Personal Dashboard

A Tauri 2 + React/TypeScript AI OS (guided vibe coding): a multi-agent assistant over personal data, protected by defined permission levels.

Windows setup

1. Install Node.js 22+ and the Rust toolchain (rustup, MSVC target — the installer will prompt for Visual Studio Build Tools if missing). WebView2 ships with Windows 11.

2. npm install
   npm run tauri dev

   First launch compiles the Rust shell, then opens the window, creates the database in %APPDATA%, and runs migrations.
3. In Settings, paste a Gemini API key (free tier), or point at a local Ollama — and save.
4. Start a chat from a preset. Documents you drop as PDFs become searchable by the knowledge agent under the permission level you select.

npm run tauri build (installer) additionally needs app icons: run npm run tauri icon <path-to-1024px-png> once first.

Verification checkpoints

Cloud/CI sessions can't open a Tauri window, so these three checks run on Windows:

#	After	Check
W1	setup	App opens; DB file exists; startup FTS5 check passes (app fails loudly if not)
W2	adding a key	A Gemini chat streams end-to-end; token meter fills in after the reply
W3	multimodal	Pasted screenshot described; dropped PDF answerable via Study preset; mic button types for you

Everything else is machine-checkable anywhere:

npm run typecheck
npm test
npm run build
npm run eval

Evals

GOOGLE_GENERATIVE_AI_API_KEY=... npm run eval
EVAL_OLLAMA_MODEL=qwen3 npm run eval

The router suite is the regression gate: ≥ 85% (npx evalite run --threshold 85) before shipping any model or prompt change. npm run eval:watch opens the local UI.

Backups

All state is one SQLite file plus attachments, in the app-data dir (%APPDATA%/com.jmvaz.dashboard/). Copy dashboard.db somewhere to backup.