No cloud round-trip, no waiting on a server. From the moment you tap the floating mic, everything happens on the device in your hand — across eleven Indian languages.
Tap the floating mic, and betterflo captures your voice. It turns speech into text on the device. Then a light pass tidies it up and inserts it where you were typing. That’s the whole loop — and none of it needs the internet.
A small mic bubble sits on top of whatever app you're in — a chat, an email, a search box. Tap it and just say what you mean. Mix languages mid-sentence freely; you don’t switch modes or pick a language first. Prefer a keyboard? betterflo’s has a mic key beside the space bar that starts the same dictation.
On phones that can run it, words land as you talk — see what your phone gets.
An on-device speech model turns what you said into text — right there on the phone. No internet, no server, nothing uploaded. It's built for budget hardware, not just flagships, and it keeps working on the metro with no signal. See exactly what your phone gets.
Offline isn’t a fallback — it’s the default path.
A quick pass adds punctuation and fixes casing, so “kal meeting hai at 5 please confirm karna” comes out as a clean sentence — then it's placed straight into the field you were dictating into. It tidies what you said — it doesn’t reword it or invent things you didn’t.
Want more polish? There’s a dial for that.
The floating mic is the default: a bubble over whatever app you’re in, no keyboard switch. It needs two Android permissions, each explained before it’s requested. Would rather skip that? The betterflo keyboard is a keyboard you type on, dictate from and edit in — with no accessibility service at all.
A bubble that sits over whatever app you’re in. Needs “display over other apps” and an accessibility service to insert text into the field you’re dictating into — both explained plainly before Android ever asks.
A full typing keyboard — QWERTY, a suggestion strip, the mic key beside the space bar — with an action bar above the keys: Fix, Tone, Rewrite and Translate, each running on the phone. No accessibility permission at all. Translate reaches every phone; Fix, Tone and Rewrite need a model you download, and the lowest device tier can’t run them — so there, those buttons aren’t drawn at all.
Behind the three taps is a simple pipeline: an on-device speech model hands off to a light cleanup step. Each part can be upgraded on its own — and the honest bit is what the pipeline can’t do.
Captured only while you hold the mic. Released the moment you’re done.
Speech recognition, tuned for Indic. Swappable — we can ship a better model without changing how you type.
swappablePunctuation, casing, obvious slips. Tuned to fix, not rewrite — and dialled to light by default.
light-touchInserted straight into the field you were dictating into — by the floating mic or the keyboard. That’s it.
Every step here runs on your device. Nothing is sent by default.
ASR is one module, not the whole app. When a stronger on-device model is ready, it slots in — you keep the same keyboard, and get better recognition without relearning anything.
By default the pipeline has nowhere to send your voice — it’s processed on-device, then released. If we add an optional cloud-quality tier later, it’s strictly opt-in, and you’ll know before anything goes anywhere.
It fixes punctuation and casing and leaves your wording alone. If you ever want more, turn the polish up. Even then, it edits what you said — it doesn’t write something new in your place.
By default the floating mic inserts your text using an Android accessibility service — which reads the field you're dictating into, locally, to place your words there, and nothing more. You'll see a plain-language disclosure before that permission is ever requested. Prefer not to grant it? The keyboard inserts text the ordinary way, with no accessibility service involved.
An optional on-device step — voice activity detection — marks where you actually paused, so a long dictation is transcribed phrase by phrase instead of as one long guess. It's opt-in and works the same regardless of phone tier; it doesn't make dictation faster, but it does make longer speech more accurate.
On the two higher device tiers, text can stream in as you speak. On the two lower tiers the pipeline keeps one model in memory at a time, so text lands the moment you stop instead — permanently, not as a "coming soon". See which your phone gets.
Set to Light by default
It fixes, it doesn’t rewrite. You decide how far it goes — and it starts at the least.
Free to start, no account, offline from the very first tap. After you install, you’ll turn on the floating mic — it’s quick, and we walk you through what each step needs and why.
betterflo