SheetDiff

Upload two spreadsheets (bank vs books). Get Matched / Only A / Only B plus the dollars that don’t tie.

Stack: Next.js 16 + Supabase (Auth/DB) + Paddle Billing + Ollama (optional AI).

Spreadsheet rows stay in the browser. Supabase stores accounts, plan quotas, and run metadata only.

Quick start

1. Create a Supabase project.
2. In SQL Editor, run supabase/schema.sql.
3. Authentication → Providers → enable Google.
   Redirect URL: https://YOUR_PROJECT.supabase.co/auth/v1/callback
   Also add site URL / redirect: http://localhost:3000/auth/callback
4. Copy .env.example → .env.local and fill keys.
5. Install & run:

npm install
npm run dev

Open http://localhost:3000.

Product surface

Route	Purpose
/	Product Hunt–style landing
/pricing	Free / Starter $9.9 / Pro $29.9 + Paddle checkout
/login	Continue with Google (+ email OTP)
/login/complete	Confirm name after first Google login
/app	Drop files → map columns → three tabs
/account	Plan, usage, portal, sign out
/api/webhook	Paddle subscription sync → Supabase

Plans

• Free — 1 lifetime run, 20-row preview, no full export
• Starter $9.9/mo — 5 runs, 5k rows/side, full export, 1 saved rule set
• Pro $29.9/mo — 40 runs, 25k rows/side, AI mapping via Ollama, audit export

Platform covers LLM cost for Pro; point OLLAMA_* at local qwen3:8b for now, swap to a cloud model later.

Paddle webhook

Point sandbox notifications to your tunnel, e.g. https://xxxx.ngrok.io/api/webhook.

Reuse the working flow from E:\Workspace\PaddleIntegration. Create SheetDiff Starter/Pro prices and put the pri_… IDs in .env.local.

Privacy note

Matching runs entirely client-side. Only filenames, row counts, and summary stats are written to recon_runs.