Using VoiceFlow

Ask mode

Select text, speak an instruction, and let a language model transform the selection in place — rewrite, summarize, translate, reformat.

Model

openai/gpt-oss-120b

Requires

A text selection

Default paste

replace-selection

How Ask mode works

In Ask mode your voice is an instruction, not content. When recording starts, VoiceFlow captures the frontmost app's selected text. Your spoken instruction is transcribed the same way as dictation, then both are sent to the model, which returns only the transformed text.

Capture selectionJXA / Accessibility
Transcribe speech→ instruction
Transformgpt-oss-120b
Paste resultreplace or insert
Ask pipeline. The captured selection becomes the source text; your transcribed speech becomes the instruction.

A selection is required

If no text is selected when recording ends, Ask mode reports that it needs a selection and nothing is pasted. The instruction is also required — silence produces no transform.

Paste behavior

The askPasteBehavior setting controls where the result goes:

ValueBehaviorUnder the hood
replace-selectiondefaultPastes over the current selection, replacing it with the transformed text.
paste-at-cursorinsertCollapses the selection to its end (a simulated Right-arrow) before pasting, so the original text is kept and the result is inserted after it.

Instruction isolation

The model is told explicitly that the selected text is source material, never a prompt. It applies only your spoken instruction, preserves factual meaning unless you ask otherwise, and returns just the transformed text — no commentary, quotes, or code fences.

  • “Make this more formal” → a polished rewrite.
  • “Turn this into three bullet points” → a reformatted list.
  • “Translate to Spanish” → a translation of the selection.
  • “Fix the grammar” → a corrected version, same meaning.