Open-source and privacy-first View on GitHub →

Stop typing, just say it

AI voice keyboard for iOS that turns speech into ready-to-send text.

One-time payment. No subscriptions.

100% Offline

Your voice is transcribed right on your phone, with zero internet required.

No Subscriptions

Pay once and keep it forever, no renewals or recurring fees.

No Servers

Your voice never leaves your phone, all processing happens on-device.

Lifetime License

Covers every future update: new speech models, features, bug fixes, etc.

Recording History

Every transcription is saved and timestamped for easy access later.

Built for iOS

Native Swift and SwiftUI app, lightweight and fast with no compromises.

See what's inside

From auto-punctuation to voice prompts — everything you need to go from speech to polished text.

AES-256 Encrypted
Voice recordings and text are encrypted with iOS Data Protection (AES-256).
I just left the clinic, the doctor said everything looks fine but I need to come back next Thursday for one more test
15 Oak Street, apartment 4B, that is the place I want to rent, remind me to call the landlord and send the deposit by Friday
•••••••••••••••••••••••••••••••
•••••••••••••••••••••••••••••••
Please do not mention this to anyone at work yet, but I got an offer from another company and I am seriously thinking about it
I love you so much and I miss you every day, I know this year was not easy but I am so proud of everything you have done
Automatic Punctuation
Your speech is instantly transformed into clean, perfectly punctuated text.
Hey, I just talked to Sarah - she said the apartment looks great! The kitchen, the balcony, everything. Can we move in on Saturday, or is Sunday better? Let me know what you think!
Smart text cleanup
Fix typos, improve grammar, and clean up your writing in one tap.
BEFORE
This idea is kind of interesting
Rewrite
Custom actions are also supported
This concept has real potential
AFTER
Voice Prompts
Tell the keyboard what to write, and it composes the text for you.
1. Prompt
Explain to my boss why I'll be late
1. Prompt
Politely decline a dinner invitation
1. Prompt
Write a short birthday greeting
2. Output
I'm running about 20 minutes late due to traffic — sorry for the inconvenience.
2. Output
Thanks for the invite, but I already have plans that evening!
2. Output
Happy Birthday! Wishing you joy and amazing adventures!
Choose Your Writing Style
Formatting that adapts to each app.
Formal
Caps · full punctuation
A
Wait, did you see that? He can't believe it happened.
Casual
Caps · less punctuation
A
Wait, did you see that? He can't believe it happened.
Very Casual
No caps · no punctuation
A
wait, did you see that? he can't believe it happened.
Choose Your Models
Speech Recognition
Parakeet
v2, v3
~450–466 MB Recommended
Whisper
Tiny, Base, Small
~69–445 MB Translates to English
Moonshine
Tiny, Base, +3 Streaming
~29–202 MB Lightweight
Text Processing
Gemma
3 1B
~851 MB Recommended
Qwen
3 0.6B, 3 1.7B
~551–2165 MB Wide Range
Llama
3.2 1B
~912 MB Meta Quality
SmolLM2
1.7B
~1056 MB English Pro

Supports 90+ languages out of the box

From widely spoken languages to regional dialects — Sayboard recognizes them all. Language detection is fully automatic. Just speak, and the model handles the rest.

Frequent questions

Yes — snippets let you create trigger phrases that auto-expand into longer text as you dictate. Great for emails, phone numbers, signatures, and other repeated text. Set them up in Settings.
iPhone and iPad with iOS 17 or later.
Yes. Sayboard is open-source and free to build yourself if you have a Mac with Xcode and some developer experience. If you’d like to support development, you can buy the app on the App Store for a symbolic price. No subscriptions ever.
New models are added regularly as better ones become available. If you’re missing a specific model, send us an email — we’ll do our best to add it if technically feasible.
Speech models range from 29 MB to 466 MB depending on the model. The app also offers optional LLM models for smart text cleanup, which take from 551 MB to 2.1 GB. You can download only the models you need.
Some models need more RAM than your device has. Sayboard checks compatibility automatically and disables models your device can’t run. Speech models work on all supported devices, but larger LLM models may require a newer iPhone or iPad.
This is an iOS limitation — keyboard extensions cannot access the microphone on their own. Sayboard opens briefly to activate the mic and returns you right back. The switch is instant and only needed once per session. To see it less often, increase the session duration in Settings.
No. That is a standard iOS indicator showing that microphone access has been granted to the app. It does not mean Sayboard is recording. Recording runs only while the mic button is active. Sayboard is open-source — you can check the code to confirm that no background recording ever happens.

Fully open-source

All source code is published on GitHub. No tracking, no telemetry, no external servers. Anyone can verify that your voice stays private and never leaves your phone.