Jarvis is a wallet-first personal banking experience that turns a connected MetaMask account into a living balance sheet. The project ships with a dark-mode React dashboard, an Express API, and a Gemini-powered copilot that surfaces allocation ideas, hedges, and playbooks tailored to the wallet in view. Everything runs locally, so you can experiment with AI-assisted portfolio coaching without handing over credentials.

• Audience – Crypto-native investors and advisors who want a banking console that understands wallets instead of checking accounts.
• Value – Unified balances, interactive analytics, and an embedded AI analyst that stays useful even when Gemini is offline.
• Stack – React 19 + Vite frontend, Express + SQLite backend, Gemini 2.5 Flash for natural-language insights.

• Wallet SSO – MetaMask connection creates and restores profiles; no password reset hoops.
• Portfolio Overview – Snapshot cards for crypto, stablecoin, and fiat-equivalent holdings with growth deltas.
• Analytics Studio – Trendlines, allocation breakdowns, and comparison charts powered by Chart.js.
• Transaction Ledger – Clean history of deposits, swaps, and yield claims with placeholders for export tooling.
• AI Copilot – Gemini-backed chat with curated fallbacks; handles prompts, prefill summaries, and context blocks.
• Health + Admin Hooks – Health check endpoint and onboarding flows for quick status validation.

jarvis/
├── backend/                      # Express API and services
│   ├── config/loadEnv.js         # Layered dotenv loader for backend/.env and project-wide .env
│   ├── database.js               # SQLite helpers + schema bootstrap
│   ├── routes/
│   │   ├── auth.js               # Wallet onboarding + profile lookups
│   │   ├── finance.js            # Portfolio, transactions, Gemini chat proxy
│   │   └── orders.js             # Strategy scheduling endpoints (placeholder)
│   └── services/
│       ├── financeService.js     # Portfolio mocks, AI orchestration, fallback logic
│       ├── geminiClient.js       # Gemini REST client (2.5 Flash default)
│       └── stockAnalyticsService.js
├── frontend/
│   ├── src/Dashboard.jsx         # Main signed-in experience
│   ├── src/components/           # Layout, analytics, Gemini panels, etc.
│   └── src/services/api.js       # REST client targeting the backend
└── README.md

• Express API with modular routes (auth, finance, orders).
• SQLite storage via database.js with helpers for queries and seeding.
• financeService composes mocked portfolio data, Gemini calls, and graceful fallbacks when the API key is missing or requests fail.
• geminiClient wraps the Generative Language API with transport fallbacks (fetch when available, https otherwise).

• React 19 + Vite + Tailwind-inspired utility classes for fast iteration.
• Dashboard experience clusters portfolio metrics, charts, market intel, and the AI copilot.
• Componentized layout enables swapping data sources without reworking the entire shell.

Jarvis reads environment variables through backend/config/loadEnv.js, which layers backend/.env over a project-wide .env when present. Create backend/.env with at least the Gemini API key if you want live AI responses.

# Required for live Gemini insights
GEMINI_API_KEY=your_api_key_here

# Optional overrides (defaults shown)
PORT=3001
GEMINI_API_BASE=https://generativelanguage.googleapis.com
GEMINI_API_VERSION=v1beta
GEMINI_MODEL=gemini-2.5-flash
GEMINI_TEMPERATURE=0.25
GEMINI_MAX_OUTPUT_TOKENS=768
GEMINI_TOP_P=0.9
GEMINI_SAFETY_THRESHOLD=BLOCK_MEDIUM_AND_ABOVE

Tip: If you omit GEMINI_API_KEY, the finance service automatically falls back to curated suggestion bundles so the UI remains useful offline.

cd backend
npm install
npm start          # launches on http://localhost:3001

The backend initializes the SQLite database (jarvis.db) on first run and logs key endpoints, including a health probe at /api/health.

cd frontend
npm install
npm run dev        # serves the React app on http://localhost:5173

Visit http://localhost:5173, connect MetaMask, and follow the onboarding prompts. The frontend hits the backend REST API at http://localhost:3001 by default (update frontend/src/services/api.js if deploying elsewhere).

1. Frontend posts prompts, historical messages, wallet context, and optional overview snapshots to /api/finance/ai-suggestions.
2. financeService assembles system instructions, generation config, and safety settings before calling generateContent in geminiClient.
3. geminiClient targets the configured model (default gemini-2.5-flash) and handles transport via native fetch or Node's https.
4. Responses are normalized into { message, note, meta, context } payloads. If Gemini is unreachable or returns an empty candidate, curated fallback suggestions are injected instead of failing the chat experience.

• Database Resets – Use the helper functions in backend/database.js or the provided seed scripts to reset sample data.
• Static Analysis – Add your preferred linter/prettier setup; the project intentionally stays lightweight.
• Testing – Endpoints are simple enough for supertest or MSW harnesses; consider adding regression coverage as you move beyond the MVP.
• Environment Safety – loadEnv only sets values when they are missing, so explicitly exporting variables in your shell will override .env entries.

1. Persist real user portfolios in a managed Postgres or Supabase instance.
2. Replace mocked balances with on-chain lookups, custodial APIs, or accounting integrations.
3. Expand the copilot with scenario planning, alerts, and streaming responses.
4. Add advisor dashboards, permissions, and collaborative planning tools.
5. Harden the auth flow with session tokens, rate limits, and audit logging.

This MVP is provided for experimentation and product exploration. Use at your own discretion.