Omnichannel AI Support Workflow: Ollama + Postgres pgvector

Turn customer messages into governed AI replies—across chat, email, and social—using Ollama + Postgres (pgvector)

This n8n omnichannel AI Support Workflow ingests inbound customer messages from multiple channels via webhook, stores and searches conversation history in Postgres with pgvector, and uses Ollama to draft a cited, guardrailed response or trigger a human handoff when needed.

What this workflow does

• Receives messages from three webhook endpoints (chat, email, or social) in a normalized payload format.
• Validates security and request size by checking a shared secret header and rejecting invalid or oversized requests with an error response.
• Upserts conversation data in Postgres, stores the inbound message, fetches recent history, and ignores duplicate provider event IDs.
• Retrieves relevant knowledge: generates an embedding for the customer message using Ollama and runs pgvector similarity search in Postgres to pull the top published knowledge snippets.
• Classifies intent + drafts a JSON reply in Ollama using the message, recent conversation history, and retrieved knowledge (including citations).
• Applies guardrails to force human escalation for sensitive requests, missing approved citations, low confidence, or when a conversation is already waiting for human support.
• Routes the outcome: creates a handoff record in Postgres and notifies a human-queue webhook for escalation; otherwise records the outbound reply, sends it via a reply-webhook channel adapter, and returns a 200 acknowledgment.

Use cases

• Helpdesk automation for SaaS support teams who want consistent AI assistance across chat, email, and social.
• AI support with citation requirements—auto-escalate when answers can’t be backed by approved knowledge.
• Omnichannel customer experience where conversation context and deduping are handled in Postgres with vector search.

Technical details

• n8n nodes/tools: webhook, respond to webhook, if, code, postgres, and HTTP Request.
• AI + retrieval: Ollama for embeddings and chat completion; Postgres pgvector for similarity search.
• Core tables: creates required tables such as cs_conversations (and related conversation/message storage).