MindType
Coming soon to the App Store
MindType is a keyboard for iPad and iPhone. Tap the mic, talk through the task, and the text lands in the app you are using.
Needs iPadOS 17 or later on iPad, and iOS 17 or later on iPhone. Typing works without Full Access.
Check how uploads are handled before we change anything.
src/routes/upload.ts. It has no rate limit today, and every request goes straight to storage.
// src/routes/upload.ts
app.post('/upload', auth, async (c) => {
const file = await c.req.blob()
await storage.put(key(c), file)
return c.json({ ok: true })
})
What limit do you want?Add rate limiting to the upload endpoint. Use a token bucket per user with 60 requests a minute.
Explain the task the way you would explain it to a colleague. Then run an action you wrote once, like the one below, to turn the transcript into a prompt an agent can follow.
okay so um I want you to look at the onboarding screens and like make the privacy one clearer, it should name every service we send data to, and uh keep the current layout, don't redesign it
Make the privacy screen in the onboarding flow clearer. Keep the current layout.
Clean up prompt is a custom action. You write its instructions once in the MindType app, and it shows on the keyboard next to Proofread, Rewrite and Shorten. The result above is an example, and model output varies.
The mic sits on the keyboard, so you dictate straight into the field: an agent chat, a terminal, a note or an email. Pause while you think, then resume. When you finish, the text lands at your cursor.
How the microphone worksSelect text, or use everything in the field, and tap an action. Proofread, Rewrite, Shorten, Reply and Postify are built in, and TTS reads the text aloud. You can also write your own actions. The result opens in a preview, where you replace your text, insert the result or copy it.
What each action doesProofread
Shorten
Your first tap on the mic opens the MindType app and starts a session. While the session is on, the next dictation starts from the keyboard. A Live Activity shows whether MindType is ready, recording or transcribing, and it has a button to end the session.
When a session ends
The Live Activity shows each state:
History keeps your last 200 dictations and AI action results. Tap an entry to copy it. MindType does not upload or sync your history, and you can clear it at any time.
What stays on your deviceAdd rate limiting to the upload endpoint. Use a token bucket per user with 60 requests a minute.
can u check the servr logs i think theres a issue
→ Can you check the server logs? I think there's an issue.
Tell the team the build is green and I will merge after lunch.
iOS does not let keyboards use the microphone. So the MindType app records the audio, and the keyboard inserts the text. That is why step 2 opens the app once.
Open MindType and follow the setup screens. Then go to Settings > General > Keyboard > Keyboards > Add New Keyboard, choose MindType, tap it, and turn on Allow Full Access.
The MindType app starts recording and a session. Tap the name of your app in the top-left corner to go back, and keep talking.
Recording started. Go back to your app and keep talking.
The text lands in the field a few seconds later. While the session is on, the next tap on the mic records from the keyboard, and MindType stays in the background.
A good prompt for an agent carries context, limits and steps. On iPad you can say all of it while the conversation stays on screen, and check the text before you send it.
Add rate limiting to the upload endpoint. Use a token bucket per user with 60 requests a minute.
Return 429 with a Retry-After header when a user goes over. Write tests for the limit and for the reset. Do not change the public API.
Steps:
1. Read the upload handler and the auth middleware.
2. Add the token bucket and the 429 response.
3. Add tests for the limit and the reset.
4. Run the full test suite and report the results.
MindType works the same way on iPad and iPhone. It needs iPadOS 17 or later.
Pause a recording, think, then resume. Tap finish when the whole prompt is ready.
Run Proofread, Shorten or your own prompt action on the transcript.
Each iPad and iPhone keeps its own history. MindType does not sync it.
MindType sends data only when you dictate, run an AI action or tap TTS. This table names every service that receives it.
| What you send | When | Where it goes |
|---|---|---|
| Audio of a dictation | When the recording stops: you tap stop, it reaches 5 minutes, or the session ends | MindType's server on Cloudflare, then Inworld, which transcribes it with Groq's Whisper model |
| Text for an AI action | When you tap the action | MindType's server, then OpenRouter, then xAI, whose Grok model writes the result |
| Text for TTS (read aloud) | When you tap TTS | MindType's server, then Inworld, which turns it into speech |
| Install ID, a random identifier | With each request | MindType's server only |
| History, settings and your actions | Never | Stays on your device |
MindType has no accounts, ads or analytics SDKs. Our server passes your audio and text on and keeps no copy.
iOS does not give keyboards access to the microphone. The MindType app records the audio, so your first tap opens it and starts a session. Go back to your app and keep talking. While the session is on, later taps start recording from the keyboard.
No. Dictation, AI actions and TTS all use servers, so they need an internet connection. Typing works offline.
Yes. MindType runs on iPad with iPadOS 17 or later. The keyboard, the AI actions and the session work the same way as on iPhone.
No. MindType has no sign-up and no account. The app creates a random install ID so that our server can limit how many requests one install sends. MindType sends no audio or text until you agree on its privacy screen.
It will be available for iPad and iPhone. Write to us with questions or feedback before launch.