| layout | default |
|---|---|
| title | Nova Assistant Docs |
On-device AI assistant powered by Gemma, built with Flutter. All inference runs locally — no chat data is sent to external servers.
| Page | Description |
|---|---|
| Getting started | Clone, setup, run, and build |
| Architecture | Services, inference pipeline, native bridges |
| Models | Built-in models, diffusion, import, HuggingFace |
| Remote LAN inference | Stream large/GGUF models from a PC on Wi‑Fi |
| Tools & MCP | Device tools and external MCP servers |
| Assistant mode | Default assistant, MediaProjection, PROJECT_MEDIA |
| Contributing | Style, tests, CI, commits |
| Roadmap | Planned features and release phases |
| Feature plan | Living feature plan and next work |
| AGENTS.md | Full agent coding guide |
Developers and coding agents can use the project skill:
.cursor/skills/nova-dev/SKILL.md (nova-dev)
Also mirrored for other agents under .agents/skills/nova-dev/ and .claude/skills/nova-dev/ (sync with scripts/link-nova-dev-skill.ps1 / .sh).
Say in any agent tool:
use the nova_dev skill to setup and build the app
or
use nova_dev to configure my own model
git clone https://github.com/involvex/nova-assistant.git
cd nova-assistant
flutter pub get
flutter run -d androidRequirements: Flutter >=3.17.0-0.1.pre (see pubspec.yaml), Dart matching
the Flutter SDK, Android API 26+, arm64 device recommended.
- Inference is 100% on-device (unless you enable Remote LAN)
- No analytics by default
- Models from trusted HuggingFace sources