What it's for
For people who write all day on a Mac, in e-mail, notes, messages and documents, and would rather speak than type. Plain dictation returns raw text that still needs fixing: punctuation, capitalisation, filler. Dictorio returns text that is ready to use, in the app you are already working in, and your voice never leaves the Mac.
How it works
- 1. Hold fn and speak. A small panel shows that Dictorio is listening.
- 2. Speech recognition on the Mac. Whisper turns speech into text locally.
- 3. Correction. A language model fixes punctuation, capitalisation and wording for the current context. On Macs with 16 GB of memory or more, this model runs locally by default.
- 4. Text at the cursor. The result is inserted where your cursor is.
Also in the app
- Context modes. Plain correction, e-mail, note and message, matched to the app you are in.
- Command mode. Select text, hold right ⌘ and say what to change.
- History. Earlier dictations stay available to copy again.
- Personal dictionary. Your names and terms are passed to both speech recognition and correction.
- Offline by default. Recognition and local correction need no network connection.
- Optional external model. Only if the user connects their own model account. The credential is stored in the macOS Keychain, the connection goes directly from the Mac to the provider, and text leaves the Mac only after the user has switched this on.
Technologies
- Swift, native macOS application
- whisper.cpp running Whisper (Turbo, Medium or Small) for speech recognition
- llama.cpp running Gemma 4 E4B for correction
- Models downloaded inside the app and stored on the Mac
- macOS Keychain for the optional external-model credential
- Distributed from dictorio.com outside the Mac App Store, because the App Store sandbox does not allow the system-wide key handling that push-to-talk needs. The public release will be signed with Developer ID and notarized by Apple.
- Interface in English, Polish, German, Spanish and French
Where AI is used
- In the product
- Speech recognition (Whisper) and text correction (Gemma), both on the Mac. An external model only if the user connects one.
- In how it was built
- An AI-agent engineering team with independent read-only reviewers, under human approval. The local correction model was chosen by a measured evaluation and checked by an independent verifier before it shipped.
Where it stands
Dictorio is in beta; a public release is planned. Product support is available through dictorio.com.