Talk DAT! (often written TalkDAT) is AI dictation for Windows and Mac by Knight AI+AV, in public beta. Install Windows from the Microsoft Store.
Windows first, runs on the PC. Talk DAT! is free to download and use. Speech is recognised on this PC with a downloaded local model, and no account is needed to dictate.
See your words while you are still holding the key. On a local model, with the switch on, the Talk DAT! bar shows the text so far about 1.2 s behind your voice, measured on a GPU and on six CPU cores alike. Nothing leaves the machine, and the text you get when you let go is the same either way.
People looking for a Wispr Flow alternative on Windows usually want one of three things: dictation that works offline, a free app, or dictation that runs on their own computer. Talk DAT! does all three.
It needs an internet connection. Wispr Flow processes speech in the cloud, so it does not work offline, for example on a plane or on a locked-down network.
Windows on ARM is not supported, per their documentation. A growing share of new Windows laptops are ARM machines, and as of September 2026 that vendor's own docs say the platform is not supported.
It has paid plans. Wispr Flow sells paid plans; check their site for current pricing. Talk DAT! is free.
Speech can travel two ways, and you pick: Local, where recognition stays on this PC; or bring your own key, using a provider you already pay for. On Local you download an on-device model once. The manager compares, downloads, selects, and deletes models, and downloads are verified and resumable. If Windows reports no network and a local model is ready, the session routes local before capture starts.
Two writing styles sit one click apart in the Talk DAT! menu. Chill tidies your words and keeps your tone; Executive makes them more formal and structured. You choose per dictation.
"ok so uh can we push the tuesday review to thursday, i still need the numbers from finance, and remind me to send bola the deck"
Can we push the Tuesday review to Thursday? I am still waiting on the numbers from finance. Also, please remind me to send Bola the deck.
Illustrative text transformation. Results vary by route, microphone, settings, and speech.
If the paste fails, you still have your words. Delivery is a verified clipboard write and paste, with Shift+Insert and simulated typing as fallbacks. There is no permanent audio archive; a limited recovery buffer is kept on this PC, so recent sessions can be replayed or transcribed again from History.
Talk DAT! entries describe the shipping public beta. Wispr Flow entries state only what that vendor publishes about itself, read on 2 September 2026.
| Capability | Talk DAT! | Wispr Flow |
|---|---|---|
| Where speech is recognised | Two routes: on this PC on Local, or a provider you supply a key for. | Cloud processing, per their site as of September 2026. |
| Works with no connection | Yes, on Local, after a one-time model download. | Not applicable to a cloud only service. |
| Price | Free. | Paid plans; check their site for current pricing. |
| Windows on ARM | No separate ARM claim published. Check the Store listing on the machine itself. | Not supported, per their documentation as of September 2026. |
| Writing styles | Chill and Executive, one click apart in the Talk DAT! menu. | Not confirmed |
| Triggers | Hold Ctrl+Win, Ctrl+Win+Space hands-free, Ctrl+Win+Esc panic stop. Rebindable, plus mouse side buttons and Xbox controls. | Not confirmed |
| Compliance paperwork | None claimed: no SOC 2, no ISO 27001, no BAA. Stated on the security page. | Not confirmed |
"Not confirmed" means we could not verify the capability from public information and would rather say so than guess. Competitor names belong to their respective owners; this page is not affiliated with or endorsed by Wispr or Microsoft. Their claims were read from their own public pages on 2 September 2026 and can change without notice.
Nothing to download. A cloud service asks nothing of your disk or processor. Local needs a model on disk first, and that download sits between installing and dictating.
Maturity. Talk DAT! is a public beta, and the terms say features may be added, altered, or withdrawn. If you need software that will not change, you may want to wait.
Phones. Talk DAT! runs on Windows and Mac only, and there is no phone version. If you need to dictate on a phone, Talk DAT! does not do that.
One. Install from the Microsoft Store, the recommended Windows route, which handles updates for you.
Two. Open the model manager and download one on-device model. This step is what makes offline dictation work.
Three. Hold Ctrl+Win wherever you were already writing, say a sentence, release. If the default chord collides, rebind it in Settings.
Four. Teach it words it does not know yet: sounds-like aliases, plus four packs, Military, Medical, Legal, and Aviation, totaling nearly 1,400 terms.
Five. Keep your old tool for a week. Talk DAT! is free, so there is nothing to cancel if it does not suit you.
Talk DAT! is free to download and use. No account is needed to dictate; sign-in is optional.
The one cost you might see is your own provider's. On the bring-your-own-key route, you add your own key for a supported provider, the request goes from your machine straight to that provider under your account, and that provider bills you at its rates. Knight is not in the path and adds nothing. Full wording is in the product terms.
Straight answers, matching what the product actually does today.
Talk DAT! has a Local route: download an on-device model once and speech recognition stays on this PC. One catch: offline dictation starts after that download, not out of the installer, and Local covers speech recognition rather than every feature.
Talk DAT! is free to download and use. Wispr Flow has paid plans; check their site for current pricing.
Wispr Flow's own documentation says Windows on ARM is not supported, as of September 2026. Talk DAT! publishes for Windows 10 and 11 through the Microsoft Store and makes no separate ARM claim, so check that listing on the machine.
No. No account is needed to dictate. Signing in with an emailed six-digit code or Sign in with Apple is optional.
On Local, recognition runs on your own machine and no provider is involved at all. On bring your own key, your chosen provider receives the request and Knight is not in that path. Your audio never goes through Knight.
It is a public beta on Windows and Mac, and features may be added, altered, or withdrawn. Knight has not completed a SOC 2 audit, holds no ISO 27001 certification, and offers no Business Associate Agreement.
Install, download one model, and dictate with the Wi-Fi switched off. It is free.
