N.O.V.A v2 / documentation
Roadmap & Status
What is built, what is being tested, and what is planned for the desktop and Android versions.
Where things stand
| Item | Status | Notes |
|---|---|---|
| Desktop prototype | In progress | Python assistant on the original N.O.V.A codebase, with Ollama so it runs offline. |
| Cloned voice | Testing | XTTS-v2 (coqui-tts) for a voice cloned from a recording; Piper if GPU memory is too tight. |
| faster-whisper + pyannote | Planned | Replaces the current speech recognition and adds speaker identity. Not applied to the codebase yet. |
| Free-form requests | Planned | The local LLM understands plain requests instead of relying on hardcoded commands. |
| Native Android app | Planned | Standalone app with its own model, wake word, and speech-to-text. Android only. |
| Django backend | Exploring | Rebuilding the backend on Django is under consideration. |
Honest note
Anything marked Planned or Exploring is not in the repository yet. The pages in these docs describe the intended design.
Decisions so far
- Fully offline. Everything runs on the device, which is also why the wake word matters: a task can be done the moment you think of it.
- Desktop first. The laptop version is built first as a testing harness, and works on Windows, macOS, and Linux.
- Native Android second. The phone gets a standalone app with its own on-device model, wake word, and speech-to-text, not a thin client for the laptop. Android only; there is no plan for iOS.
- One assistant, two devices. Both run in the background and behave largely the same, each independent of the other.
- Plain-language requests. The goal is to move away from a fixed command list toward free-form requests, as with ChatGPT or Claude.
For how the pieces fit together, see Stack & Tools and Voice Lane.