FlowVoice charges nothing and has no server of its own: it talks directly to the AI provider you choose. All it takes is a free API key. Or none, if you prefer to use it offline.
If you only want to dictate today: create a key at Groq and you're done. It transcribes your voice and polishes the text, it is very fast, and the free plan covers a whole day of normal use.
The other keys are optional. Having more than one provider works as a backup: if one fails or hits its limit, FlowVoice moves on to the next on its own.
No need to memorize anythingInside the app, each provider has this same step-by-step, a button that opens the right page and a Test button that confirms the key works.
Each one takes a few minutes. Start with Groq; the others serve as backup or add features.
Voice and text. Very fast transcription and polishing.
gsk_). It is shown only once.Voice and text. Google's AI, with a generous free plan.
AIza) and paste it into the app.Text only: polishing, translation and questions. Gives access to dozens of models, and the ones ending in :free cost nothing.
sk-or-) and paste it into the app.Voice and text. European AI, with a very good transcription model.
Optional. Gives the voice assistant (Ctrl Shift U) access to the internet: it searches and answers, citing its sources.
tvly-) already shows up on the dashboard.In the Windows Credential Manager. Only in rare cases, when the vault is unavailable or rejects a very long key, it goes into a local file in your user folder. It only leaves your PC to go to the provider itself, with each request.
You can add several keys for the same provider. When one hits its limit or fails, FlowVoice moves to the next one automatically.
Providers swap models often. FlowVoice notices when a model goes offline, picks a replacement from the provider's current list, tests it and starts using it on its own. It also checks this once a day, in the background. You don't have to touch anything.
Every free plan has a ceiling: so many requests per minute, so many minutes of audio per day. For everyday dictation, it's plenty. The exact numbers change from time to time, so check each provider's page.
| Provider | The limit applies to | Another key on the same account |
|---|---|---|
| Groq | The whole account (organization) | Doesn't raise the quota |
| Gemini | The Google project, not the key | Doesn't raise it, if it's in the same project |
| OpenRouter | Requests per day on :free models; goes up if the account has purchased credits | Doesn't raise it |
| Tavily | Credits per month; each search costs 1 or 2 | Doesn't raise it |
Several keys on the same account don't add upOn Groq, the limit applies to the whole account: five keys on the same account give exactly the same quota as one. What really helps is having different providers as backup, for example Groq and Gemini.
Prefer not to use any cloud? In Settings → Transcription, choose 100% local and download a speech recognition model (Whisper). Transcription runs on your processor and the audio never leaves your PC.
The app shows which models are recommended for the language you speak. The numbers below are approximate:
| Model | Download | Memory in use | Best for |
|---|---|---|---|
| Tiny | 31 MB | ~150 MB | Very weak PCs; makes more mistakes |
| Base | 57 MB | ~300 MB | Fast; also used for the live preview |
| Small | 181 MB | ~600 MB | Good balance for the most widely spoken languages |
| Distil-Whisper pt-BR pt-BR | 513 MB | ~950 MB | Tuned for Brazilian Portuguese |
| Large v3 Turbo | 547 MB | ~1 GB | Maximum accuracy in any language |