n8n Multi-Channel Support Bot: OpenAI RAG + Supabase

Answer support questions across Email, Live Chat, WhatsApp, Slack, and Discord using your own knowledge base—then escalate when needed

This n8n Multi-Channel Support Bot uses OpenAI RAG with a Supabase vector store to deliver grounded answers from your documentation. It categorizes and logs every conversation, reports weekly, and routes unclear or risky interactions to a human via Zendesk.

What this workflow does

• Normalizes incoming messages from Email, Live Chat, WhatsApp, Slack, and Discord into a single schema, with an explicit branch per channel.
• Runs five “guards” before any model call to avoid wasted tokens and bad experiences: auto-replies and bounces, empty bodies, duplicate deliveries, oversized messages, and explicit request-for-human detection (in six languages).
• Uses your documentation for answers by retrieving relevant context from a Supabase vector store connected as a tool.
• Scores confidence on grounding—based on whether retrieval found relevant results (not on answer length).
• Short-circuits greetings and thanks with a lightweight test so routine politeness doesn’t trigger support escalation.
• Replies back to the originating channel when confidence is at/above the threshold and sentiment is not negative (including threaded replies in Slack).
• Escalates when necessary: ungrounded answers, negative sentiment, or explicit human handover requests create a Zendesk ticket.
• Logs and reports: every conversation is categorized, stored, and summarized weekly.

Use cases

• SaaS support teams reducing inbox load by answering from internal docs—while keeping Zendesk as the safety net.
• Automation engineers deploying a single workflow that handles multiple channels with consistent logic.
• Teams that want grounded, retrieval-backed responses instead of long, unsupported “chatty” answers.

Technical details

• Integrations/channels: Email, Live Chat, WhatsApp, Slack, Discord
• RAG stack: OpenAI RAG + Supabase vector store
• Memory: Postgres-backed chat memory to persist history across restarts
• Zendesk escalation for negative sentiment, ungrounded answers, and human requests
• n8n nodes used: if, set, code, no op, slack, switch