SaypeBetav0.1.0Private offline dictation · Linux, Windows & Mac
Your words never leave the room.
Cloud dictation uploads every word. Saype (say + type, pronounced “sayp”) doesn't: press a hotkey, speak, and a blind typer emits only your transcript at the cursor in any app. True streaming Nemotron 3.5 — words land as you speak, not after you pause. Super fast, fully offline, 90%+ right first time on clear speech — the rest fixed live by Backspace revision — and a Whisper mode hits 95%+ on the toughest audio.
one-time · tax included · FOUNDER BETA — FIRST 100 LOCK THIS PRICE
14-day money-back guarantee — if it doesn't hear you, email for a refund, no forms.
Founder beta: the first 100 lock $29 before it goes to $99. Whisper mode covers 95%+ accuracy tasks today; word boosting ships next, free for founders.
Who it's for
- 01
Researchers & journalists
Field notes and interviews with no server ever holding your source material. Fast, 90%+ accurate drafts straight from the field.
- 02
Lawyers
Privileged memos that stay on your machine — no cloud copy, no vendor to subpoena. Whisper mode when a passage must be close to exact.
- 03
Clinicians & healthcare teams
Notes where audio never leaves the device. No compliance badges claimed, just architecture: no cloud mode exists.
- 04
Developers & heavy typists
Dictate prompts, docs, messages, and agent chats anywhere your cursor lands. It's prose, not code syntax — a fast 90%+ draft you glance over beats typing it all.
Not for: verbatim record or captions — use Whisper mode, or wait for word boosting.
Why it's different
- 01
Private offline dictation on Linux, Windows, and Mac
Local streaming transcription, typed at your cursor through the OS-approved channel. Cache-aware RNNT processes each new chunk once — no overlapping recompute — so latency stays sub-second on CPU. One hotkey flow on all three OSes; no other offline option covers them all.
- 02
Blind typer: emits only your words
Only transcript characters plus Backspace for corrections. No uinput, no evdev, no keylogging, no shortcuts — holding Ctrl/Alt/Super/Shift safely skips instead, the text waits in history. Your clipboard is never touched.
- 03
No account, no cloud, no calling home
Fetch the ~1 GB voice model once, then fully offline. Decline the input permission and it still works, history only — your clipboard is never overwritten.
Only your transcript. Nothing else.
NEVER: UINPUT / EVDEV · KEYLOGGING · RETURN SYNTHESIS · CLOUD AUDIO
Pricing · one-time
Own it once. No subscription.
Cloud dictation costs $12–15/mo forever. Saype is one payment, priced fairly for your region — shown automatically below.Founders lock the beta price; everyone after pays $99. Refunds revoke the license key and the app stops activating — your audio never left your machine either way.
one-time · tax included · FOUNDER BETA — FIRST 100 LOCK THIS PRICE
14-day money-back guarantee — if it doesn't hear you, email for a refund, no forms.
Roadmap · free for owners
- 01
Whisper mode for 95%+ accuracy — available now
Switch to --model whisper for large-v3-turbo accuracy on tough audio, fully offline — plus translate. Pastes land every ~2s.
- 02
Word boosting for Nemotron
Teach it your names, terms, and jargon so streaming gets your vocabulary right the first time.
Demo · true streaming, unedited
Raw cut: watch words land mid-sentence — that's streaming, not chunked Whisper. A tighter edit is coming.
- 01Words appear while you're still speaking — interim results, typed live.
- 02When you pause, the endpoint finalizes the sentence.
- 03If the model revises itself, it Backspaces only the changed tail and retypes — same blind-typer channel, nothing else it can do.
Requirements
- 0164-bit Linux, macOS, or Windows
- 02A microphone — built-in is fine
- 03~1 GB free disk for the one-time voice-model download
- 04Internet once, to fetch the model — fully offline after that
Accuracy figures are founder-measured across real dictation sessions, counted harshly — expected errors per 100 words on clear mic speech. Measured WER bench to follow.