Your microphone is not a product.

1. Privacy you do not have to read a policy to get

When recognition runs on your device, there is no recording on anyone else's server to retain, no transcript stored in someone's data center, and no third party to subpoena. Lawyers, clinicians, journalists and anyone under an NDA can dictate without checking a compliance matrix first.

2. Latency you can feel

Round-trips to a data center add a beat between speaking and seeing text. On a modern laptop the local model keeps up as you talk, and it keeps working on a plane, in a basement, or on hotel Wi-Fi.

3. Costs that do not scale with your words

Cloud recognition costs the vendor per minute, which is why cloud dictation tends to be a subscription with tight caps. Local recognition costs nothing per word, which is why Pro has no word cap and no per-minute meeting limit.

The honest limitation

Optional features that use a large cloud language model do send text off the device if you turn them on: a transform you point at a cloud provider, and, in Pro, Cloud Polish and meeting summaries, which send the text of a dictation or a meeting transcript to Vaitly's server and OpenAI. Vaitly Voice labels every one of them, audio never goes with them, and the default rewriting runs locally, on a Mac or a Windows PC.

Read the privacy policy for the exact list of what we do and do not process.