AI Keyboard — User Guide

An Android keyboard (IME) that lets you speak instead of type. Voice is transcribed by Groq Whisper (recommended), Gemini Live, or Android’s built-in recognizer, and you can optionally run text through AI “prompts” to fix grammar or rewrite tone using your own Groq or Gemini API key. Typing assistance — suggestions, auto-fix, and next-word prediction — runs entirely offline, and hold gestures put digits, punctuation, and word-by-word delete on the letter keys.

Get started See dictation workflow Install on Google Play Send feedback

Optional AI features require your own API key — Groq (recommended; its generous free tier covers everyday voice typing and corrections) or Gemini. Get a key in minutes with the Groq guide or the Gemini guide. Gemini Live streaming transcription may be limited or unstable on the free tier and can incur costs for native audio models — see Gemini API pricing or the cost estimator. Android’s built-in transcription is always free, and typing assistance never touches the network at all.

Features

What AI Keyboard can do:

Voice-to-text dictation inside any app — record-then-transcribe with Groq Whisper (recommended) hears your complete sentence for far better accuracy; Gemini Live streaming and Android’s built-in recognizer are also available
AI correction/rewriting using Groq or Gemini, with prompt chips for tone/style (Fix Grammar, Professional, Casual, custom)
Field Correction — hold a prompt chip to correct everything in the text box, preview the result, then replace it in place with a five-second Undo
Typing assistance, fully offline — a suggestion strip as you type, auto-fix on space, next-word prediction, and a learned-words manager; no network, no API key
Hold gestures — hold a top-row key for its digit (q for 1 through p for 0), hold . or , for a sixteen-mark Punctuation Palette, and hold backspace to clear whole words after the first second
AI Transform — select text in a supported app or share text to AI Keyboard, apply a prompt, then replace, copy, or share the result
Correction preview with a model badge naming the AI that answered, and actions: Insert / Copy / Share / Dismiss
Settings for auto-correct, auto-insert, and auto-copy
Token usage statistics stored locally on-device

What’s new

Recent releases at a glance:

v1.6 — Digits and punctuation without leaving the letters

Hold any top-row key for its digit — q for 1 straight across to p for 0, with the digit shown on each keycap. No more trips to the 123 page.
Hold . or , and the Punctuation Palette opens: sixteen marks in two rows. Slide to one and lift to type it. Release without moving and you simply get the key you were pressing — a slip costs nothing.
A held backspace starts on characters and switches to whole words after about a second, with a buzz when it switches and a stop at every line break. A tap always deletes exactly one character.
Every key now responds to a hold in 300 ms instead of half a second.
Removed: the accented characters on a e i o u n c — digits and accents wanted the same keys, and the built-in dictionary never recognized accented words. If you type them regularly, keep a system keyboard alongside and switch with the globe key.

v1.5 — Field Correction

Hold a prompt chip to correct everything in the text box, preview it, then tap Replace. Tap a different chip in the preview to compare styles on the same text.
Built to be paranoid: it re-reads the box before replacing and asks if you typed in the meantime, refuses truncated results, and shows a five-second Undo bar after every replace.
Off by default (Settings → Field Correction), and refused outright in password, email, address, phone, number, and date fields.
Long dictations no longer come back silently cut off — truncated replies are caught and marked.

v1.4 — Typing Assistance, entirely offline

A suggestion strip shows candidates for the word you’re typing; obvious misspellings fix themselves on space (“teh” → “the”), and one backspace undoes any fix and teaches the keyboard to leave that word alone.
Next-word prediction offers what usually comes next — tap it and it lands with a space.
Runs against a 37,000-word dictionary built into the app: no network, no API key, works in airplane mode. Learned words stay in a private on-device database you can review and delete in Settings.
Switches itself off in password, email, URL, phone, and number fields.

v1.30 — Groq Whisper voice and model transparency

Groq Whisper speech-to-text: your full recording is transcribed with complete sentence context for dramatically better accuracy, with a generous free tier.
A model badge on corrected text and live transcription names the model that actually answered, so silent fallbacks are visible.
Gemini correction uses Gemini 3.5 Flash-Lite by default, with an advanced model override and a “Fetch available models” picker in Settings.
AI Transform from text selection or share, Android 16 support, and Groq-first onboarding. The legacy localhost/Moonshot integration has been removed.

Setup (first time)

AI Keyboard is a system keyboard. You only need to enable and select it once.

1) Open AI Keyboard

Open the app from your launcher. You’ll see the setup wizard.

Setup wizard welcome screen — Setup wizard — Welcome

2) Enable the keyboard

Follow the wizard and enable AI Keyboard in your device’s input method settings.

Setup step: enable AI Keyboard — Setup wizard — Enable AI Keyboard

3) Select AI Keyboard as your current keyboard

Tap Select Keyboard, then choose AI Keyboard from the system picker.

Setup step: select AI Keyboard — Setup wizard — Select AI Keyboard

Android input method picker showing AI Keyboard — System UI — Choose input method

4) (Optional) Add a Groq or Gemini API key

Add an API key to enable Groq Whisper voice typing and AI correction/rewriting. Setup recommends Groq — its free tier covers everyday use. You can skip this and add a key later in Settings. Need help getting one? See the Groq guide or the Gemini guide.

Setup step: add Gemini API key — Setup wizard — API key (optional)

5) Done

Open any app with a text field and start using the microphone.

Setup complete screen — Setup wizard — You’re all set

How to use

Use AI Keyboard anywhere you can type.

Choose a prompt (style)

Prompt chips control how AI rewrites your text (for example: Fix Grammar vs. Professional tone).

AI Keyboard inside Messages with prompt chips above the keys and digit hints on the top-row keycaps — Keyboard view — prompt chips + QWERTY with digit hints on the top row

Voice dictation

Tap the microphone button and speak. When you’re done, stop the recording.
If Auto-correct after voice is enabled, you may see two stop options: Stop + AI and Stop (no AI).

AI correction preview (Insert / Copy / Share)

When AI correction is enabled and your API key is configured, the corrected text appears in a preview panel. Insert it into the current field, or copy/share it.

Correction preview panel showing corrected text with Copy and Share actions — Correction preview — corrected text + actions

Fix the whole message with one hold (Field Correction)

Hold — don’t tap — a prompt chip and the keyboard corrects everything in the text box, shows a preview, and replaces it when you tap Replace. Tap a different chip in the preview to compare styles on the same text, and tap Undo within five seconds to get your original back. Off by default: turn it on under Settings → Field Correction. A plain tap on a chip still just picks that prompt.

Typing assistance (offline)

As you type, the suggestion strip shows candidates for the current word — tap one to swap it in. Obvious misspellings fix themselves when you type a space, and a single backspace restores your word exactly and stops the keyboard correcting it. After a word, the strip offers the likely next word. All of this runs on-device against a built-in dictionary — no network, no API key.

Suggestion strip showing the typed word tomoro with candidates tomorrow, tomato, and tumor — Suggestion strip — “tomoro” with its candidates

Next-word prediction offering morning, afternoon, is, and at after the word tomorrow — Next-word prediction — after “tomorrow”

Digits, punctuation, and fast delete by holding

Hold a top-row key to type its digit (q for 1 through p for 0). Hold . or , to open the Punctuation Palette — slide to one of sixteen marks and lift to type it; release without moving and you just get the key you were pressing. Hold backspace and deletion switches from characters to whole words after about a second, with a buzz when it switches and a stop at each line break. Taps are unaffected — one tap is always one character.

Holding the u key types the digit 7, completing the phrase tomorrow at 7 — Hold u — the digit 7, no layout switch

Punctuation Palette open over the keyboard with sixteen marks and the question mark highlighted — Punctuation Palette — slide to ? and lift

Settings & customization

API keys & voice behavior

In Settings, you can add/clear your Groq and Gemini API keys, toggle auto-correct after voice, and control whether corrected text is auto-inserted or auto-copied. Need help getting a key? See the Groq guide or the Gemini guide.

Settings screen showing Gemini and Groq API key fields and voice input toggles — Settings — Gemini + Groq API keys and voice input

Choose your speech provider

Choose your speech recognition provider: Groq Whisper (recommended) records your full thought first, then transcribes it with complete sentence context — the most accurate option, and Groq’s free tier covers typical daily use. Android (Free) uses the device SpeechRecognizer with no key at all. Gemini Live provides streaming transcription, but on the free tier it may be unstable or quota-limited. Free Gemini keys are still great for output styles (Fix Grammar / Professional / Casual). For native audio pricing details, see Gemini API pricing. For a quick monthly estimate, see Cost estimator.

Settings screen showing speech recognition provider options: Android (Free) and Gemini Live (Uses API Key) — Settings — speech recognition provider (Android vs Gemini Live)

Advanced Gemini model & model badge

The recommended text model (Gemini 3.5 Flash-Lite) is selected automatically. Power users can opt into a custom Gemini model: tap Fetch available models to load the models your key can actually use, or type an ID by hand. Invalid or retired IDs safely fall back to the recommended default, and the model badge on every result names the model that actually answered.

Field Correction & Typing Assistance

Field Correction (off by default) enables the hold-a-chip gesture that corrects the whole text box — it sends the box’s contents to your AI provider, so it stays off until you turn it on, and is always refused in password, email, phone, number, and date fields. Typing assistance has a master switch with subordinate Auto-fix and Next-word prediction toggles (all on by default), plus a Learned words screen where you can review and delete everything the keyboard has picked up from you.

Settings screen showing the Field Correction toggle and the Typing Assistance switches for auto-fix, next-word prediction, and learned words — Settings — Field Correction + Typing Assistance

Manage prompts

Enable/disable built-in prompts and keep only the styles you want to see on the keyboard.

Settings screen showing prompt toggles — Settings — prompts list

Create your own prompts

Add a custom prompt (e.g., “Make it more polite” or “Summarize”) and tune temperature.

Create custom prompt dialog with fields for name, description, and system prompt — Custom prompts — create and save your own instructions

Usage statistics

Track recent AI requests and token usage. This data is stored locally on your device.

Token Usage screen showing overview cards — Token usage — overview

Permissions & data (summary)

Microphone: used for voice dictation when you tap the mic — audio is only recorded then, never in the background.
Internet: required for AI correction and transcription requests (Groq or Gemini); voice dictation may also use network depending on your device’s speech recognition service.
API keys: stored encrypted on-device using AndroidX Security.
Typing assistance: fully offline — suggestions, auto-fix, and predictions never make a network call, and learned words live in a private on-device database you can clear anytime.
Field Correction: off by default because it sends the text box’s contents to your AI provider; it is refused in password, email, phone, number, and date fields.
Usage stats: token usage is stored locally on-device.

Troubleshooting

The keyboard doesn’t appear: go to Setup/Settings → enable AI Keyboard, then select it as the current input method.
Microphone not working: grant microphone permission and try again.
AI correction not working: ensure a Gemini API key is configured and you have internet access.
Gemini Live not working: if you’re on the free tier, Live streaming can be unstable or quota-limited. Try again, confirm your API key and plan/quota, or switch Settings → Speech Recognition → Provider to Android (Free) for transcription and keep using Gemini for output styles. See Gemini API pricing, or use Groq Whisper (recommended) for voice instead.
Holding a letter no longer types an accent: the accented characters on a e i o u n c were removed in v1.6 — the top row now types digits on hold. If you type accented words regularly, enable a system keyboard alongside and switch with the globe key on the space bar.
A long hold on backspace deletes whole words: that’s the v1.6 hold-to-delete escalation, and only a held key speeds up — a tap always deletes exactly one character. Deletion also stops at each line break.

Support contact

Developer: YCK
Email: [email protected]
Google Play: play.google.com/store/apps/details?id=com.kengblog.aikeyboard
Feedback: https://forms.gle/EVX64Rimp2fQsJxh8
Closed testing: How to join & help test