Chat Architecture

Two separate chat experiences exist: a static site chat (dumb, no auth) and a portal chat (AI-powered, auth required).

---

1. Static Site Chat (Dumb)

Purpose: SEO-friendly landing page with a simple Q&A widget. No AI, no login.

How it works:

• Business publishes a static site via POST /api/site/publish (Firebase Function)
• template.ts renders index.html + sub-pages as pure HTML/CSS/JS
• QA data is generated at publish time from the business knowledge base and baked into the HTML as <script type="application/json" id="chat-qa-data">
• Client-side JS does fuzzy pattern matching (bigram similarity) against the baked QA data — zero API calls
• Fallback answer links users to the portal for full AI chat

Hosting: GCS bucket humanlike-sites → {uid}/index.html, {uid}/{slug}.htmlURL: https://storage.googleapis.com/humanlike-sites/{uid}/index.htmlAuth: None Firebase SDK: None AI: None — pattern matching only Runtime cost: Zero (static file serving)

Files:

• functions/src/site/template.ts — HTML generation (landing page, sub-pages)
• functions/src/site/generate-qa.ts — QA generation from knowledge base (at publish time)
• functions/src/site/storage.ts — GCS upload
• functions/src/site/index.ts — Express routes: preview, publish, unpublish, import

Chat button in navbar: Links to portal (https://app.humanlike.world/portal/{slug}) for full AI chat.

---

2. Portal Chat (AI-powered)

Purpose: Full conversational AI chat. Requires sign-in. Talks to our API, uses tools (calendar, etc).

How it works:

• User visits https://app.humanlike.world/portal/{slug}
• React app resolves slug → business via GET /api/portal/resolve?slug=X (Cloud Run API)
• User signs in (Google OAuth or email/password via Firebase Auth)
• On sign-in, portal sets visitor custom claim and writes to visitors/{uid}/businesses/{bizId} in Firestore
• ChatGPT component fetches system prompt (GET /api/portal/prompt) and tool definitions (GET /api/portal/tools)
• Messages sent to POST /api/chat (OpenAI Chat Completions with streaming)
• Tool calls executed via POST /api/portal/tools (server-side, same tool registry as phone)
• On unmount, interaction logged via POST /api/portal/interaction → Firestore businesses/{bizId}/interactions
• Chat history persisted in localStorage per business

Hosting: Firebase Hosting (React SPA) Route: /portal/:slug → Portal.tsxAuth: Firebase Auth (Google OAuth + email/password) Firebase SDK: Auth only (for sign-in + ID tokens). Firestore used server-side only. AI: OpenAI Chat Completions API (via our Cloud Run proxy) Runtime cost: OpenAI API usage per message

Files:

• web/src/components/Portal.tsx — Portal page: slug resolution, auth gate, layout shell, interaction logging
• web/src/components/ChatGPT.tsx — Reusable chat component: streaming, tool calls, message display
• web/src/components/TexterHome.tsx — "Your Chats" page: lists businesses the visitor has chatted with
• services/api/serve.js — API endpoints: /api/portal/resolve, /api/portal/prompt, /api/portal/tools, /api/portal/interaction, /api/chat

---

Comparison

Aspect	Static Site Chat	Portal Chat
Auth required	No	Yes (Firebase Auth)
AI	No (pattern matching)	Yes (OpenAI Chat API)
Tools (calendar, etc)	No	Yes
Firebase SDK in client	No	Auth only
Firestore reads at runtime	No	Yes (server-side via API)
SEO crawlable	Yes	No (SPA, auth-gated)
Hosting	GCS (static files)	Firebase Hosting (React)
Runtime cost	Zero	OpenAI API per message
Chat history	None (stateless)	localStorage + Firestore interaction log
Conversation depth	Single Q&A turn	Multi-turn with context

---

Flow: Static Site → Portal

1. Visitor lands on static site (SEO, shared link, etc.)
2. Uses dumb chat for basic questions (hours, location, services)
3. For deeper questions, clicks "Chat" button in navbar → opens portal in new tab
4. Signs in → full AI chat with tool access (booking, lookups, etc.)
5. Interaction logged to business dashboard