Voice typing that polishes prose before it hits your cursor

What it does

OpenTypeless is a desktop voice-typing assistant built in Rust and Tauri. Hold a global hotkey, speak naturally, and it transcribes your speech—then optionally routes it through an LLM to rewrite, translate, or answer questions before typing the result into the app that is currently active. It runs on macOS, Windows, and Linux, and supports workflows for dictation, selected-text editing, translation, and one-shot voice Q&A.

The interesting bit

The app detects the active window locally to adapt tone and structure for email, chat, code, or documents, but it only sends an abstract app category and style metadata to the LLM rather than raw window titles. That is a rare privacy-conscious design choice in a category that usually hoover up context indiscriminately.

Key highlights

• Bring-your-own-key support for STT (Deepgram, AssemblyAI, Whisper-compatible, Qwen3) and LLMs (GPT-5/o-series, Claude, OpenRouter), plus a managed-cloud option.
• App-aware polish that reshapes dictation based on whether you are in Gmail, Slack, Cursor, or an issue tracker.
• One-shot “Ask Anything” voice Q&A that returns a concise answer in a floating panel without chat history or input boxes.
• Local dictionary with import/export and per-app style overrides to fix recurring transcription errors and mismatched tones.
• Linux ARM64 packages and Wayland-aware direct typing, though global hotkeys stay conservative on Linux due to Wayland fragmentation.

Caveats

• Linux global hotkey handling is less reliable across desktop environments, especially on Wayland, so the app sticks to conservative Ctrl-based defaults there.

Verdict

Developers who want AI-assisted voice input without surrendering their window titles to a cloud LLM should look here; if you just need basic transcription and never touch an LLM, your built-in OS dictation is probably enough.