Can I vibe code Intercom?

Build a full-stack, single-tenant AI Customer Support and Shared Inbox platform designed for e-commerce stores using Node.js/TypeScript, PostgreSQL (Prisma), Redis, and React (Next.js App Router).

CORE ARCHITECTURE & ENTITIES
1. Database Models:
   - Workspace (id, name, domain, api_key)
   - User/Agent (id, workspace_id, email, role [ADMIN, AGENT], status [ONLINE, OFFLINE])
   - Customer (id, workspace_id, external_id, email, phone, name, custom_attributes [JSONB])
   - Conversation (id, workspace_id, customer_id, assigned_agent_id, status [OPEN, CLOSED, PENDING], channel [CHAT, EMAIL], created_at, updated_at)
   - Message (id, conversation_id, sender_type [CUSTOMER, AGENT, BOT], content, attachments [JSONB], created_at)
   - KnowledgeArticle (id, workspace_id, title, content, vector_embedding [vector(1536)])
   - WorkflowProcedure (id, workspace_id, trigger, action_payload [JSONB], is_active)

REAL-TIME CHAT WIDGET ENGINE
1. Client Widget Snippet:
   - Provide an isolated, lightweight vanillajs shadow-DOM script that embeds on external storefronts.
   - Establish a persistent WebSocket connection to a dedicated server gateway using Socket.io or ws.
   - Maintain conversation session continuity across browser reloads via secure HttpOnly session tokens or localStorage keys.
   - Support rendering rich UI cards inside chat (e.g., Shopify Order Status cards with tracking links, action buttons).

AI SUPPORT RESOLUTION (FIN CLONE)
1. Automated Agent Pipeline:
   - When a new customer message arrives, trigger an async job via BullMQ.
   - Run vector search using pgvector against KnowledgeArticles to retrieve top matching contexts.
   - Fetch customer order context from an external mock e-commerce API using the customer's email.
   - Invoke an LLM (e.g., OpenAI gpt-4o) with structured system instructions, article context, and order status.
   - If confidence is high and action requires tool execution (e.g., "Where is my order?"), invoke function calling to query order status and formulate a response.
   - If confidence is below threshold or customer requests a human, escalate the conversation: change status to OPEN, assign to a available agent (round-robin), and issue a real-time event to the agent inbox.
   - Track resolution status: mark outcome as 'SUCCESS' if user closes or does not reply within 10 minutes.

SHARED AGENT INBOX UI
1. Real-Time Dashboard:
   - Split-pane layout: Left pane for filterable conversation queue (Unassigned, Mine, All), middle pane for active chat thread, right pane for customer profile & order context.
   - Real-time updates via WebSockets when new messages arrive or when another agent claims a ticket.
   - Agent AI Copilot side-panel: One-click "Draft AI Answer" button using context from knowledge base and recent conversation turn.
   - Support private internal notes visible only to agents.

EXPLICITLY OUT OF SCOPE
- Outbound bulk email/SMS marketing campaigns or series builders.
- Voice/Phone calling integrations.
- Native mobile SDKs (React Native/Swift/Kotlin wrappers).
- Complex multibrand data isolation beyond standard workspace multitenancy.

$ each button prefixes agent-specific run instructions · build your own product, never copy proprietary code, trademarks or designs