Yonder — Hack the 6ix 2026

AI-powered spending tracker that turns everyday spending into live, real, bookable travel opportunity costs. Full spec: docs/PRD.md. Shared context: CONTEXT.md.

The WTF moment: Yonder Exchange

We treat Stay22 accommodation inventory as a securities exchange:

• Every destination is a ticker (YBNF, YMTL...) whose price is the cheapest live nightly rate. Stay22 is snapshot-only, so we poll on their 10-min cache and build our own price time series — our market data is genuinely ours.
• The four suppliers (Booking, Vrbo, Expedia, Hotels.com) are market makers: the same room priced differently across them is a bid/ask spread, surfaced as live arbitrage alerts ("same room, $31 cheaper on Vrbo").
• Your cut spending (Chexy feed) is your buying power. The AI coach is the desk analyst. BUY is a real Stay22 affiliate link — every market order pays commission, so the demo is the business model.

One-liner for judges: hotel inventory as a stock market where your bad habits fund the deposits and every order pays Stay22 commission.

Team split (own your folder, don't touch others' — see PRD section 11-12)

• /frontend — Person A
• /backend — Person B (owns Stay22 integration)
• /ai-service — Person C (owns FreeSolo integration)
• /chexy-integration — Person D (owns Chexy integration + Deloitte sustainability logic)

Env setup (do this first)

Each service has a .env.example. Copy it to .env in the same folder and fill in your keys. Everything runs without keys — services fall back to mock/stub mode so you can build immediately:

Service	Key	Without it
backend	STAY22_API_KEY	Serves realistic mock Stay22 results
ai-service	FREESOLO_API_KEY (+ base URL/model in .env.example)	Stub coaching messages
chexy-integration	CHEXY_API_KEY (TBD)	Seeded mock transactions
frontend	none	—

Never commit a real .env — they're gitignored.

Quick start

# Terminal 1
cd backend && cp .env.example .env && npm install && npm run dev

# Terminal 2
cd frontend && cp .env.example .env && npm install && npm run dev

# Terminal 3
cd ai-service && cp .env.example .env && pip install -r requirements.txt && uvicorn main:app --reload --port 5001

# Terminal 4
cd chexy-integration && cp .env.example .env && npm install && npm run dev

Open http://localhost:3000 — the departure board should show live rows fed by backend → chexy-integration spend data, mock Stay22 prices, and a stub coach message. Each service also exposes GET /health.

What's already wired (mock mode)

• chexy-integration GET /spend-summary → categorized mock transactions
• backend GET /api/opportunity → pulls spend summary, matches categories to destinations, prices them (mock Stay22 until a key is set, with the 10-min cache already enforced), calls ai-service /coach for the nudge
• ai-service POST /coach → FreeSolo-trained coaching when keyed, else stubs. Backend sends trusted facts (recoverableMonthly, remaining, estimatedMonths, …)
• frontend → polls the backend, renders the board, falls back to demo rows if the backend is down

Everyone builds against the API contracts in docs/PRD.md section 12. Real integration happens by pointing at live endpoints, not by editing someone else's folder.