Verba turns your voice into clean, formatted text — using
WhisperKit and on-device LLMs. No cloud API, no account,
no data uploaded. Ever.
Hold a key. Speak. Release. It's already typed.
Verba doesn't phone home. Speech recognition runs on WhisperKit
via Apple Neural Engine. Text formatting uses MLX models loaded
into local memory. There's no server, no analytics, no telemetry.
It works on airplane mode. That's the whole point.
Voice input is messy — fillers, repetition, broken sentences. Verba's on-device LLM cleans it up while preserving what you meant. No cloud API key needed.
Choose from built-in prompts (general, meeting notes, email) or write your own.
Supports Qwen3, Gemma 3, and SmolLM3
models via Apple MLX.
Hold fn and speak. When you release, Verba transcribes your audio, optionally cleans it up with an LLM, and pastes the result into whatever app you're using.
The floating indicator shows real-time streaming preview so you always know what's being recognized. Tap the stop button to confirm or the discard button to cancel — icon-only controls with instant tooltips keep the UI clean. Minimize to a slim bar when you need to see content behind it, click to expand back. Works in any text field — editors, Slack, email, browser.
Speech-to-text on Apple Neural Engine
On-device LLM for text formatting
Auto-detect or set manually
Quick-switch mode & prompts, recent history
One-time purchase. No subscription. No hidden fees.
Verba is open source. You can also build from source for free.
Open source, on your terms.
macOS 14+ · Apple Silicon · ~75 MB