Mispher

Mispher transforms your voice into polished text directly on your Mac, offering a suite of on-device capabilities without requiring cloud access or accounts. Key features include: • On-device speech-to-text with multiple language models • Dictation cleanup for improved text quality • In-place text rewriting and translation • A local agent for task planning and tool execution • Customizable controls and user interface This application provides robust transcription services using models like Parakeet EOU, Parakeet TDT (25 European languages), CTC (Chinese), Nemotron (~40 languages), and Qwen3-ASR. Beyond simple transcription, it intelligently cleans up dictated text, corrects punctuation, capitalization, and removes filler words. Users can also select text in any application and speak an edit or translation, which Mispher then applies directly. The integrated local LFM2.5 agent is a powerful addition, capable of planning and interacting with tools, Apple Notes, the clipboard, and local files. It supports MCP tools, allowing for expanded functionality through custom servers. Mispher emphasizes privacy and control, operating entirely on Apple Silicon devices with no data leaving your machine. Designed for anyone needing efficient and secure voice interaction on macOS, from writers and researchers to developers and general users. Its open-source, MIT-licensed nature ensures transparency and community contributions. Future model additions like Gemma 4, Qwen3.6, and Ornith 1.0 are planned.