Can’t post video here, but it’s actually quite polished overall — phaser array fire effects, sound design, the warp-stretch jump effect, and so on.

Star Trek Computer is a desktop voice assistant, but it differs from most Star Trek–themed (and Python-based) voice assistants in several concrete ways. Its entire runtime is Rust: there is no Python, conda, or virtualenv at all —cargo build --release produces a single binary. Wake-word detection, voice-activity detection and speech recognition all run locally, so raw microphone audio never leaves the machine; only the transcribed text goes to the cloud for the LLM and TTS. The audio front-end is genuinely engineered — anti-aliasing, phase-continuous resampling, an adaptive noise floor with RMS gating, and a two-stage VAD — instead of feeding raw mic data straight to a model. It also avoids a classic pitfall by refusing to wake itself on its own TTS playback, and it exposes a diagnostic endpoint (/kws_diag ) so a failed wake-up can be debugged rather than guessed. On top of that it is a full assistant — agent orchestration, skills, long-term memory, scheduled tasks, multi-engine web search and multi-turn conversation — with switchable LCARS/classic themes and a Chinese/English UI

That was exactly my original vision. I’m working on developing a hardware badge, but running the model locally is extremely challenging. TTS however can be migrated to run locally.