Built around your words.
And your way of working.

From a thought to a text field.

Hold a shortcut, speak, then release. A native engine creates the final transcription; streaming models can show a preview while you speak.

• A compact recording indicator with visible listening and processing states.
• Cancellation and a recoverable result when insertion cannot be confirmed.
• Original recognition stays separate from optional edits.

The right amount of polish.

Literal keeps generative rewriting off. Clean and Polished can use an optional local model, with checks around important values and meaning.

• Protect numbers, URLs, email addresses, paths and dictionary terms.
• Keep the original when cleanup fails or a change is rejected.
• Download a cleanup model only if you choose to use one.

A vocabulary that feels familiar.

Save preferred spellings, aliases and spoken forms for names, projects and technical terms. Scope entries to a speech language, or use them everywhere.

• Dictionary changes take effect without rebuilding the speech model.
• Save complete phrases as voice-triggered snippets.
• Date, time and explicitly permitted context can fill snippet variables.

A useful memory. On your terms.

Search local history, copy the original or final text, and revisit a saved result. Audio retention is a separate choice.

• Incognito and per-app exclusions control retention.
• Choose a retention period, delete entries, or clear history.
• Reprocess saved text, or retained audio, through the available local paths.

Choose simply. Tune when you need to.

Automatic prepares a compatible installed model for your language and available hardware. Manual selection remains available, with explicit downloads and checksum verification.

• Whisper models offer different storage and memory footprints.
• Nemotron provides genuine streaming on supported runtimes and hardware.
• Parakeet offers an additional offline transcription path.
• Repair, unload and remove model files from the app.