Zero cloud. Zero telemetry. Instant keyboard injection into any Windows or Mac app with local Whisper speech recognition across 11 world languages.
Built for developers, writers, and privacy purists.
Verbtm gives you the speed of modern deep learning without surrendering your conversations to third-party clouds.
Fast-Whisper neural models run entirely on your local GPU or CPU. No audio bytes or transcripts ever leave your device. Fully functional even when completely offline.
Uses native Win32 batch SendInput with UTF-16 surrogate safety. Injects text into whichever application has focus without altering your clipboard.
A sleek, floating OLED-style pill HUD that stays cleanly docked at the top of your screen. Displays real-time 16-band audio levels and processing status without stealing window focus.
Compiled directly against NVIDIA CUDA 12 and cuDNN 9 libraries. Transcribes 10 seconds of speech in under 350 milliseconds on modern RTX GPUs, with auto-fallback to AVX2 CPU.
Eliminates Whisper repetition loops with custom n-gram deduplication, hallucination silence thresholding, and built-in Voice Activity Detection (VAD).
Pick between Push-to-Talk or Toggle mode. Choose from 6 WinUI 3 color presets, select your audio microphone device, and calibrate transcription sensitivity in seconds.
Hold Alt + Win, say any app name, and release — Verbtm fuzzy-matches and launches it instantly. Works with all your installed apps, including Start Menu shortcuts and Store-installed apps like Netflix, Copilot, and Spotify. A confirmation chime plays on success; a distinct tone on no-match — without touching your mouse or Start menu.
Multilingual Speech Engine
Verbtm brings private, zero-cloud speech recognition to the world’s most spoken languages. Automatic language identification, native script punctuation, and specialized speech models—100% offline on your Mac or PC.
Full vocabulary dictionaries, script-biasing prompts, and native punctuation formatting
Switch languages effortlessly mid-day. Verbtm’s two-stage acoustic classifier detects your language on the fly in milliseconds—no manual dropdown toggling required.
Prompt-biasing delivers accurate punctuation for non-Latin scripts (Japanese 。, Devanagari ।, Arabic ،) and full Latin accents without IME interference.
From universal multilingual powerhouses like large-v3-turbo to featherweight base and tiny, select the ideal balance of accuracy and memory for your Mac or PC.
A seamless floating HUD. Uncompromising control.
Try the live interactive floating pill below, then explore how Verbtm integrates natively with Windows 11.
Verbtm feels right at home on Windows 11 and macOS Apple Silicon. With background system tray / menu bar integration and hardware-accelerated overlays, enjoy fine-grained control over speech models, audio input hardware, visual themes, and global push-to-talk hotkeys.
tiny or base for ultra-low latency, or switch to small for studio-grade accuracy.
How Verbtm stacks up against alternatives.
| Feature | Verbtm | Cloud AI Dictation | Built-in Voice Typing |
|---|---|---|---|
| Audio Privacy | 100% Local (Never leaves device) | Sent to Cloud Servers | Sent to Cloud Servers |
| Subscription Cost | 100% Free & Local ($0) | $15 – $30 / month | Free with OS |
| Air-Gapped / Offline | Yes (Fully Functional) | No (Requires Internet) | No (Requires Internet) |
| Text Injection Method | Direct Keystroke (Win32 & Quartz) | Often overwrites clipboard | Direct Keystroke |
| Hallucination Guard | Smart VAD + Repetition Penalty | Varies by service | Limited |
| Multilingual Dictation | 11+ Languages (100% Offline Auto-Detect) | Cloud-dependent | Requires separate pack downloads |
| Visual HUD | Dynamic Island (Windows & Mac) | Standard Window / None | OS Floating Bar |
Runs on modern laptops, rigs, and workstations.
Smart NVIDIA Acceleration: Verbtm compiles against NVIDIA CUDA 12 and cuDNN 9 for instant tensor core execution on GeForce RTX GPUs, with automatic fallback to AVX2 CPU vectoring for maximum offline performance with zero cloud latency.
Smart Apple Silicon Acceleration: Verbtm is compiled as a pure native ARM64 binary optimized directly for Apple Silicon M1–M5 chips. It executes whisper inference on dedicated ARM NEON vector pipelines with zero cloud latency and no battery impact.
What's New in Verbtm
Recent performance upgrades, accuracy enhancements, and release notes.
Frequently Asked Questions
Yes. Verbtm is 100% free to download and use for all personal and professional workflows. Speech recognition runs offline on your local GPU or CPU with zero cloud subscriptions and zero data collection.
Only once during the very first run to fetch your chosen Whisper speech model (e.g., ~140 MB for base.en). Once cached, Verbtm operates 100% locally and completely offline—no audio streams, telemetry, or transcripts ever leave your device.
Verbtm natively supports 11 top world languages: English, Mandarin Chinese, Hindi, Spanish, French, Arabic, Bengali, Portuguese, Russian, Japanese, and German. With smart language auto-detection enabled, Verbtm identifies your spoken language automatically on the fly without requiring you to manually toggle settings.
Verbtm utilizes native Win32 SendInput keystroke synthesis on Windows and low-level Quartz CGEvent injection on macOS. It types transcribed characters directly into whichever application currently has focus (VS Code, Word, Chrome, Slack, Discord, Xcode) just as if you typed them on a physical keyboard—leaving your system clipboard completely untouched.
On Windows, Verbtm runs smoothly on standard Intel and AMD systems using optimized multi-threaded CPU inference with INT8 quantization (~250–350ms). On Mac, Verbtm runs natively on Apple Silicon (M1–M5) leveraging dedicated ARM NEON vector pipelines for instant sub-150ms transcription without requiring a discrete graphics card.
Verbtm is distributed as a verified standalone application to prevent malicious third parties from creating compromised clones bundled with keyloggers.
Privacy is guaranteed and verifiable directly at the network layer:
base.en) directly to your local drive. After that, speech recognition operates 100% offline.
Operating system security mechanisms (Microsoft Defender SmartScreen on Windows, and Gatekeeper on macOS) flag newly released executables that haven't yet accumulated widespread global download telemetry. Verbtm is completely safe, private, and 100% offline.
xattr -cr /Applications/Verbtm.app in Terminal). When prompted, grant Microphone and Accessibility permissions so Verbtm can hear your voice and inject keystrokes into your active apps.
You can submit feedback, bug reports, and feature requests directly through our Feedback Portal, or right-click the Verbtm tray icon in Windows (or click the menu bar icon on macOS) and select Feedback & Bug Report....
Download Verbtm today. No credit cards, no logins, no cloud tracking.