Opening the build…
Anna turns speech into clean, formatted text in the desktop app you are already using.
Inside the build
Raw transcription is only part of useful dictation. Spoken text still needs punctuation, formatting, and the right names—and it needs to arrive in the application where the work is happening. Anna brings that complete interaction together.
Hold a hotkey, speak, and release. Anna captures audio, transcribes it, cleans up the text, and inserts the result into the focused app. We combined existing speech models with a fast language model, then built the native desktop behavior, onboarding, billing, and update system around them.
The hotkey starts recording without a trip to another window. Transcription and cleanup run as a pipeline before text is inserted into the focused application. macOS and Windows each have platform-specific integration work behind that simple interaction.
Open full capture ↗Inside the buildIn a recorded March 7, 2026 benchmark on an approximately 340-word clip, Claude cleanup took about 2.3 seconds. The full pipeline took about 10 seconds. This is one recorded benchmark, not a guarantee for every recording.
Anna orchestrates speech-to-text providers rather than training a transcription model. That keeps the product work focused on the experience, vocabulary, and desktop integration.
Reliable text insertion needed accessibility helpers, focus handling, and a platform adapter for Windows. Browser-based selected-text capture remains an application-specific limitation.
The supplied September 2, 2026 intake records real team use and a shipped macOS/Windows product. The git history places the Windows port ten days after the first commit. That is a development milestone, not a claim that the whole product was finished in ten days.
The benchmark describes one recorded test, not a guaranteed time for every dictation. No before-and-after productivity baseline was captured.