Skip to main content

AI-prompt

You are a Senior Full-Stack Engineer and System Architect. Your task is to scaffold the initial codebase for "OxaWay," a B2B SaaS AI Concierge platform for the hospitality industry.

We are adhering to a strict System Architecture Design Document (SADD). Please read the following context and requirements carefully before generating any code.

1. Project Structure (Monorepo)​

We will use TurboRepo for monorepo management. The structure must be:

  • apps/dashboard: Next.js 14 (App Router) for the Tenant Admin Panel (Agents/Admins).
  • apps/widget: Vite + React for the embeddable chat widget script (Client-facing).
  • packages/database: Shared Prisma schema and Supabase client configuration.
  • packages/trigger: Trigger.dev (v3) workflow definitions for backend business logic.
  • packages/ui: Shared UI components (Shadcn/UI).

2. Core Technology Stack​

  • Language: TypeScript (Strict mode).
  • Database: Supabase (PostgreSQL) with pgvector enabled.
  • ORM: Prisma.
  • Orchestration: Trigger.dev (v3) Cloud.
  • Chat Engine: Typebot (Self-hosted via iframe).
  • AI Logic: Flowise (Self-hosted API).
  • State Management: Supabase Realtime (for session status sync).

3. The "Wrapper Pattern" Requirement (CRITICAL)​

The apps/widget application is NOT a standard chat UI. It is a Wrapper Component that orchestrates the user experience. It must implement the following logic:

  1. Default State: Render the Typebot Standard Embed (via Iframe) for AI automation.
  2. Context Injection: It must parse the URL parameters (e.g., ?property_id=123) and pass them as "Hidden Variables" to the Typebot instance.
  3. Real-time Listener: On mount, subscribe to Supabase Realtime for the current session_id.
  4. Handover Logic:
    • Listen for changes to the status column in the Session table.
    • IF status changes to 'human_active', unmount the Typebot Iframe and mount a custom React Chat Interface (using Supabase for messages).
    • IF status changes back to 'resolved', swap back to the Typebot Iframe to trigger a feedback flow.

4. Data Model (Multi-Tenancy)​

In packages/database/prisma/schema.prisma, define the following 3-tier hierarchy:

  • Tenant: (e.g., "Emma Villas") -> Includes stripe_id, config (JSON).
  • Property: (e.g., "Villa Corbaia") -> Belongs to Tenant. Includes property_id (external ID), name, knowledge_base_id (Flowise ID).
  • Session: (The Chat) -> Belongs to Property. Includes status (enum: ai_active, human_active, resolved), guest_info (JSON), platform (enum: web, whatsapp).

5. Implementation Steps​

Please generate the shell commands to initialize the TurboRepo and then provide the code for the following key files:

Step A: Database Schema

  • Write the schema.prisma file implementing the hierarchy above.

Step B: The Widget Wrapper (apps/widget/src/App.tsx)

  • Write the React component that handles the conditional rendering between the Typebot embed and the custom Human Chat UI based on Supabase Realtime events.
  • Note: Do not fully implement the "Human Chat UI" yet, just a placeholder component. Focus on the switching logic.

Step C: Trigger.dev Setup (packages/trigger/src/trigger.ts)

  • Initialize the Trigger.dev client.
  • Create a placeholder task onboard-tenant that simulates creating a Stripe customer and inserting a Tenant into Supabase.

Constraints:

  • Do not hardcode secrets. Use process.env.
  • Ensure strict logical isolation (Row Level Security ideas can be commented, but enforce tenant_id in Prisma queries).
  • Use functional React components with Hooks.

Start by proposing the folder structure and the schema.prisma.