What a profile holds
Open Settings → Profiles. TypeWhisper starts with Email · Professional, Calendar Event, Note, Message, Cleaned Text and, unless your phone is set to English, Translation · English. Tap a profile to edit it, or tap Add profile.
- Name (optional): without a name, the profile is named after its settings, for example “Email · Professional”.
- Format: what the writing service makes of your dictation.
- Tone (Message, Email, Note): Automatic, Casual, Friendly or Professional.
- Length (Email, Note, Summary): Short, Medium or Detailed.
- Translate into (Translation): one of 20 target languages.
- Custom instruction (Custom): your own instruction, for example “Translate into English and keep it short.”
- Use automatically in these apps and Use automatically on these websites: see below.
- Speech recognition and Language: As in settings, or a fixed engine and language for this profile. The model comes from that engine’s model page.

Formats
Dictation in the profile menu means no profile: the text is used as recognized, after text clean-up, dictionary and snippets. Plain dictation never goes to a writing service.
- Cleaned Text: fixes punctuation, grammar, casing and spacing, and keeps your wording.
- Message: the exact message to paste into a chat. If you dictate “tell Alex that…”, it writes the message to Alex. No greeting or emoji unless you dictated one.
- Email: a complete email with greeting, body and closing, and the subject separately.
- Note: a short, readable note that keeps the important details.
- Calendar Event: a calendar draft with name, time, place and notes.
- Checklist: one item per line, each starting with “•”.
- Summary: a short, accurate summary.
- Translation: a faithful translation into the target language, without shortening.
- Custom: follows your custom instruction.
Note
Premium
The Email and Calendar Event formats need Premium, as does a profile that uses Nemotron Live. Such profiles show a lock without Premium. Dictating without a profile stays free.
Use a profile while you dictate
- Record screen: tap the “Profile: …” chip below the engine name and choose Dictation or a profile. It stays active until you change it.
- Keyboard: tap the profile chip. Your choice there also becomes the active profile. Calendar Event is not offered in the keyboard, because its result goes to the calendar app.
- Apps: with Use automatically in these apps, the keyboard and the dictation bubble switch to this profile in the apps you pick. Without assigned apps, they use the active profile.
- Websites: with Use automatically on these websites, add domains such as example.com. You can paste a full URL; only the domain is kept and subdomains are included. Websites are recognized in Chrome, Edge, Brave, Opera, Vivaldi and DuckDuckGo while the dictation bubble is on. If apps are chosen too, both must match.
- Order: a profile for the app and the website wins over one for the website alone, which wins over one for the app alone. Within each step, the order in the profile list decides, which is the order you created the profiles in.
- Text you already wrote: select it in any app, choose TypeWhisper in the selection menu, pick a profile under Rewrite with a profile and tap Replace (Copy for read-only text). In the TypeWhisper keyboard, select text and tap Rewrite.
Note
Password fields
In password fields, the keyboard and the bubble insert plain dictation. The text is never sent to a writing service.
Connect an AI writing service
Open Settings → AI writing and tap Set up step by step (once the service is set up, the button reads Run the setup again). Every step saves right away.
Service
Choose OpenAI, OpenRouter, Groq, Meta, Anthropic (Claude), Google Gemini, xAI (Grok) or Own server for an OpenAI-compatible server such as Ollama or LM Studio.
Access
Paste your API key. The link Create a key at… opens the provider’s key page where one is known. The key is stored encrypted on your phone and is only sent to that service. For Own server, enter the server address including /v1: HTTPS, or HTTP on your local network. A key is only needed if your server asks for one.
Model
TypeWhisper loads the service’s model list and preselects a recommended model. You can change it any time.
Consent
Allow sending dictations with a writing profile to the service. Without your consent, TypeWhisper sends nothing there.
Test
TypeWhisper checks that the service answers with your settings. Later you can run Test connection on the AI writing page.
- The AI writing page also lets you change the service, the model (Load models fetches the list) and the switch Send dictations to … for rewriting.
- Switching to another service resets the model and your consent; a new server address resets your consent.
- Groq, OpenAI, Meta and Own server share one key between speech recognition and AI writing; Own server also shares its Server URL. All keys are also listed under Settings → Cloud providers.
Note
What a writing service receives
Only the text of dictations with a writing profile and text you choose to rewrite, after clean-up and dictionary corrections, plus the profile’s instructions and, for calendar events, the current date, time and time zone. It never receives your audio, your history or your dictionary. If the service is not set up or fails, TypeWhisper inserts the plain dictation and tells you why.

Email and calendar drafts
- Email: the result shows “Subject: …” above the body. Email draft, on the Record screen and in History, opens your mail app with subject and body filled in. In the keyboard and the bubble, only the body is inserted; the subject stays in History.
- Calendar Event: the draft shows name, time, place and notes. Add to calendar opens your calendar app’s new-event screen with everything filled in, and you save the event there. TypeWhisper needs no calendar permission for this.
- The calendar app does not report back whether you saved. When you return, tap I saved the event, or Open draft again.
- Missing details are marked: without a date or time, the next full hour is a placeholder; without an end time, the event lasts one hour.
- After a quick dictation with a Calendar Event profile, the notification says Calendar draft ready and offers Add to calendar.
- History shows the recognized text under “Before corrections” when processing changed it. Report AI text in the ⋮ menu lets you report a bad result by email.
Text clean-up
Settings → Text clean-up works without a writing service. It applies to dictations, the keyboard and imported files, before the dictionary and profiles.
- Remove filler words (on by default): ähm, äh, uh, erm and the like; “um” only in English, “euh” only in French.
- Collapse stutters (on by default): a word repeated three times or more is kept once, “I, I, I think” becomes “I think”.
- Spoken punctuation (off by default): “comma”, “period”, “question mark”, “new line” and “new paragraph” become marks, also in German.
- Numbers as digits (off by default): spoken numbers from ten up become digits, “twenty three” becomes 23.
- Chinese characters: As recognized, Simplified, Traditional, Traditional (Taiwan) or Traditional (Hong Kong).
- In email, web and number fields, the keyboard never adds a final period.