AI-powered Zendesk support responses with RAG, OpenAI, and Supabase knowledge base

⚡ How it works

This workflow automates first responses to new Zendesk tickets with the help of AI and your internal knowledge base.

Webhook trigger fires whenever a new ticket is created in Zendesk.

Ticket details (subject, description, requester info) are extracted.

Knowledge base retrieval – the workflow searches a Supabase vector store (with OpenAI embeddings) for the most relevant KB articles.

AI assistant (RAG agent) drafts a professional reply using the retrieved KB and conversation memory stored in Postgres.

Decision logic:

If no relevant KB info is found (or if it’s a sensitive query like KYC, refunds, or account deletion), the workflow sends a fallback response and tags the ticket for human review.

Otherwise, it posts the AI-generated reply and tags the ticket with ai_reply.

Logging & context memory ensure future ticket updates are aware of past interactions.

🔧 Set up steps

This workflow takes about 15–30 minutes to set up.

Connect credentials for Zendesk, OpenAI, Supabase, and Postgres.

Prepare your knowledge base: store support content in Supabase (documents table) and embed it using the provided Embeddings node.

Set up Postgres memory table (zendesk_ticket_histories) to store conversation history.

Update your Zendesk domain in the HTTP Request nodes (<YOUR_ZENDESK_DOMAIN>).

Deploy the webhook URL in Zendesk triggers so new tickets flow into n8n.

Test by creating a sample ticket and verifying:

AI replies appear in Zendesk

Correct tags (ai_reply or human_requested) are applied

Logs are written to Postgres