ToolPiper turns speech into text in 127 to 173 milliseconds on the Apple Neural Engine. Hold the right Option key, speak, release - the words land in whatever app you are using. No cloud, no account, no subscription. Your audio never leaves the Mac.
Transcription latency, flat regardless of how long you speak.
Audio sent to the cloud. The model runs on your Neural Engine.
Dictation is on the free tier. No account, no subscription.
| Feature | ToolPiper | Wispr Flow | superwhisper | MacWhisper | Apple Dictation |
|---|---|---|---|---|---|
| Price | Free tier | Subscription | Free + paid Pro | Free + one-time Pro | Free (built in) |
| Runs on-device | Yes (Neural Engine) | Cloud | Yes | Yes | Yes |
| Account required | No | Yes | No | No | Apple ID |
| Push-to-talk hotkey | Right Option | Yes | Yes | Hotkey | Double-tap |
| Types into any app | Yes | Yes | Yes | Paste | Yes |
| Voice commands that run system actions | Yes (right Command) | No | No | No | Limited |
| Part of a full local AI suite (chat, vision, MCP) | Yes | No | No | No | No |
superwhisper and MacWhisper also run on-device and do dictation well. The difference is that ToolPiper dictation is free and part of a full local AI suite.
Dictation writes text. Hold the right Command key instead and ToolPiper routes your speech through a local model that parses intent and runs a macOS action - "turn down the brightness," "open Safari to my email," "turn on Do Not Disturb." Around 142 actions across 26 domains, all on-device. No other dictation app does this.
ToolPiper transcribes dictation in 127 to 173 milliseconds on the Apple Neural Engine, and the latency stays flat regardless of how long you speak because the model pads to a fixed encoder window. That is fast enough that text appears as soon as you release the key, with no cloud round-trip.
It is fully on-device. The speech-to-text model (Parakeet) runs on your Mac's Neural Engine, so your audio never leaves the machine and works with no internet connection. There is no account and nothing to upload.
Dictation is on the free tier. Download the app, hold the right Option key, and speak - no subscription and no account required. ToolPiper Pro is $10/month for the advanced pipeline, agent, and creator features, but dictation itself is free.
Dictation (hold the right Option key) types what you say into whatever app has focus. Voice commands (hold the right Command key) route your speech through a local model that parses intent and runs a macOS system action - changing brightness, opening an app, toggling Bluetooth, and around 142 other actions. Dictation writes text; voice commands do things.
Yes. Dictation inserts text at the cursor in any app that accepts keyboard input - your editor, browser, Slack, Notes, the terminal. It is system-wide, not limited to one window.
superwhisper and MacWhisper also run on-device, and they are good at it. Wispr Flow runs in the cloud and needs an account and a subscription. ToolPiper's difference is that dictation is free and built into a full local AI suite - the same app does chat, vision, OCR, 300+ MCP tools, and voice commands that execute system actions, not just transcription.
Apple Dictation, Whisper, and the on-device option compared.
No cloud, no data collection, no compromise.
Dictate without an internet connection on Apple Silicon.
Dictate into Cursor, VS Code, and the terminal.
Local voice AI or cloud dictation. Where your audio goes.
The privacy and latency case for on-device voice.
Download ToolPiper, hold the right Option key, and speak. Free, on-device, no account.
Download ToolPiper