Preheat
Order management for home bakers.
One place for home bakers to track orders, see real cost and profit, know exactly what to bake and buy, and keep customers in the loop — instead of juggling spreadsheets, notes apps, and DM threads.
Built for home bakers (cottage-food operators) and piloted with Tart Theory, a real Seattle egg-tart bakery taking weekly pre-orders with Friday/Saturday pickups. The owner shared her real order and cost sheets, demoed her weekly routine, and co-designed the app with us.
Live demo: home-bakery-lovat.vercel.app/welcome (sample data, anonymized customers)
Why This Matters
Tart Theory's patchwork is the norm, not the exception:
- Solo businesses are being left behind on AI. Only 15.3% of businesses with no employees pay for AI tools, vs. 26.1% of businesses with employees (Dec 2025), and the gap holds even at similar revenue. Researchers suggest time and capacity, not cost, may be the barrier. (JPMorganChase Institute, 2026)
- Food businesses are among the slowest adopters. AI adoption in accommodation and food services is 2.9%. (Brookings, cited by JPMorganChase Institute)
- Orders are scattered across channels. Small businesses (2–99 employees) sell through a median of 3 channels; 28% use 5 or more, 51% sell through social media, and 47% plan to add channels next year. (SBE Council Small Business Technology Use Survey, 2026)
- Accuracy is the top worry, and time is the payoff. 45% of small businesses cite accuracy and reliability as a concern about AI, while those using it report saving a median of 5 hours of the owner's time per week. (SBE Council, 2026)
That's why Preheat keeps every number in plain code, uses Claude only for messy human input, and connects to the tools bakers already use instead of asking them to switch.
Features
- Bake sheet — exactly how many of each flavor to bake per pickup day, counting paid orders only. Flavors that share a base recipe are grouped and rounded up to whole batches (extras shown so you can sell them), with a per-flavor ingredient breakdown and check-offs for baked flavors and packed orders.
- Insights — weekly sales, cost, and profit trends, best sellers, orders by pickup place, and heads-ups when sales or margin swing.
- Cost & profit per order — a reusable ingredient + recipe library computes cost per item; profit uses what actually reached your Venmo (after discounts and fees), not the list price.
- Order tracking — every order in one list, with payment status. Orders from your website land automatically through a webhook; DM/text orders are added in seconds (paste the message and Claude structures it); existing sheets can be imported as CSV.
- Production plan — the week as a timeline with check-offs and calendar reminders.
- Shopping list — one aggregated list for the week in store units ("buy 2 × quart heavy cream"), with estimated cost and a restock forecast.
- Pickup sheet (Deliver) — per pickup slot: who's collecting what, paid status, tap-to-check-off.
- Customer messaging — bilingual (EN/中文) templates and Claude-drafted messages you copy into WeChat/SMS. The app never sends on your behalf.
AI features are optional. Without an Anthropic API key everything works except the three Claude helpers (paste-an-order, message drafting, recipe import). Claude handles messy human input (a WeChat DM, a free-text recipe) and drafts messages; all money and quantity math is plain code, and every AI result is a draft you confirm.
How Preheat Is Different
- Built with a real bakery, on real data. Tart Theory's owner brought her actual order and cost sheets, demoed her weekly routine, tested every version, and designed the UI herself. Every feature traces back to a problem she has every week.
- Onboarding that meets bakers where they already sell. Most tools ask bakers to move their sales to a new platform. Preheat starts with "Where do people find and order from you?" and lets bakers list every channel they use. Today, a Google Sheet link shows the sheet inside Preheat and leads straight to CSV import, and the bakery's existing website order format is accepted as-is. One-click connectors for Squarespace, Square, Instagram and WeChat are designed and next.
- From orders to the oven. Costing tools know recipes; order tools know orders. Preheat connects the two, so the week's paid orders turn directly into what to bake, what to buy, and what you actually earned after discounts and Venmo fees.
- AI where it helps, code where it must be right. Claude reads messy input (a WeChat DM in Chinese, a free-text recipe) and drafts bilingual messages. Every number is plain code, and the baker confirms every AI result.
- Free to run. Open source and runs on free tiers, with no per-order fees or monthly subscription.
| Preorder storefronts (e.g. Hotplate) | Costing tools (e.g. Craftybase, CakeBoss) | Preheat |
|---|
| Keep your existing website and sheet | ❌ Move sales to their storefront | ⚠️ Re-enter orders | ✅ Accepts your site's orders; syncs your sheet |
| Orders → what to bake and buy | ⚠️ Order lists | ⚠️ Separate from orders | ✅ Batches, grams and a shopping list |
| Real profit per order | ❌ | ⚠️ Cost per item | ✅ After discounts and Venmo fees |
| Cost | Per-order fee | Monthly subscription | Free on free tiers |
Current Status
- Working: Bake sheet, Insights, costing, shopping list, plan, orders, Deliver, website order webhook, CSV import, Google Sheet view and write-back, Claude helpers.
- Concept: onboarding connectors (Squarespace, Square Online, Instagram, Google Form, …) show the planned flow; the connections are not built yet. The Insights "Conversion & sources" panel uses sample data until order sources are tracked.
- Next: pre-filling your menu from your website link, and matching Venmo payments to orders automatically (today you tap Mark paid).
Tech Stack
Next.js (App Router, TypeScript) · Supabase (Postgres + auth) · Anthropic Claude API · Vercel · Tailwind CSS
Who This Is For
Home bakers who want to stop losing money to untracked costs and stop losing orders to scrolled-away DMs. You don't need to be a developer: if you can follow steps and copy-paste values, you can deploy your own private copy in about 15 minutes.
Option 1: Try the Demo
Open home-bakery-lovat.vercel.app/welcome (deployed from this repo with Tart Theory's sample data). The demo resets periodically — don't enter real customer information.
Option 2: Deploy Your Own Copy (Recommended)
Your own copy keeps your data private, and you control the costs.
About 30 minutes the first time. You'll need free accounts on: GitHub · Vercel · Supabase · Anthropic Console (only for the AI features)
Step 1: Create your database
- Log in to Supabase and click New project. Pick a name, set a database password, choose the region closest to you, and wait ~2 minutes for setup.
- Open the SQL Editor, paste the contents of
supabase/schema.sql, and click Run. This creates the tables.
- (Optional but recommended for a first look) Run
supabase/seed.sql the same way — it loads a sample bakery (Tart Theory's menu and a week of test orders) so the app isn't empty. You can clear it later; to start blank instead, skip this step.
- Go to Project Settings → API and copy three values for the environment table below: Project URL, anon public key, and service_role key (keep that one secret, like a password). If you don't see
anon and service_role, look under the Legacy API keys tab.
- Go to Authentication → Sign In / Providers → Email and turn off "Confirm email". Otherwise the confirmation link points to the wrong address and you can't log in.
Step 2: Get a Claude API key (optional)
Needed only for the AI features — everything else works without it.
- In the Anthropic Console, go to API Keys and create one.
- Set a monthly spending limit under Billing so you're never surprised.
Step 3: Deploy to Vercel
Click the button. It copies this project to your GitHub account and sets up hosting:
When Vercel asks for environment variables, paste the three Supabase values from this table. To turn on the AI features later, add ANTHROPIC_API_KEY in Vercel → Settings → Environment Variables, then Redeploy.
Environment Variables
| Variable | Where to find it | Required | Secret? |
|---|
NEXT_PUBLIC_SUPABASE_URL | Supabase → Project Settings → API → Project URL | Yes | No |
NEXT_PUBLIC_SUPABASE_ANON_KEY | Supabase → Project Settings → API → anon public | Yes | No |
SUPABASE_SERVICE_ROLE_KEY | Supabase → Project Settings → API → service_role | Yes | Yes |
ANTHROPIC_API_KEY | Anthropic Console → API Keys | AI features only | Yes |
INTAKE_TOKEN | You invent it (any random string) | Optional | Yes |
Never share your secret keys or commit them to GitHub. Anyone with the service_role or Anthropic key can access your data or run up charges.
Step 4: Create your baker account
- Open your new site and click Sign up (or "First time? Create the baker account" on the login page).
- Enter your bakery name, email, and password — that's your account.
- Then, in Supabase → Authentication → Sign In / Providers, turn off "Allow new users to sign up" so nobody else can register on your deployment.
Step 5: Connect your website's order form (optional)
If your website already takes orders (e.g. a form that posts to a Google Apps Script/Sheet), point it at your app instead — Settings → Connect your website shows your intake URL:
POST https://<your-app>.vercel.app/api/intake
Content-Type: text/plain
{"orderNum":"TT-1234","items":"2x Corn Egg Tart ($6);\n1x Classic Portuguese Egg Tart ($4)",
"total":"$16","pickupDay":"Fri 10/30","pickupTime":"3:30–4:00 PM",
"pickupPlace":"U District","lang":"zh","weekLabel":"Oct 30-31"}
- Same-order retries are deduplicated by
orderNum.
- Live capacity for a week:
GET /api/intake?action=remaining&week=Oct%2030-31 → {"ok":true,"orderedQty":42}.
- If you set
INTAKE_TOKEN (add it in Vercel and redeploy), append ?token=<your token> to the URL.
- Item names are matched against your menu, including alternate spellings (add aliases in the database's
product_aliases table).
Step 6: Keep your Google Sheet in sync (optional)
If you still use a Google Sheet (or want it as a backup), the app can mirror its changes back into the sheet: new orders you add in the app are appended to the right weekly tab, and marking an order paid updates the Paid / Amount Received cells. Setup is five minutes:
- Open your orders Google Sheet → Extensions → Apps Script.
- Paste the contents of
scripts/sheet-writeback.gs, replacing any starter code.
- Change the
SECRET line to a random string you invent.
- Click Deploy → New deployment → type Web app → Execute as: Me · Who has access: Anyone → Deploy, and approve the one-time authorization prompt.
- Copy the web app URL (ends in
/exec) and paste it plus your secret into the app's Settings → Sheet write-back.
Notes: the script runs with your Google permissions — the app never sees your Google account; it only calls the URL with the secret. Orders arriving via the webhook or CSV import are not mirrored back (they came from the sheet's side already — mirroring them would create duplicates). To see the sheet inside the app, also fill Settings → Google Sheet embed URL (File → Share → Publish to web → Embed).
Your bakery app is ready to use.
Option 3: Run Locally (For Developers)
Requires Node.js 20+.
git clone https://github.com/honeychawla/home-bakery.git
cd home-bakery
npm install
cp .env.example .env.local # then fill in the values from the table above
npm run dev
Open http://localhost:3000.
Project Structure
├── app/
│ ├── (app)/ # Baker-facing pages: home, products (Menu), insights, plan,
│ │ # shopping, orders, bake, schedule (Deliver), settings
│ ├── api/intake/ # Order webhook (website → app)
│ ├── api/import/ # CSV import
│ ├── api/ai/ # Claude helpers: parse-order, draft-message, parse-recipe
│ ├── welcome/ # Marketing page
│ ├── signup/ # Onboarding (account + order sources)
│ └── login/
├── components/ # Client components (forms, check-offs, composer, insights)
├── lib/ # Domain logic: costing, bake sheet, shopping list, restock, insights
├── scripts/
│ └── sheet-writeback.gs # Apps Script: mirrors app changes into your Google Sheet
├── supabase/
│ ├── schema.sql # Database schema — run once in the SQL Editor
│ └── seed.sql # Sample bakery data (optional, anonymized)
└── .env.example # Environment variable template
How the weekly flow works
- Orders arrive from your website (webhook) or you add DM orders by hand/with Claude in Orders.
- Customers pay you on Venmo; you tap Mark paid and enter the amount received (fees tracked automatically). Only paid orders count toward baking and shopping.
- Before shopping day: open Plan for the week's timeline and Shop for the aggregated grocery list.
- On baking day: open Bake for per-day batch counts, and check flavors and orders off as you go.
- On pickup day: open Deliver, check orders off as they're collected, copy reminder messages per customer.
- Home shows this week at a glance; Insights shows how the business is trending week to week.
Troubleshooting
I signed up but can't log in / the confirmation link is broken. Turn off "Confirm email" in Supabase (Step 1.5) and sign up again. If the account already exists, delete it in Authentication → Users first.
The site loads but shows no data. Check NEXT_PUBLIC_SUPABASE_URL and NEXT_PUBLIC_SUPABASE_ANON_KEY in Vercel, then redeploy.
AI buttons show "not configured". Add ANTHROPIC_API_KEY and redeploy — the rest of the app works without it.
My Supabase project is paused. Free projects pause after inactivity. Open the Supabase dashboard and click Restore.
Webhook orders aren't appearing. Test it yourself: curl -X POST <your-app>/api/intake -d '{"orderNum":"TEST-1","items":"1x <exact menu item name> ($6)"}'. If unmatchedItems is non-zero, the item name doesn't match your menu — add an alias.
Imported orders show 0 on Bake and Shop. The CSV's item text must look like 2x Matcha Tart ($6); 1x Original Egg Tart ($5) and match your menu's names. Fix names in Menu or add aliases.
Changes to environment variables don't take effect. In Vercel, redeploy the latest deployment.
Costs
- Vercel & Supabase: free tiers comfortably cover a small bakery.
- Claude API: pay-as-you-go; each AI action costs a fraction of a cent. Set a spending limit.
Contributing
Contributions welcome — please open an issue first.
License
MIT
Built with Claude at the Seattle AI Week Claude Impact Lab Hackathon, piloted with Tart Theory.
