n8n RAG Chat Agent: Multi-Knowledge Base with Supabase & Ollama

Build a multi-knowledge-base RAG chat agent in n8n—without writing routing code

This n8n RAG Chat Agent automatically routes user questions across multiple Supabase pgvector knowledge bases using only tool descriptions. It delivers semantic answers with source attribution, keeps persistent conversation memory in Postgres, and runs locally with Ollama.

What this workflow does

• Receives a message from the built-in n8n Chat interface via the LangChain Chat Trigger.
• Uses an agent powered by n8nn8n-nodes-langchainagent and guided by a system prompt to read each Supabase tool’s name and description.
• Automatically decides which knowledge base(s) to query—no routing code needed.
• Performs semantic retrieval against Supabase pgvector tables (e.g., Primary, Reference, and an optional Additional knowledge base).
• Generates a concise response grounded in retrieved results and appends a source line naming the knowledge base(s) used.
• Saves and reuses conversation history using Postgres Chat Memory, so follow-up questions retain context.
• If the workflow errors, an Error Trigger captures the failure and formats an execution summary for notification.

Use cases

• SaaS support assistants that search both product docs and reference policies in Supabase.
• Internal engineering Q&A spanning multiple knowledge sets (primary docs, reference standards, optional additional corpora).
• Operator-friendly RAG chat for teams who want local inference using Ollama instead of hosted APIs.

Technical details

• n8n nodes: n8nn8n-nodes-langchainagent, n8nn8n-nodes-langchainchat trigger, n8nn8n-nodes-langchainlm chat ollama, plus Error Trigger, Set, and Sticky Note.
• Supabase setup: enable pgvector and create vector tables plus match_* SQL functions (e.g., match_documents, match_reference_documents, match_additional_documents).
• Extensible design: add unlimited knowledge bases by defining additional tools and pgvector tables.