Local-first dictation utility · Windows-first beta

Speak instead
of typing.

Press your hotkey, speak naturally, and Whispa turns your voice into text — copied to your clipboard, ready to paste directly into the field where you're already working.

Download the beta See how it works

✓ Local transcription with whisper.cpp ✓ Global hotkey · clipboard-ready ✓ Consent-based model setup

PRESS MOUSE 5 OR CTRL+SHIFT+SPACE

REC · 00:04 whisper.cpp · ggml-small.bin

README.md prompt.txt notes.md

1Draft a launch tweet for a local-first

2desktop dictation app called Whispa.

3Tone: calm, confident, no hype. Mention

4the global hotkey workflow

listening · mouse button 5 to stop ● 00:04

// active app

// the four-step loop

Ready · Recording · Transcribing · Pasted.

One tight loop you can run in seconds, anywhere your cursor lives — without leaving the app you're working in.

STEP_01READY

Place your cursor.

Whispa sits compact in the background. Drop your cursor into any text field — chat, editor, doc, terminal — and it's ready.

STEP_02RECORDING

Press & speak.

Hold your global hotkey or mouse button. A timer and waveform confirm Whispa is listening.

STEP_03TRANSCRIBING

Local transcription.

Whispa transcribes on your machine using whisper.cpp. Normal runtime stays local after the consent-based model setup.

STEP_04PASTED

Text where you work.

The transcript copies to your clipboard — or, with Auto-Paste, drops directly into the field your cursor was in.

// what's inside

A small, sharp tool for people who think faster than they type.

Built as a compact desktop utility — not a SaaS product, not a cloud assistant. Power-user behaviour, premium indie polish.

01 / Global hotkey recordingHOTKEY

Press once. Speak freely.

Trigger Whispa from anywhere on your desktop with a global keyboard shortcut or a mouse button. Set your own in Settings.

CTRL

SHIFT

SPACE

MOUSE 5

02 / Local transcriptionWHISPER.CPP

Your voice stays on your machine.

Normal transcription runs locally with whisper.cpp once you've completed the first-run model setup. No cloud round-trips during day-to-day use.

03 / Auto-Paste into the active fieldFOCUS_TARGETED

Speak into any text field.

For the intended public build, Whispa can drop your transcript directly into whichever app and field your cursor was in — chat, prompt, doc, IDE, or terminal — with user-controlled safety settings.

"draft a launch tweet for whispa…" → cursor in: prompt.txt PASTED

04 / Clipboard-readyCOPY

Always copied, automatically.

Auto-copy the finished text to your clipboard so you can paste it wherever you like.

⌘V

            ready to paste
// in any app

05 / Consent-based setupFIRST_RUN

No blind model bundling.

On first run, Whispa asks before downloading a Whisper model — or point it to one you already have locally.

✓ ready · transcribing locally

06 / Tune to your workflowSETTINGS

Toggles, not dashboards.

Auto-copy. Delete recordings after success. Always on top. Custom hotkey. Paste behaviour. Honest, scoped controls.

Auto-copy after transcriptionsend finished text to clipboard

Delete recording after successkeep transcripts only · save disk space

Always on topkeep whispa above other windows

07 / File transparencyTOOLS_&_FOLDERS

Stay in control of your files.

Direct access to your output folder, recordings, logs, and one-click cleanup. Nothing hidden, nothing buried.

~/Whispa/output/42 files

~/Whispa/recordings/auto-deleted

~/Whispa/logs/scoped · rotating

› Run cleanup2 clicks

08 / Local-first stanceWORKFLOW

Built for fast desktop work.

Compact window. Always-on-top option. Privacy-conscious logging. Optional deletion of recordings once a transcript lands.

◐

          normal transcription runs
locally after first-run setup
        

// before / after

From "thinking out loud" to text on screen.

Two states of the same blinking cursor: one where you're typing, one where Whispa just put the words there.

// before · typing

draft a launch tweet for a
local-first desktop dict|
_

≈ 38 wpmtyped · in progress

// after · whispa

Draft a launch tweet for Whispa — a local-first desktop dictation app. Calm tone, no hype. Mention the global hotkey workflow and that transcription runs locally after the first-run model setup.

⌘ pasted in 6.2sspoken · 1 take

// the visual system

Calm, warm-black, amber-lit, mono-tagged.

A desktop-utility palette. Premium indie software vibe — not corporate SaaS, not generic AI cloud branding.

Background

#0E0C0A

warm‑black · base canvas

Surface

#181410

cards · panels

Amber primary

oklch(0.78 0.13 70)

accent · CTA · status

Rec red

oklch(0.62 0.20 25)

recording state only

Ink 100

#F3ECE1

primary text on dark

Ink 70

#C8BFB1

body · subheads

Ink 45

#8A8073

labels · meta

Ready green

oklch(0.72 0.14 150)

"ready" pill only

// display + uiINTER

Speak instead
of typing.

Press once. Speak freely.

Inter at 15–18px for body. Tracking −0.01em on display sizes, balanced wrapping for hero. Weight 600 for headlines, 400 for body, 500 for UI.

// labels + codeJETBRAINS MONO

// REC · 00:04

whisper.cpp · ggml-small.bin

Mono is reserved for: status pills, file paths, captions, kbd shortcuts, "//" callouts, and any text that points back at the desktop-utility nature of the product.

record dot

waveform

clipboard

cursor

hotkey

local shield

// LAYOUT

12-col grid · generous gutters

Everything sits on a 12-column grid with calm 24–48px gutters. Cards are framed; backgrounds breathe. Section heads use mono eyebrows + display heads for rhythm.

// SCREENSHOT FRAMING

App always anchored, never floating alone

Screenshots sit inside dark warm-black surfaces with a 1px line and a soft drop shadow. Add a mono caption + status pill so the state is obvious at a glance.

// LOCAL-FIRST, NO CLICHÉS

Show local with files, not clouds

Use file paths, hotkey chips, waveforms, cursor carets, and the actual app window to imply local-first. Avoid generic AI robots, glowing brains, server racks, neon clutter.

// what we say (and don't)

Honest copy. Honest claims.

Until the final clean test on a non-dev Windows machine ships, we keep claims scoped to what the build actually does today.

Safe to say

✓ Local-first transcription
✓ Global hotkey recording
✓ Clipboard-ready output
✓ Auto-Paste workflow (intended public build)
✓ Consent-based first-run model setup
✓ Privacy-conscious logging
✓ Built for fast desktop workflows
✓ Normal transcription runs locally after setup
✓ Windows-first initial build · broader desktop vision

Avoid (until verified)

✕ 100% private
✕ No data ever leaves your device
✕ Fully secure
✕ GDPR compliant
✕ Enterprise-ready
✕ No internet ever
✕ Available on all platforms now
✕ Production-verified on all Windows systems

// SHORT PRODUCT INTRO · DOWNLOAD PAGE~60 WORDS

Whispa is a compact local-first desktop app that turns your voice into ready-to-use text.

Press your global hotkey, speak naturally, and Whispa transcribes locally with whisper.cpp. The finished text is copied to your clipboard — and, in the intended public build, can be pasted directly into the active text field with user-controlled safety settings.

// BETA NOTERELEASE_CAVEAT

Beta build · Windows-first

Whispa is currently shipping as a portable Windows ZIP for early testers. The final clean test on a non-dev Windows machine is still pending; broader desktop platforms are part of the longer-term vision, not a current promise.

Speak instead
of typing.

Ready · Recording · Transcribing · Pasted.