Kazán Kecskemét — Árajánló chat widget

A Hungarian gas‑boiler quoting chat widget. The customer answers a short set of questions (clicking suggested options or typing freely), gives their contact details, then immediately sees an itemised estimate. The company owner receives the same quote + the customer's details by e‑mail.

---

How it works (architecture)

public/widget.js   ──POST──►  api/faq-agent.js  ──►  OpenAI (model via .env)  = conversation only
   (chat UI)                       │
                                   ├──►  PRICES table  = deterministic price calc
                                   └──►  Resend         = e-mail to owner

The AI never does arithmetic. It only runs the Hungarian conversation and, once every answer is collected, emits a hidden JSON block of the customer's choices (not prices). The backend looks each choice up in the fixed PRICES table, sums it, and builds the quote. This is why the total can never be miscalculated by the model.

---

How the price is calculated

All prices are in HUF and are shown as‑is (no VAT note, no base/call‑out fee). The total is simply the sum of the applicable items below. Source of the numbers: the company's price sheet (milan.xlsx). Edit them in one place: the PRICES object at the top of api/faq-agent.js.

Step	Question	Options → amount added
1	Csere vagy új beépítés?	replacement / new install — controls step 2 & the demolition line
2	Jelenlegi kazán (only if replacement)	nyílt égésterű +60 000 · kondenzációs +0 · turbós +60 000
3	Új kazán típusa	kombi 24 kW +450 000 · tárolós 46 L +900 000 · külső 125 L +900 000
4	Kémény / égéstermék‑elvezetés	tetőn ki +380 000 · tégla kéménybe +600 000 · társasházi gyűjtőkémény +600 000
5	Életvédelmi (Fi) relé	van +50 000 · nincs +100 000
—	Always added (not asked)	vizes rendszerre kötés +300 000 · gyári üzembe helyezés +50 000
—	Added on replacement only	régi kazán/kémény bontása +90 000 (csak csere esetén)

Total = sum of the selected rows + the always‑added standard costs.

Decisions baked into the logic (confirmed by the company)

• Prices are GROSS (ÁFA included) and include the boiler appliance + full installation — the displayed total is what the customer pays (APPLIANCE_INCLUDED = true).
• Standard costs (wet‑system +300 000, commissioning +50 000) are always added and not asked. Demolition +90 000 is added the same way but only on a replacement (no old boiler to remove on a new install).
• Current boiler only counts on a replacement (a new install → nincs, +0).
• No gázterv line — the company confirmed it isn't needed.
• Only 24 kW appliances; exact brand/model is decided at the site survey.
• Contact details collected in one form at the end (name, phone, postal code required; e‑mail optional).
• Any "Nem tudom" answer is priced at its cheapest variant, so the total is shown as "Induló ár: X Ft-tól" instead of an estimate.
• Chip clicks are answered without the AI (scripted next question); only typed messages go to the model.

Quote delivery

• The itemised estimate is shown in the chat as soon as all answers are in, split into easy‑to‑read bubbles (price → "just an estimate" note → a recap of everything the customer answered).
• The owner always receives it by e‑mail (with the client's details).
• The customer is offered a button to have the quote e‑mailed to them too.
• Completion is decided by the backend (isQuoteReady) from a running hidden state block the model maintains — so the quote always appears even if the model phrasing varies. The model never computes the price.

---

Setup & deploy

1. Keys — copy your secrets into .env (already git‑ignored):
   • OPENAI_API_KEY — from https://platform.openai.com/api-keys
   • OPENAI_MODEL — copy the exact model ID from your OpenAI dashboard. The task is tiny, so the cheapest tier is plenty (e.g. gpt-5.4-nano, or gpt-5.4-mini). The model never affects price accuracy — that's computed in the backend.
   • RESEND_API_KEY — free at https://resend.com
   • LEAD_EMAIL_TO — where quotes are sent (default [email protected])
   • LEAD_EMAIL_FROM — leave as the [email protected] test sender to start; later verify your own domain in Resend and change it.
2. Run locally: node server.js → http://localhost:8888
3. Deploy (Vercel): push the repo; set the same env vars in the Vercel dashboard. api/faq-agent.js is the serverless endpoint, public/ is static.
4. Embed on a site:

   <script>
     window.KAZAN_CONFIG = {
       apiUrl: "https://YOUR-APP.vercel.app/api/faq-agent",
       assetsUrl: "https://YOUR-APP.vercel.app"
     };
   </script>
   <script src="https://YOUR-APP.vercel.app/widget.js"></script>
   

---

Leads → Google Sheet

Every completed lead is also appended as a row to a Google Sheet (timestamp, name, phone, e-mail, postal code, budget, timeline, the four choices, and the estimated total). This is optional — leave SHEETS_WEBHOOK_URL empty and the quote/e-mail flow still works exactly as before.

Setup (one-time, ~3 minutes, no Google Cloud project needed):

1. Create a Google Sheet. In the first row, add these headers (this exact order — it matches the row array in sendLeadToSheet):

   Időbélyeg | Név | Telefon | E-mail | Irányítószám | Tervezett keret | Tervezett kivitelezés | Jelenlegi kazán | Új kazán | Kémény | Életvédelmi relé | Becsült végösszeg (Ft)

2. In that sheet: Extensions → Apps Script. Replace the contents with:

   function doPost(e) {
     var sheet = SpreadsheetApp.getActiveSpreadsheet().getActiveSheet();
     var data = JSON.parse(e.postData.contents);
     sheet.appendRow(data.row);
     return ContentService
       .createTextOutput(JSON.stringify({ ok: true }))
       .setMimeType(ContentService.MimeType.JSON);
   }
   

3. Deploy → New deployment → Web app. Set Execute as = Me, and Who has access = Anyone. Deploy, authorise, and copy the Web app URL (ends in /exec).

4. Paste that URL into SHEETS_WEBHOOK_URL in .env (and into the same env var in the Vercel dashboard). Done — new leads now land in the sheet.

The sheet only fills in once the customer completes the whole flow (same point the owner e-mail is sent). It runs in parallel with the e-mail, so neither blocks the other.

---

Customising

• Prices: edit the PRICES object in api/faq-agent.js.
• Questions / wording: edit SYSTEM_PROMPT in api/faq-agent.js.
• Colours / branding: public/style.css (navy #111827, gold #FBBF24, blue #025888). Replace public/logo.svg with the company's real logo.
• Phone number: PHONE constant in public/widget.js.