# The voice task flow end to end: speak, mark, merge, hand to the agent

> Every screen relates to the others; this page merges them into one structure. The flow below is not separate images: it is a single journey that tells what a…

Canonical: https://optionos.app/guide/en/voice-task-flow
Published: 2026-07-19
Source: optionos-product:r19

Every screen relates to the others; this page merges them into one structure. The flow below is not separate images: it is a single journey that tells what a person does from start to finish, with the screens linked to each other. Open your terminal, start the speech with ⌥A and walk the flow click by click.

![Desktop with terminals, the capture menu, the speech watch bar and paste targets](https://optionos.app/guide/voice-flow-desktop-start.webp)
*11 screens in one flow: start the speech, watch, copy, record the screen, finish, preview, merge, edit and paste to the agent.*
Interactive scene: https://optionos.app/guide/en/voice-task-flow#voice-task-flow-tour-en

Small notes that live inside the flow: there is also a cursor follower — while speaking you see the recording state next to your cursor; no photo needed, you will run into it. Content is copied as it is pulled; the photos and clipboard update inside the speech as you copy, and the system sees it all as one session.

**Your whole speech archive lives at ⌥H.** The Dictation History panel collects your speeches on one searchable, tagged, starred and badged surface; since audio is saved every time, no speech is lost even if the app breaks. The second flow below walks that panel end to end.

![Dictation History panel: search, live-updating speech, tags, star and the badge popover](https://optionos.app/guide/voice-history-panel.webp)
*From ⌥H to the audio bar: search, star-versus-tag, badges, multi-select, including, recording safety and listening from text.*
Interactive scene: https://optionos.app/guide/en/voice-task-flow#voice-history-tour-en

**The trick for not losing yourself in a long speech: make it count.** I have the AI enumerate the tasks — so my own speech is not forgotten. At the end it hands everything back in one pass. When I paste the output I also make the agent tell me what it understood; every round sharpens itself. The pattern I use:

> Enumerate every task as [n/m] and proceed; change the counter when the goal changes. At the end, hand everything back to me in one pass.