// DevFlare product · v1.0 Windows 11 · Hold Alt to talk

Hold Alt. Speak.
It pastes.

VoiceFlare is a tiny always-on-top overlay. Hold Alt to record, release to stop — it transcribes, copies, and pastes into the app you were using. Local Whisper by default. Cloud when you want it.

Download for Windows See how it works
  • Hold-Alt PTT
  • Auto-copy + paste
  • Local Whisper
  • Alt+Tab safe

// Features

What VoiceFlare actually does now

Built for dictation into any Windows app — browser, IDE, Slack, Notes — without switching windows.

Alt 01

Hold-Alt push-to-talk

Default shortcut is bare Alt hold-to-talk. Hold to record, release to stop and transcribe. Alt+Tab and other system combos cancel cleanly — no hijacking.

⇄ 02

Toggle or push-to-talk

Prefer press-to-start / press-to-stop? Switch modes in Settings. Bind any Ctrl/Alt/Shift combo, or keep Hold Alt.

↵ 03

Auto-copy + auto-paste

On by default: cleaned text hits the clipboard, then VoiceFlare restores the previous app and pastes at the cursor. Misses still leave text on the clipboard.

◎ 04

Tiny always-on-top pill

Real UI is a ~92×44 dark pill: drag grip + mic only. Stays above every window, remembers position, no transcript chrome in your face.

≈ 05

Smart audio defaults

Noise suppression and AGC on by default. Silence timeout and max duration keep sessions from running away. Mic pulses while you hold Alt.

🔒 06

Local Whisper first

Default mode is local with the tiny model for speed. Upgrade to base / small / medium. Optional GPU path. Language fixed for faster dictation.

☁ 07

Cloud when you need it

OpenAI, Groq, Deepgram, or a custom OpenAI-compatible endpoint. API keys live in Windows Credential Manager — never plain text on disk.

✦ 08

Cleanup that ships clean text

Spoken punctuation, space cleanup, optional fillers/repeats/capitalization, custom replacements, and vocabulary. Raw preserved when you want it.

♪ 09

Sound cues

Start, stop, success, and error sounds so you know the pipeline fired without looking at the panel.

☰ 10

History, tray, onboarding

Optional SQLite history (off by default). System tray for record, hide, settings, history, pause shortcut, quit. First-run wizard for mic, model, and paste test.

⛨ 11

Safe paste

Heuristic skip for secure/password fields. Simulated paste never pretends to be Ctrl+C — clipboard is written directly.

⚡ 12

One pipeline, every path

Shortcut, tray, and panel all share the same stop → transcribe → clean → clipboard → paste workflow — consistent every time.

// How it works

Dictation without leaving your app

  1. 01

    Hold Alt

    Keep typing in Slack, VS Code, or a browser. Hold Alt — the overlay starts recording. (Alt+Tab still works; system combos cancel recording.)

  2. 02

    Speak, then release

    Watch levels. Release Alt to stop. Audio is resampled and sent through local Whisper or your cloud provider.

  3. 03

    Copied & pasted

    Text is cleaned, written to the clipboard, and auto-pasted into the window you were in. Done — no context switch.

// Transcription

Private by default. Cloud optional.

Default

Local Whisper

  • Starts on tiny for fast dictation
  • base · small · medium when you need accuracy
  • CPU by default · optional GPU path
  • Fixed language (EN) for speed — auto-detect available
  • Works offline after model download
  • Models under %AppData%\voiceflare\…\models

Optional

Cloud API

  • OpenAI · Groq · Deepgram · Custom URL
  • Keys in Windows Credential Manager
  • Audio only uploaded when you choose cloud
  • Same cleanup + paste pipeline afterward

// Privacy

Local-first by design

Local mode never uploads audio. History is off until you enable it. Temporary recordings are deleted unless you choose to keep them. No telemetry by default. Diagnostics stay on-device when enabled.

// Requirements

What you need

  • Windows 11 (x64)
  • WebView2 (included on Win 11)
  • Microphone access
  • Disk for Whisper models (local mode)
  • Internet + API key only if using cloud

// Out of the box

Sensible defaults after install

ShortcutHold Alt (push-to-talk)
TranscriptionLocal · tiny model · English
OutputAuto-copy + auto-paste on
AudioNS + AGC · silence stop 2s
HistoryOff until you enable it
PanelAlways on top · ~85% opacity

// Get VoiceFlare

Install once.
Dictate everywhere.

Windows installer from GitHub Releases. Installs under %LocalAppData%\VoiceFlare. Onboarding covers mic, model, shortcut, and paste test.

Download for Windows Talk to DevFlare

Version 1.0.0 · MIT · Tauri + React · A DevFlare product · voice.devflare.site