Summaries and translation

The four summary modes, how often summaries update (the window), output language, glossary and context hints, live translation.

When the summary updates

The transcript is split into “segments” by punctuation and length. When enough segments have accumulated, the most recent ones are handed to the summary engine and the summary is rewritten. The summary pane header controls this:

Control	What it does	Default
Mode	The shape of the summary (below)	Short summary
Window	How many recent segments go into one summary	3 segments
Output language	The language the summary is written in	The interface language (your choice is remembered)

A larger window updates less often but reads better as a unit. Use a small window for fast-moving meetings, a larger one for lectures where one person talks for a long time.

The four modes

Mode	Output	Good for
Short summary	One or two sentences	Following along during a meeting; a display for an audience
Conversational summary	Slightly longer, keeping the spoken phrasing	Reading back later; keeping nuance
Keywords	A list of terms	An index of what came up; search handles
Translation	The text in the output language	Following a foreign-language meeting in your own

The “Edit summary prompt” button next to the mode picker opens the prompt editor, where you can rewrite the built-in modes or add your own.

Output language

Pick the output language from the language menu in the summary pane. If it differs from the transcription language, the summary is also a translation — for example, an English meeting followed through a Japanese short summary.

Glossary and context hints

Open the glossary from the session in the sidebar. Each session has its own.

Context hint (“About this conversation”) — one to three lines on what the conversation is, who’s in it, and the tone. Passed to the prompt as background for summaries and translation. It is a place to describe the conversation; for formatting instructions (“write it as minutes”), edit the summary prompt instead.

Glossary — one term per line: names, technical terms, abbreviations you want written correctly. The format is correct spelling(how it's said).

Kubernetes(k8s,kube)
Tatsuya(Tatchan)
CI

Left of the parentheses is what appears on screen; inside is how it’s actually said (readings, nicknames, variants; comma-separated)
No commas outside the parentheses
English words and abbreviations can be listed on their own (case and width are handled automatically)

The glossary feeds three things: recognition hints for transcription (only on engines that support them — Apple Speech does, WhisperKit doesn’t), spelling normalization after commit (every engine), and the summary/translation prompt. Only the first 30 lines reach the prompt, so put important terms at the top. Around 100–200 lines is a sensible ceiling for recognition hints.

Live translation (second stream) Pro

Splits the summary pane in two: a summary in the first, a translation into another language in the second, both live. Available with Pro.

Pop-out windows Pro

The “↗” button in a summary pane’s header moves Stream 1 or Stream 2 into its own window, or shows both side by side in one (Stream 1 + 2). It’s meant for putting just the summary and translation on a meeting-room display or a lecture-hall projector — the transcript itself is not shown.

The button at the top right of the window sets the font, text size, and which display to use; the settings are saved per window. Full screen hides every button and menu, and the header appears only when the pointer is at the top of the window. The pop-out doesn’t close on its own when the main window comes to the front.

The machine-generated notice

For the first three minutes of a recording or import, a notice that the text is machine-generated and may contain errors stays on screen. After that, the interval (1 / 3 / 5 / 10 minutes, always, never) is set under Settings > General > Display. The wording can be edited per language.