Working now
The tray application builds, packages, validates its checksum, and passes 89 credential-free tests. It supports push-to-talk, conditional wake detection, and an xAI provider path.
Project 02 · In progress
Hey Computer
A lightweight Windows voice assistant designed around local wake detection, visible microphone state, and AI services that can be swapped without rewriting the application.
- Status
- Integrated prototype
- Platform
- Windows x64
- Stack
- C# · .NET 10 · WPF
- Automated tests
- 89 passing
Target experience
One clear loop from wake phrase to spoken response.
Push-to-talk remains available when a local wake model is not configured.
- 01Wait locally
Listen for “Hey Computer” on-device without continuously streaming audio.
- 02Capture
Show recording state, normalize the utterance, and stop cleanly.
- 03Respond
Transcribe, call the selected language model, and stream speech playback.
- 04Reset
Return to local listening without retaining history by default.
Architecture
Providers are replaceable. Application state is not.
The coordinator owns the conversation lifecycle while microphone, wake-word, transcription, model, and speech providers stay behind interfaces.
WASAPI audio
→Local wake word
→Conversation coordinator
→STT · LLM · TTS
Privacy boundary
Credentials use Windows protection. Audio, transcripts, and telemetry are not saved by default. Computer control and persistent memory remain intentionally out of scope.
Before release
Real microphone behavior, wake accuracy, a live end-to-end conversation, signing, installation, and update rollback still need validation.
Follow the build ↗