Speech into words.
Share a reel, a video or a voice memo. Get back a clean, timestamped transcript you can search, copy or hand straight to Claude.

What it does
Nov 2026- 01Share anything with a voice in itAn Instagram reel, a TikTok, a YouTube video, a voice memo or a file. Jot fetches it, listens and writes it down.
- 02Timestamps you can tapEvery line keeps its time, so you can jump straight to the bit you want.
- 03Search everythingOne search over every transcript you've ever jotted, with Links, Voice and Files filters.
- 04Send to ClaudeOne button hands the transcript to the Claude app, as text or as a file.
- 05Record in the appTap the mic, talk, stop. It keeps recording with the screen locked.
- 06iPhone and MacThe library syncs through iCloud. On the Mac, Jot lives in the menu bar and does the heavy lifting.
Screens
The library
A transcript, in dark mode
Listening
Settings, with tiny Jot home screens for each look
The mark
A j that's also a quote markA closing quote mark is a round head with a tail curling down and to the left. So is a lowercase j. Jot's j is drawn as a long comma: the head flows straight into the tail in one outline, tilted 10°. One reads as a j; two side by side make ”, the app icon.
1From j to ” and backSix glyphs from most j to most quote mark. The long comma was the only one that read clearly as both.
2One outline, tiltedThe head runs into the tail with no seam, tried at 0°, 6°, 10° and 14°. At 10° it reads as ” and still stands up like a j.
3SpacingThe t's crossbar reaches towards the o, so the j is pulled in to match: 6.5 units at both closest points.
4TypeFour pairings tried on the real screens. Onest and Martian Mono shipped: a touch technical, and a mono that's wide and square, like a tool.
Signature
Pull to refresh
Pull down the list and the j splits into an opening “ and a closing ” that spread apart. Let go and a waveform plays between them, turns into words saying what the pull found, and the quotes snap shut. Speech becoming words, which is the whole app in one gesture.

How it works
- Fetching, logged outLinks are fetched with yt-dlp with no cookies and no account, so Jot only ever sees what a logged-out visitor sees.
- Three places to transcribeOn the iPhone with whisper.cpp, in the cloud on a GitHub Actions runner with faster-whisper (about 40 seconds for a 26-second reel), or on the Mac from the menu bar, which pushes the result to the phone itself.
- A proofread and a headlineAn optional pass with Claude Haiku fixes misheard names and keeps what Whisper actually heard alongside. Each transcript gets a short title written on device by Apple's Foundation Models.
Design
Onest · Martian MonoAa0:42
#0A1620#18CFFF#5E6F7C#F4F7F9
- One accent: cyan, only for things you act on and for timestamps. Everything else is ink and grey.
- Onest for all the text, Martian Mono for times and durations only.
- No containers: rows under hairlines, Liquid Glass only on the controls.
- Thinking orbs instead of spinners: Listening while it transcribes, Searching while it fetches.
