LazyTyper is a lightweight desktop app for voice typing — speaking instead of typing — that lets you pick from 12 different AI speech recognition models. These include cloud services such as ElevenLabs, Groq Whisper, Mistral Voxtral, AssemblyAI and DouBao Voice, plus five local engines that run completely on your device with no audio leaving your computer. You can switch engines at any moment depending on whether you want maximum speed, the best multilingual accuracy or total privacy for sensitive work.
The app is designed to be fast and unobtrusive: it sits in the background, uses very little memory and works system-wide, so you can dictate into emails, documents, chat apps and code editors. LazyTyper claims around 90% of dictation needs no corrections and that speaking is roughly three times faster than typing. It also handles mixed-language sentences — for example switching between Chinese, English and Japanese mid-sentence — which is a common pain point with standard operating-system dictation.
LazyTyper is free to download and use with no ads, and all 12 models are available at no cost. It runs on Windows and macOS (direct download) as well as Linux (packages via GitHub). Cloud models require an internet connection because audio is sent directly to your chosen provider's API rather than through LazyTyper's servers; local models work offline. Settings and API keys are stored locally.
