# Dicty.io

A free desktop voice-to-text app for macOS and Windows that dictates into any app and formats speech with an LLM into commit messages, PR notes, architecture decisions, or meeting summaries.

Dicty.io is a desktop dictation app that transcribes speech and can format it with an LLM into structured output such as commit notes, pull-request descriptions, architecture decision records, or meeting summaries, syncing the result into a Git workflow. Live dictation can also be typed directly into an editor such as Cursor or VS Code.

Speech recognition runs through local Whisper models, from a small fast model to a larger accurate one, or an NVIDIA model, chosen for a balance of speed against accuracy and shown with benchmarks before downloading; a fully offline mode runs with zero telemetry. For the language-model formatting step, it can call offline local models, such as Qwen, Llama, Mistral, or DeepSeek, or cloud models including Claude, GPT, Gemini, or Grok using the user's own API keys.

The app exposes its model choices, custom prompt modes, integrations, and shortcuts directly in its settings, and a Windows build is in early alpha alongside a Mac build.

Website: https://dicty.io/