macOS menu bar app

Voice to polished text,
right at your cursor.

Hold a hotkey, speak naturally, release — ready-to-send text appears wherever your cursor is. No app switching. No copy-pasting.

$ curl -sSL /install.sh | sh

Requires Python 3.10+ · macOS 12+ · OpenAI & Anthropic API keys

1 Hold Alt and speak
2 Release to process
3 Polished text at cursor

Features

Built for people who write a lot

Best-in-class transcription meets best-in-class enhancement — under your own API keys.

Works everywhere
Terminal, IDE, browser, Slack, email, Notion — anywhere there's a text cursor.
Your API keys
No middleman servers. API calls go directly from your machine to OpenAI and Anthropic.
One-line install
Install with a single curl command. Updates with pip. Fits in your dotfiles.
Neural VAD
Silero VAD trims silence with a neural network. Works on whispers and noisy environments.
Custom prompts
Override any mode with your own prompt files. Add context.txt with your name and writing style.
Hot-reload config
~/.vaani/config.yaml is watched for changes. Edit and save — no restart needed.
Encrypted history
Transcription history stored in a local SQLite database with AES-256 encryption.
Keychain storage
API keys stored in macOS Keychain — never written to disk in plaintext.

Pipeline

From voice to polished prose in ~2 seconds

Every step runs in the background. The menu bar icon shows your current state.

🎤
Mic capture
sounddevice
VAD trim
Silero VAD
Gain norm
RMS-based
Transcribe
Whisper API
Enhance
Claude Haiku
Paste
pynput

Enhancement modes

Match your tone to the context

Switch modes from the menu bar. Each mode sends a different system prompt to Claude.

Minimal
Least changes possible. Fix grammar, remove filler words, preserve your original wording.
Professional
Formal rewrite for business emails, reports, and async communication. No contractions.
Casual
Friendly, conversational tone. For Slack, DMs, and informal writing.
Code
Code-aware formatting, shell command recognition, and code generation from speech.
Funny
Witty, humorous rewrite that still delivers your core message.

Override any mode with your own prompt file in ~/.vaani/prompts/modes/


Comparison

Where Vaani fits in

Same output quality as subscription tools — at a fraction of the cost.

Tool Price AI Enhancement Custom Prompts Your API keys
Vaani ~$0.50/mo API costs 5 modes + custom File-based
Wispr Flow $15/mo Context-adaptive
SuperWhisper $8.49/mo or $249 Multi-model GUI-based ~ Optional
Whisper Dictation Free Local only
macOS Dictation Free

Privacy

No Vaani servers. Ever.

All cloud calls go directly from your machine to OpenAI and Anthropic under your own account.

Audio
Sent to OpenAI
Under your API key. Subject to OpenAI's data retention policy.
Transcribed text
Sent to Anthropic
Under your API key. Subject to Anthropic's data retention policy.
API keys
macOS Keychain only
Never written to disk in plaintext. Stored in the system credential store.
History
Local SQLite + AES-256
Stored on your machine. Encrypted with Fernet. Nothing leaves your disk.

No subscription. No servers. Just your voice.

API costs run ~$0.50/month at typical usage. You own your keys.

Copy install command View on GitHub