Windows · checked 11 September 2026

Voice control for AI agents on Windows: what actually ships today

When you run several coding agents at once, the keyboard stops being the bottleneck. Waiting does. One session is thinking, another finished two minutes ago, a third is asking you something — and you read three windows to find out which one deserves you.

Voice fixes half of that on Windows today. This page is about the half it doesn’t, and it names what SpeakApp does and does not do, app by app.

Talking to your agent is a solved problem

If all you need is to speak instead of type, you have three good options on Windows, and we would rather say so than pretend otherwise. Wispr Flow, Willow and Superwhisper all ship Windows builds — their own home pages say it, and Superwhisper even runs a dedicated Windows page. Dictation into a prompt box is no longer where anyone wins.

The half nobody ships on Windows: the way back

Exactly one vendor in this category built the full agent loop — the floating window that surfaces when an agent calls you, so you can answer without hunting for the right window. That is Superwhisper, and their own page for it states the boundary in three words: “Requires Superwhisper for macOS.” Their Windows page offers dictation, and only dictation.

The other two don’t offer the return trip on any platform. On Wispr Flow’s Claude page and Willow’s — 1,037 and 1,240 visible words, read on 11 September 2026 — the words listen, read back, out loud, hear and text-to-speech appear zero times between them. These are dictation tools, and they say so clearly.

So on Windows the outbound trip is crowded and the return trip is empty. That is the part SpeakApp was built for.

What SpeakApp does while your agents work

SpeakApp keeps the source of every answer and reads it back to you at your pace — in full, or as a summary — so you can keep your eyes on what you were doing. Answering in the session and supervising which one needs you are separate capabilities with their own reach, and the table below says which is which, line by line.

Dictation is independent of that table: it types wherever your cursor is, in any Windows application — an editor, a chat box, a terminal window. Two modes: Fast, where text appears as you speak, and Precise, for maximum quality. It covers 40 languages, and you can dictate in one and have it written in another. Terms it gets wrong once — library names, variables, commands — you correct once, and the personal dictionary writes them correctly from then on.

Where it works, and the one place it doesn’t

Dictation is a Windows-level capability: it types wherever your cursor is, in any application — an editor, a chat box, a terminal window. Listening to an answer and supervising a session work on the desktop assistants and, through the browser extension, on their web versions.

The exception is a session whose execution is offloaded to the cloud. Ask Claude to run a task remotely rather than locally and SpeakApp will not read it back — and answering by voice falls with it, because there is nothing left to aim at.

The reason is worth one sentence, because it tells you where the line will always be: reading takes the content at the source — the conversation trace on disk for desktop apps, the page itself for the browser. A remote session leaves its results on your disk but never its dialogue. There is nothing to read, and nothing announces it: it simply stays quiet.

The list app by app lives on the features page — one table, one place. This page does not carry a second copy of it.