Talk DAT! (often written TalkDAT) is in public beta for Windows and Mac. Install the Windows build from the Microsoft Store, or take the notarized Mac disk image for Apple silicon on macOS 14 or later.
Use a speech provider you already pay for. Paste your own API key once and Talk DAT! sends your audio to that provider under your account, and nowhere else. The other route stays on the same switch: a downloaded on-device model.
These speech providers work in the current version: Deepgram, OpenAI, ElevenLabs, xAI, Groq, Mistral, AssemblyAI, Google Gemini audio, Smallest.ai, Soniox, OpenRouter, and any OpenAI-compatible endpoint you supply a base URL for. Each exposes that provider's own model and variant choices in Speech settings, so you are not stuck with whichever default we picked.
Text finishing is a separate list with its own key: OpenAI, Anthropic, Gemini, Groq, OpenRouter, or a custom compatible endpoint. A local Ollama formatter can also be prepared, which keeps the writing pass on your machine even when speech goes out.
Not finished yet. Settings also lists providers that are not finished: Google Cloud Speech, Azure Speech, AWS Transcribe, Speechmatics, Cohere, Gladia, Rev AI, NVIDIA, and Alibaba. If you pick one, you will see an error when a session starts, so choose a provider from the list above.
On Windows, supported provider keys go into Windows Credential Manager instead of sitting in a plain configuration file. Environment variables remain supported for development and automation. Provider setup screens link straight to the page where that provider issues keys, so you are not hunting for the right console.
The diagnostics ZIP strips secret-looking fields before it is written, and the in-app feedback form attaches no transcript, recording, API key, dictionary, or device log. A settings backup ZIP is different: it can carry sensitive configuration and text, so treat one as private.
What we have not certified. Talk DAT! has not completed a SOC 2 audit, holds no ISO 27001 certificate, and Knight does not offer a business associate agreement.
The Talk DAT! bar offers two choices, Local and Your key, and shows which one is active while it works. Local runs a downloaded on-device model on your own PC, with NVIDIA Parakeet TDT 0.6B v3 as the default selection. Bring-your-own-key sends the audio to the provider whose key you configured, and Knight is not in that request path. There is no third route: Knight does not forward your audio anywhere.
If Windows reports no network and a local model is ready, the session routes local before capture even starts. Repeated provider failures inside one session can promote local temporarily instead of paying for the same failed round trip again, and a connectivity probe can move the session back without rewriting your saved choice.
What still uses the network. Running speech recognition locally does not make the whole application offline: update checks, optional sign-in, and any bring-your-own-key provider you choose still use the network. The offline dictation page sets out which parts are local.
Open Settings, then Speech. Local-only privacy is on by default, so first turn off Keep speech and finishing on this computer under Privacy. Then choose the provider, paste your key, and pick the model or variant you want from that provider's list. A custom base URL is available for OpenAI-compatible endpoints, and an advanced options field passes route-specific JSON straight through for anything the normal controls do not reach. Batch requests retry network failures and retryable 408, 429, and 5xx responses within bounded limits, so one blip does not cost you a dictation.
Then test it the way you will use it: hold the trigger, say a real sentence, and watch where the text lands. The push-to-talk page covers triggers and delivery; the Windows overview covers the install.
Talk DAT! itself is free. Using your own key does not change that, and no account is needed to dictate. Sign-in, by emailed six-digit code or Sign in with Apple, is optional.
The only bill comes from the provider you chose, at their rates, on their account. We do not print per-minute speech prices here, because they change and a stale number is worse than none: check the provider's current price list before you commit. The Local route carries no per-minute cost from anyone once the model is downloaded, and the comparison page puts the routes side by side.
Straight answers, matching what the product actually does today.
The wired remote speech routes are Deepgram, OpenAI, ElevenLabs, xAI, Groq, Mistral, AssemblyAI, Google Gemini audio, Smallest.ai, Soniox, OpenRouter, and any OpenAI-compatible endpoint you point at with a custom base URL. Text finishing can use OpenAI, Anthropic, Gemini, Groq, OpenRouter, or a custom compatible endpoint. Further adapters appear in settings but are not finished, so treat the wired list as the real one.
On Windows, supported provider keys use Windows Credential Manager rather than a plain configuration file, and environment variables remain available for development and automation. The diagnostics ZIP strips secret-looking fields, and the in-app feedback form attaches no transcript, recording, API key, dictionary, or device log.
Yes. Talk DAT! is free to download and use, with or without your own key, and no account is needed to dictate. The only bill is the one your provider sends for the requests made with your key, at its own rates.
That depends on the provider you choose, and it is billed to your account with them rather than to us. We do not publish per-minute figures here because they change, so check the provider's current price list before you commit. The Local route is the one with no per-minute cost from anyone after the model download.
Yes. Compatible routes accept a user-supplied base URL, model name, and key, and an advanced field passes route-specific JSON options straight through. That covers self-hosted OpenAI-compatible services and gateways the named provider list does not. Provider setup screens also link to the page where each provider issues keys.
No. When you supply your own API key the request goes to your chosen provider and Knight is not in that path, so retention and training terms are governed by that provider. Knight is not in that path either way, and stores nothing of what you dictate.
Install Talk DAT!, paste your key, and keep dictation on an account you control. Talk DAT! itself is free.
