[11]entry 11 of 29
Brainiac
this entry is written twice, in full
- 1.in brief
voice-activated AI assistant in your terminal
- 2.in general
Press space, speak, get a spoken answer back.
- 3.in particular
A real-time voice↔AI loop in Rust: mic → Deepgram STT → DeepSeek → Aura TTS, in a ratatui TUI.
- 1.in brief
a voice-controlled AI assistant that lives in a terminal
- 2.in general
Press space, talk, and get a spoken answer back.
- 3.in particular
A full loop from your voice to an answer and back out loud, running in a text window.
note
A full voice pipeline in the terminal. cpal3 captures PCM on a dedicated thread (the stream is !Send on CoreAudio), Deepgram6 streams transcripts over WebSocket, DeepSeek7 streams tokens, and Aura reads the reply aloud, with a live waveform TUI and a minimal SSH-friendly mode. The audio channel doubles as a lifecycle signal for clean shutdown without cancellation tokens.
note
The whole voice loop, in a terminal. The microphone is captured on its own dedicated thread, because on a Mac that recording cannot be handed between threads at all. What you said is transcribed as you say it rather than after you stop, the answer is written out word by word as it is thought of, and it is read aloud. There is a live waveform to watch, and a stripped-back mode for when you are working on a machine somewhere else. The audio channel doubles as the shutdown signal, so closing it cleanly needs no separate machinery.