RCLI

Why we included this project

For anyone running macOS on Apple Silicon, RCLI turns the terminal into a voice-controlled assistant that can actually do things: dictation, app launches, volume and Spotify control, and asking questions about your own PDFs or DOCX files. The whole pipeline, speech-to-text, local LLM, text-to-speech, and vision for camera or screen analysis, runs on-device with no cloud and no API keys. A hybrid retrieval layer handles document search, and the sub-200ms latency claim applies to the recommended hardware, which is an M3 chip or newer; on older Macs it falls back to llama.cpp and runs slower but still locally. That privacy angle matters if you do not want conversations or documents leaving the laptop. It is also a useful example of a complete Metal-based STT+LLM+TTS stack, if you are building something similar.