Splay

2026 · Fork, design, build

Simplify, then add lightness. How light can a voice dictation app for Mac and iOS be, and how native can it feel?

In my words

Splay (alpha) for Mac and iOS began as a fork of MacParakeet, a great open-source app for local voice transcription.

I love local transcription. It’s fast, it’s safe, and I genuinely love it when my hardware is put to work. Since I speak into my machine a lot, I was tempted to give my own version a go.

In the spirit of the old Lotus philosophy, “Simplify, then add lightness”, I set out to build an app that is as light, and feels as integrated into macOS, as the Dynamic Island does on the iPhone.

Google MeetFileEditWed 14:42

Dormant

The Splay island, animated. It changes size and colour as it records, transcribes and finishes. Click a state to hold it.

How it went — 2026

Jun 1Fork
Upstream. MacParakeet by Daniel Moon, the app Splay started from.
Jun 2First skin
Pick a green. First try: new colours on the same window.
Jun 3Iteration
A quieter window. Two calmer layouts. Still a window.
Jun 3Turn
The island. A small pill at the bottom of the screen. The big window stays closed.
Jun 3No window
An ambient app. No main window at all. Transcripts are plain files in Finder.
Aug 6Named
Splay. The island moves up to the notch and turns black and violet. It shows status and has no buttons.
Aug 7Handoff
Nine states. Every state drawn to exact size, ready to build. Testing picked the fn key.
Aug 8Theming
Seven accents. Seven colour themes. Red, amber and green keep their meaning in all of them.
Aug 22Shipped
0.1.0. Signed, notarized and in daily use. No screenshot of it yet.

From the repomacparakeet_MC / docs/fork-product-model.md

The fork’s north star. Where it disagrees with upstream, this document wins.

Why this fork exists

Upstream MacParakeet is a capable but broad app: dictation + file + YouTube + meetings + Transforms + calendar + Discover. For this owner’s actual use it carries weight that doesn’t earn its place. The fork keeps the engine, sheds the surface area, and re-themes.

The owner’s real workflow

Everything is capture → transcript document → export. The upstream split between “dictation” and “meeting recording” is artificial for this owner — they are the same intent (record a conversation) with a different audio source.

Capture model (the spine)

One key — Fn — two gestures, distinguished by audio source. Single tap: mic only. Double tap: mic + system audio. Both produce a transcript document. No push-to-talk.

A transcript is never polished. A summary can be.

Flagging preserves doubt; polishing hides it.

Fork point: 2026-06-01
Commits: 50 · 11 threads
Diff: +15,103 / −137,760
State: 0.1.0 notarized · alpha
Engine: Daniel Moon · MacParakeet

← Back to Products

This site is under construction. Where the text sounds like Claude, it was written by Claude. I'll rewrite it by hand. Code of honor.