AI summary
Floating dictation orb that transcribes speech into any app's text field using your own Gemini API key. Features customizable tones, per-app profiles, a personal dictionary, and optional clipboard translation. Redesigned with Material 3 styling, easier settings, and a self-populating dictionary in version 1.5.3. Requires draw-over-apps and accessibility permissions to function.
Generated by AI. May contain inaccuracies.
About this app
Speak into any Android app. RMBLR writes it down, tidies it up, and puts the text straight into the box you were typing in.
No account. No subscription. Your own Gemini API key, talking directly to Google.
What it does
A small orb appears only when a text field has the cursor — never sitting on your home screen. From there:
- Gesture: Tap What happens: Dictate. The transcript is written into the field.
- Gesture: Hold What happens: An arc of tones fans out. Slide to one, release, speak.
- Gesture: Flick What happens: Fires that direction's tone immediately, skipping the menu.
- Gesture: Drag What happens: Park it anywhere. It remembers.
Per-app profiles
The tones on the arc change with the app you are in. WhatsApp offers casual ones, Gmail offers professional ones, Slack sits in between. Pick the apps and the tones once, in the Profiles tab.
Your own tones
A tone is a name and a system prompt, nothing more. The built-ins are a starting point — write your own ("Rewrite this as a short LinkedIn post, no hashtags") and it appears on the arc alongside them.
A dictionary of your own words
Names, places and jargon get guessed at, and differently every time. Add them once and they are handed to whichever engine is running before it hears anything — with an optional spelling for the ones that never come out right. The list suggests itself from what you have already dictated, so you tap rather than type.
Translate what you copy
Optional, and off until you switch it on. Copy any text and the orb appears wearing a translate glyph; tap it and the translation comes back in a bubble you can read or copy. It goes away again a minute later.
The clipboard is read at the moment you tap and at no other time. Android does not let an app read the clipboard in the background, and RMBLR does not try to work around that — it borrows input focus for a fraction of a second on your tap, reads, and hands focus straight back.
Setup
1. Draw over other apps — so the orb can sit above what you are typing in. 2. Accessibility → RMBLR orb — the only Android API that can tell an app a text field has focus, and write text back into another app's field. Without it the orb cannot appear on cue and can only copy to the clipboard. 3. A Gemini API key — free to start, from Google AI Studio. Paste it on the last onboarding page or in Settings.
The key is stored on your device and used to call Google directly. Nothing is routed through any server of ours, because there isn't one.
Models
Transcription is a choice between three engines, in Settings.
- Engine: Gemini Flash Lite (default) What you get: Cheap and accurate, and it handles switching languages mid-sentence.
Tones: Yes
- Engine: Gemini Live
What you get: Audio streams to the model while you are still speaking, so the words are already there when you let go.
Tones: No
- Engine: Groq Whisper What you get: The fastest of the three, and free to start. It is Whisper, so it is strong in one language and weaker across two.
Tones: Yes
Transcription is not the part that needs a large model, so the default is the cheapest one that does the job well. If it is out of quota the client walks down a fallback chain rather than losing what you said.
Gemini Live streams as you speak. The socket opens when you press, audio goes up in 200ms chunks, and the transcript accumulates while you are still talking — there is no upload after you stop. The trade is that a Live model is speech-to-speech and cannot rewrite text, so tone actions are unavailable in that mode. The app says so rather than quietly ignoring your tone.
Languages
Whatever Gemini can hear, which is considerably more than its documented list — the language this was built for is not on that list and transcribes fine. Leave the language picker on auto unless you are being transcribed into the wrong language.
Translation providers
Translation can run on Gemini, Mistral, OpenRouter, Groq, OpenAI — or anything that speaks the OpenAI chat-completions API, including a server on your own machine. Give it a base URL and a model name.
Appearance
Light, dark, or OLED black — true #000000, so the pixels are switched off on an OLED panel. Material You is on by default from Android 12: the app takes its palette from your wallpaper. OLED black keeps its black background and borrows only the accent colour.
License
MIT
What's new
- A redesigned interface, a personal dictionary, and settings that are easier to navigate.
- The app now uses a more expressive Material 3-inspired design, with a floating bottom bar, rounded shapes, smoother touch feedback, and full-height content. - Your own dictionary is now available under History → Dictionary, so names, places, and jargon can be saved once and reused during transcription and streaming. - The dictionary can suggest repeated or capitalized words you already dictated, making it easier to add terms you want the app to learn. - Settings are now organized into four clearer sections: Models & keys, Translation, Appearance, and Typing. - Translation is now switched on from the main page instead of being buried in settings. - The Voice Studio page has been removed, and its Mode and Style controls are now on the main page where they belong. - The home screen no longer starts with a setup test prompt; trying the app now comes later and feels less intrusive.
About this version
- Version
- 1.5.3 (10)
- Size
- 16.7 MB
- Requires Android
- 7.0
- Target SDK
- 24
- Architecture
- arm64-v8a, armeabi-v7a, x86, x86_64
- Downloads
- 15
- Updated
- Aug 26, 2026
- Package
- io.github.pastdaking.rmblr
Similar apps
Ratings & reviews
- 50
- 40
- 30
- 20
- 10