OpenVerba quietly checks for a newer version and notifies you when one is ready. Choose 'Check for updates' and it downloads the new version, verifies it with SHA-256, and installs it — no admin rights needed, and your settings and downloaded model are kept. The automatic check can be turned off in Settings.

Free · 100% local · open source

Voice typing that never
leaves your computer.

Press a key and speak — OpenVerba types your words into whatever app you're using, live as you talk. Hold the key instead and it edits your text by voice. Every word is transcribed locally on your own PC — no cloud, no account, no subscription.

Download for Windows See how it works

Free forever · Windows 10 & 11 · No sign-up · MIT licensed

0words sent to the cloud

$0free forever

Anyapp you can paste into

100%runs on your machine

How it works

Three steps. No setup after install.

OpenVerba sits quietly in your system tray and waits for your trigger. From press to typed text, everything happens on your own machine.

1
Tap to talk, hold to edit

Pick a key combo (like Ctrl+Shift+Space) or a mouse button. Tap it to start dictating; hold it for a second to give a command or edit text with your voice.
2
Speak naturally

Your microphone audio is transcribed on your own CPU or GPU. The audio is processed and then discarded — never saved, never uploaded.
3
It types anywhere

The text drops into whatever window is focused — your editor, browser, chat box or terminal — then your clipboard is restored exactly as it was.

Features

Everything you need. Nothing you don't.

A focused dictation tool that respects your privacy, your hardware and your wallet.

100% local & offline

Audio is transcribed on your PC and discarded. Nothing is sent to any server. After the model downloads once, OpenVerba works fully offline.

Works in any app

OpenVerba types into the focused window by pasting, so it works everywhere a normal paste works — editors, browsers, chat, even the terminal.

GPU-accelerated

Uses your NVIDIA GPU for near-instant transcription when one is present, and falls back to the CPU automatically when it isn't.

Preserves your clipboard

Your previous clipboard — text, images, files, even HTML — is saved before OpenVerba pastes and restored right after. Dictation never clobbers your copy.

Smart model picker

On first run it scans your GPU, VRAM, CPU and RAM, then recommends the best speech model for your machine — with a clear comparison of size, speed and accuracy.

Live as you speak

Turn on real-time mode and watch your words appear the moment you say them — it only commits text it's sure of, so it never rewrites what's already on screen. Prefer to review first? Classic press-speak-paste still works.

Voice commands

Hold your trigger and just say what to do — "delete last word", "select all", "press enter", "undo" — and OpenVerba runs it instead of typing it out.

Edit with your voice

Select some text, hold your trigger, and say how to change it — "make this sound more natural", "fix the grammar", "delete the last sentence." A local AI model rewrites it in place. Optional and 100% on-device.

Flexible triggers

Use a keyboard combo, a mouse button, or a two-button chord. The live trigger picker shows what it detected and warns you if a choice is conflict-prone.

Stays up to date

OpenVerba checks for new versions and updates itself in a click — the download is verified before it installs, no admin rights needed, and your settings and model carry over.

Free & open source

No account, no trial, no subscription. OpenVerba is MIT licensed — read the code, build it yourself, or send a patch.

On-device AI

A speech model that runs on your computer.

OpenVerba uses faster-whisper, an optimized build of OpenAI's Whisper speech model — running entirely on your own hardware.

Downloaded once, then yours. The model downloads a single time on first run and is cached on your PC. After that, no internet is required.
Right-sized for your machine. OpenVerba detects your hardware and picks a model that fits — from a tiny CPU-friendly build to a high-accuracy GPU model.
GPU when you have one. With an NVIDIA GPU it uses CUDA for near-instant results; without one it runs on the CPU.
Truly offline. Flip on offline mode and OpenVerba never touches the network again — your voice and your text stay on the device.

Edit by voice

Don't just dictate — edit out loud.

Hold your trigger instead of tapping it and OpenVerba switches from typing to doing. Run quick commands, or have a local AI model rewrite whatever you've highlighted — without ever touching the keyboard.

Rewrite what you've selected. Highlight a sentence, hold, and say "make this clearer" or "sound more professional" — it's replaced in place.
Fix and reshape, in plain English. "Fix the grammar", "remove the word actually", "turn this into bullet points."
Hands-free commands. "Select all", "delete last word", "press enter", "undo" — the mechanical stuff, spoken.
Local, free, and optional. The AI runs on your own machine — no cloud, no API keys, no cost. Switch it on in one click; dictation works great without it.

Models

Pick your trade-off: speed vs. accuracy.

OpenVerba recommends a model automatically, but you can switch any time in the built-in model manager. Bigger models are more accurate; smaller models are faster and lighter on your hardware.

Model	Languages	Disk	Accuracy	Speed	Good for
`tiny.en`	English	~75 MB	Low	Fastest	Old / low-power CPUs
`base.en`	English	~145 MB	Good	Fast	CPU-only default
`small.en`	English	~480 MB	High	Fast	4 GB+ GPU default
`distil-small.en`	English	~340 MB	High	Fastest	Low-latency English
`medium.en`	English	~1.5 GB	Very high	Medium	6 GB+ GPU (fits ~4 GB tight)
`distil-large-v3`	English	~1.5 GB	Very high	Fast	Best high-end English
`large-v3`	Multilingual	~3.1 GB	Best	Slow	6–10 GB+ GPU, multilingual

No GPU? base.en is the sweet spot on CPU. A GPU makes the larger, more accurate models practical for real-time dictation.

Requirements

What you need to run it

Windows 10 or 11

Installs per-user — no administrator rights needed. A macOS and Linux build is in progress.

A microphone

Any mic Windows recognizes works — built-in, USB, or headset.

Optional NVIDIA GPU

Speeds up transcription dramatically. Not required — OpenVerba runs on the CPU too.

FAQ

Questions, answered

Is it really free?

Yes — completely free, with no account, trial, or subscription. OpenVerba is open source under the MIT license.

Does my voice get sent anywhere?

No. Your microphone audio is transcribed locally on your CPU or GPU and then discarded. It is never written to disk or uploaded. The only time OpenVerba uses the network is to download the speech model once on first run — after that you can run it fully offline.

Do I need a powerful computer?

No. OpenVerba runs on the CPU with a lightweight model on ordinary machines. If you have an NVIDIA GPU it'll use it automatically for faster, larger, more accurate models — but a GPU is optional.

Which apps does it work in?

Any app you can paste into — documents, browsers, chat apps, IDEs, even the terminal. OpenVerba inserts text into whatever window is focused.

Will it overwrite what I copied?

No. OpenVerba saves your existing clipboard — including images, files and formatted content — pastes the transcript, then restores your clipboard exactly as it was.

Can I change the trigger?

Yes. Use a keyboard combo, a single mouse button, or a two-button chord. A live picker shows you exactly what it detected and warns you if a choice might conflict with normal clicking. Tap the trigger to dictate; hold it to switch into command and editing mode.

Can it edit text I've already written?

Yes. Select some text, hold your trigger, and say how to change it — like "make this sound more natural", "fix the grammar", or "delete the last sentence" — and a local AI model rewrites it in place. It's optional and runs entirely on your own machine, with no cloud, account, or cost.

How do updates work?

OpenVerba quietly checks for a newer version and lets you know when one's ready. Choose "Check for updates" and it downloads the new version, verifies it (SHA-256), and installs it for you — no admin rights needed, and your settings and downloaded model are kept. You can turn the automatic check off in Settings.

Does it need admin rights to install?

No. OpenVerba installs per-user to your local app data, so no administrator rights are required. It also won't run inside elevated windows unless you run OpenVerba as admin too.

Is there a Mac or Linux version?

Windows is available today. A cross-platform build for macOS and Linux is actively in progress.

Start dictating in minutes.

Free forever, private, and entirely yours. Download OpenVerba for Windows and talk to any app.

Download for Windows View source on GitHub

OpenVerba 1.0.0 · Windows 10 & 11 · ~73 MB · MIT licensed