar words get similar forms (some/someone/something → s/sn/st) — as a manual step on top of the machine's top-200 list, and he kept the curated YAML in git to sync across his computers.

How it ended up

No company, funding or revenue appears anywhere in the material: Compress stayed a personal open-source tool rather than a startup. What the author did report in the 2023-08-30 thread was that it was one of the rare side projects that actually saved him more time than he put in, and the repository remained public with 1.4k stars and 84 commits at the 2026-09-04 crawl.

Background

Compress is an open-source Linux utility that mines a corpus of your own writing for phrases you repeat and turns them into typing shortcuts. Developer eschluntz had been using it for about a year before the Show HN on 2023-08-30: he parsed six months of his Slack messages, found the phrases he typed most, and generated keyboard shortcuts for them, mainly for email and Slack. Autokey's simulated key input beat browser extensions because it did not clash with sites' own JavaScript; configs ran only while Chrome was focused, since one-letter abbreviations collided with shell commands elsewhere.

The founding bet was that the right abbreviations are a data problem, not a taste problem: the code finds common n-grams and ranks them by characters-saved × frequency. The ranking delivered a genuine surprise — very short, frequent words such as 'the → t' topped the list over longer phrases the author typed often, such as 'what do you think → wdytk'. He layered memorable-abbreviation heuristics on top, building families where plurals gain an 's', phrases inherit their words' abbreviations, and similar words get similar forms, keeping the curated list in git to sync across computers.

The thread on 2023-08-30 scored 815 points with 313 comments, and the author noted in it that the project was one of the rare side projects that saved him more time than he put in. The repository showed 1.4k stars, 30 forks and 84 commits in the 2026-09-04 crawl and was still live on 2026-09-05. Nothing in the repository or the thread documents a company, funding, revenue or commercial intent — Compress remained a solo, open-source productivity tool rather than a startup.

What has to be true

  • The bet is falsifiable and data-driven: rank candidates by characters-saved × frequency, and the corpus produced a counterintuitive winner — 'the → t' beat the long phrases that felt repetitive.
  • Concrete dated traction: the 2023-08-30 Show HN (item 37326870) drew 815 points and 313 comments, and the repo showed 1.4k stars with 84 commits in the 2026-09-04 crawl, still live a day later.
  • It is a real usage arc: the author parsed six months of Slack history, used the shortcuts for about a year, and said the project saved him more time than he invested.

What can be applied

Before optimizing a habit, measure its actual frequencies: the highest-yield typing shortcuts came from tiny common words, not the long phrases that felt repetitive — data beat intuition.

Aftermath

As of 2026-09-05 Compress remains a public open-source repository — 1.4k stars, 30 forks, 9 watchers and 84 commits — but shows no sign of becoming a business. The README still frames it as a personal tool for email and Slack, pinned to Linux/X11 Autokey, and neither the repository nor the HN thread documents a company, funding, revenue or further product work. The idea it demonstrated, ranking text-expansion snippets from your own corpus instead of hand-picking them, stayed a solo productivity hack rather than a product or a startup.

Sources

spotted an error? The archive wants to know.

Your turn

You just read one. Describe what you are building, and see who is betting on the same thing.

Free account · 3 free questions · no card

Related cases