🔧 PitCrew

Four real Auth0 accounts. Every one uses the same password:

PitCrew-Demo-2026!

Three separate customers, each pre-seeded at a different repair stage — so you can open any screen in the workflow directly, instead of clicking through the whole flow to reach it.

A fourth customer has no login at all. Priya Nair's Ford F-150 appears on the adviser's dashboard but belongs to no demo account — sign in as any customer and you'll find it unreachable. That's the multi-tenancy boundary, and you can check it yourself.

⚡ Fastest path to the money shot: sign in as [email protected] and pay. Stripe is in test mode — card 4242 4242 4242 4242, any future expiry, any CVC.

👥 See both sides at once: open the adviser in one browser profile and a customer in another. State is shared through Redis, so approving on one screen updates the other.

♻️ Reset everything: /api/demo/reset restores all four accounts to their starting stages.

1. Sign in as the adviser → land on a real repair-order dashboard
2. Open the 2019 Honda Accord → see the raw technician note ([RED] DIAGNOSE CHECK ENGINE LIGHT ON…)
3. Hit Generate PitCrew Report → AI rewrites it in plain English, priced from the shop's service menu
4. Click Find parts on any line item → AI writes a parts query, real vendors and prices come back
5. Send to customer
6. Sign in as [email protected] → approve/decline each item, watch the total update
7. Pay → Stripe Checkout → order flips to paid on both screens

Short on time? Sign in as [email protected] — that account is already at step 6.

🤖 Jargon → plain English. Technician shorthand becomes a report written at an 8th-grade reading level: what we found, what you'll notice, what happens if you wait. Never invents prices — those come from the shop's own service menu.

🛒 Live parts sourcing. For each repair, an AI writes a parts-store query from the symptom ("misfire on cylinder 2" → "2019 Honda Accord iridium spark plugs set"), then pulls real listings from Google Shopping — NAPA, RockAuto, AutoZone, Walmart — scoped to the shop's city. The adviser attaches one and sees their margin.

✅ Line-item approval. The customer approves or declines each repair individually. The total is always recomputed server-side — the client never sends a price.

💬 Page-aware AI assistant. Answers questions grounded in what you're actually looking at. Context is built server-side from your own orders only, so the model structurally cannot see another customer's data.

🔒 Two real roles. Advisers see the shop; customers see only their own cars. Requesting someone else's order returns 404, not data.

Auth0, OpenRouter and the Vercel project were provisioned through Stripe Projects.

Every external call has a fallback, and none of them show the user an error:

• AI report → 20s timeout, then a pre-generated report from the seed
• Parts search → missing key, rate limit, or zero results → seeded offers, marked "Showing saved results"
• Parts query → if the model is slow, a deterministic query from the service menu
• Stripe → clean demo mode when no key is present
• Redis → falls back to an in-process store for local dev

The service menu also pins each finding's parts query, so the same demo returns the same results every run.

Next.js 16 (App Router, Turbopack) · React 19 · TypeScript · Tailwind v4 · shadcn/ui · server actions as the entire write API

src/lib/ai.ts        raw notes → structured report
src/lib/parts.ts     finding → parts query → live vendor offers
src/lib/catalog.ts   shop service menu: prices findings, pins parts queries
src/lib/store.ts     Redis-backed shared state
src/app/actions.ts   every write, session-checked

npm install
cp .env.example .env      # fill in the keys you have
npm run dev

Auth0 keys are required — every screen sits behind a login. Everything else is optional: without Stripe, OpenRouter or SerpApi keys the app still runs end to end on seeded data, so you can light up live features one at a time.