Private voice dictation for developers and product teams

Where dictation saves keystrokes

• Issue reports: describe reproduction steps while the failure is visible.
• Reviews: draft the reasoning behind a requested change before tightening the wording.
• Documentation: capture a first explanation, then edit commands and identifiers by hand.
• Status updates: turn a spoken progress summary into text at the active cursor.
• Support: answer recurring questions without retyping complete sentences.

Designed for prose around code

Speech recognition is useful for natural language, not for reliably producing punctuation- dense source code. Secretary42 works best for the prose around engineering: intent, trade-offs, test results, documentation and collaboration. Dictate the first draft, then use the keyboard for exact symbols, paths, API names and formatting.

Local processing for sensitive context

Dictation may contain unreleased product names, internal incident details or customer information. Secretary42 processes microphone audio and transcript text on the device with whisper.cpp and has no cloud-transcription fallback. Local history can be viewed, copied or deleted from the app.

The app is not wholly disconnected: licensing, model downloads, updates, checkout, support, and optional diagnostics are separate online operations. None is an alternate transcription route. Review the Privacy Notice before using any dictation tool with confidential material.

What this workflow does not include

Secretary42 does not interpret “Hey agent” commands, rewrite text with a reasoning model, read the surrounding editor context, or operate development tools. It supplies a faithful local transcript at the cursor; the developer remains in control of the final text and any action taken from it.