AI dictation

AI dictation that never leaves your machine

Two models, both on your device. One turns your speech into text. A second cleans it up: dropping filler, fixing punctuation, restoring code identifiers, matching the style your work needs. No API key, no cloud round-trip, nothing uploaded.

Download for Mac See pricing

Mac & Windows · Free local speech-to-text · On-device AI cleanup on Pro (beta)

The short version

AI dictation is voice typing with a second model. Speech-to-text produces a raw transcript; a language model then polishes it: removing "um" and "you know", fixing punctuation, restoring code identifiers, and adjusting tone. SnailText runs both models on your device, so the transcript is never uploaded the way it is in cloud AI dictation tools. Five topic profiles and a choice of identifier styles control the cleanup, and you can turn it off for verbatim output.

Two models, both running locally

Plain dictation uses one model: speech-to-text. You speak, it types what it heard, filler and all. AI dictation adds a second model, a language model (the same class behind ChatGPT and Claude), that edits the transcript into something you can actually send.

In nearly every AI dictation app, that second step runs in the cloud. So even when the speech recognition happened on your device, the transcript still gets uploaded to OpenAI, Anthropic, or Google for the cleanup. "Local speech recognition" and "local AI dictation" are not the same claim.

SnailText runs both. Whisper (or Parakeet TDT) does speech-to-text in RAM. Then a local Gemma model does the cleanup, on your own hardware. No API key, no network call, nothing uploaded at either stage.

What the second model does for you

The cleanup is tuned to preserve your meaning, not rewrite it. Here is what it handles.

Cleanup & correction

Filler words dropped, punctuation and capitalization repaired, grammar slips fixed, brand names cased correctly (github → GitHub, postgres → Postgres). A transcript you can send, not one you have to fix.

Topic profiles

Five profiles tune the cleanup to what you dictate: General, Development & IT, Writing, Business, and Academic. A developer dictating code and a novelist dictating prose get different handling.

Code identifier styles

Dictate code and the model restores symbols in your convention: snake_case, camelCase, kebab-case, or PascalCase. Say "recording completed", get recording_completed.

Tone & translation

Shift casual speech into a professional message, or speak your native language and get the text in another, translated on your machine, not sent to a cloud API.

You said

so umm i pushed the fix to githab and the the latency droped on postgress

SnailText gives you

So I pushed the fix to GitHub, and the latency dropped on Postgres.

Plain speech-to-text vs cloud AI dictation vs SnailText

Plain speech-to-text vs cloud AI dictation vs SnailText, compared across what you get, filler and punctuation handling, code identifiers, where cleanup runs, offline support, and whether the transcript is uploaded.
Aspect	Plain speech-to-text	Cloud AI dictation	SnailText
What you get	Raw transcript	Cleaned text (uploaded)	Cleaned text (on-device)
Filler & punctuation	Left as-is	Fixed	Fixed
Code identifiers	No	Sometimes	snake_case / camelCase / …
Where cleanup runs	No cleanup	Cloud LLM (GPT/Claude/Gemini)	Local LLM on your device
Works offline	If STT is local	No	Yes, both models local
Transcript uploaded	Maybe	Yes, every dictation	Never

Frequently asked questions

What makes SnailText AI dictation and not just speech-to-text?

Two models. Speech-to-text turns your voice into a raw transcript; a second model, a language model, then cleans it up: dropping filler, fixing punctuation, restoring code identifiers, and adjusting style. Both models run on your device.

Does the AI step send my text to the cloud?

No. The cleanup model is a local Gemma model running on your own hardware. Most AI dictation apps run this step in the cloud, so the transcript is uploaded even when the audio was processed locally. SnailText uploads nothing at either stage.

Can I control how it edits my text?

Yes. Pick one of five topic profiles, choose the identifier style for code, or turn the cleanup off entirely for verbatim speech-to-text. AI dictation is a mode you switch on, not a filter you are stuck with.

Does it work on Windows and Mac?

Both, as equal first-class platforms. GPU acceleration via Vulkan on Windows and Metal on Mac, with a CPU fallback that still runs the compact models in real time.

How much does it cost?

The free tier gives you unlimited local speech-to-text with no account. The on-device AI cleanup (topic profiles, identifier styles) is a Pro feature, currently in beta. Pro is $7.49/mo or $89/yr, covering up to 3 devices.

Try AI dictation that uploads nothing

Free local speech-to-text, no account needed. Add the on-device AI cleanup on Pro.