JotDownload for Mac

Jot writes every dictation, capture and meeting as a folder on the Mac, and puts this guide at the top of the recordings folder, in ~/Documents/Jot/README.md. It's here so an agent can read it before it's handed a recording.

Reading these recordings

This folder holds recordings made by jot, a recorder that runs on this computer. Each recording is a folder of its own, and the file to read in it is recording.xml. This page says how to read it well.

The folders

  • Meetings/ holds conversations: a call, or people talking in a room.
  • Captures/ holds notes spoken aloud, often with screenshots.
  • Dictations/ holds words spoken to be typed somewhere else.

A recording’s folder is named for the moment it started, such as 20260918_140000. In each kind’s folder, latest points at the newest one. index.db is jot’s search index; there is no need to read it.

Inside a recording

  • recording.xml is the whole story: what was said, who said it and when, and what was on screen.
  • events.jsonl is jot’s own log, with more detail than most readers need.
  • settings.toml says how the recording was made.
  • speakers.json, when it is there, holds the names a person gave the voices.
  • shots/ holds the screenshots, and audio/ the sound, when it was kept.
  • Capture data 2026-11-03 at 20.36.12.zip or Meeting data …, when it is there, is a copy of the words, timings and shots to hand to someone. There is no need to read it.

A small example

<recording format="1" type="meeting" started_at="2026-09-18T14:00:00+01:00" status="finished">
  <sources>
    <source id="mic" type="microphone"/>
    <source id="sys" type="system_audio"/>
    <source id="screen" type="screen"/>
  </sources>
  <context>
    <calendar starts_at="2026-09-18T14:00:00+01:00">Weekly sync about the audio pipeline</calendar>
  </context>
  <timeline>
    <speech source="sys" predicted_speaker="sys.a" voice="7496a" start="72.4" end="75.9">Um, shall we look at the plan?</speech>
    <speech source="mic" predicted_speaker="mic.a" start="76.2" end="78.0">Sounds good.</speech>
    <shot source="screen" start="80.0" file="shots/01.png" caption="The plan"/>
  </timeline>
</recording>

Here mic is the person recording, and sys is the call: everyone heard through the computer’s speakers.

Reading it well

  • Speakers are a guess within one recording. predicted_speaker tells voices apart, as sys.a and sys.b, but sys.a in one recording has nothing to do with sys.a in another.
  • The microphone and the call sound different. mic.a and sys.a are never matched, and the same person heard both ways will look like two people.
  • voice is a hint across recordings. It appears on a meeting’s speech once that voice has spoken for about a minute. Codes that are close, digit by digit, may be the same voice heard the same way; they never prove it.
  • dictated="true" was dictated, not said to the room. The person held a key to type those words somewhere else during the meeting.
  • <context> is what jot was told, not what was said. It holds notes and the person’s Your words, given to the speech model to help it spell names.
  • The words are as heard. They keep the ums and ahs; jot’s exports leave them out.
  • Times are seconds from the start. start="72.4" is 72.4 seconds after started_at. In pasted text, [image 2] marks where the second screenshot was taken.
  • Names in speakers.json are a person’s own labels, given by hand after the recording, such as {"sys.a": "Andrej"}.
  • A recording may still be going. While status says recording, the file keeps growing.

To share a recording, copy its whole folder.

Drawings on screenshots

A screenshot someone drew on has three files. shots/01.png is the picture as they saw it, drawing and all. shots/01-clean.png is the screen alone, and shots/01.svg is the drawing alone, the same size, so its points are the clean picture’s pixels. The shot element names them in clean and drawing.

Each thing drawn is an ink element in recording.xml and a group with the same id in the SVG. tool says what drew it: highlighter, box, arrow or number. A highlighter stroke also has shape, jot’s guess at what it is: a loop round something, an arrow (with its arrowhead), an underline, a line or a scribble. A number is listed as a mark, and its number is the one the person says, as in “mark 2”.

To tidy the drawing up, draw over the clean picture rather than the flattened one. https://capturewithjot.com/workflows/clean-up-capture shows how to turn a capture into feedback a person can send.

When a name is spelled wrongly

The speech model can get a name or a term wrong, such as a colleague’s name or a product’s. When you see one, it helps to suggest the person adds the right spelling to Your words in Jot’s window. Jot gives those words to the speech model in every later dictation and meeting, so they come out spelled that way from then on. Recordings already made keep the spelling they have.

jot writes this page and updates it when its format changes.