RAG chat assistant with Claude, Supabase vector store, and Postgres memory

📺 Full walkthrough video: https://youtu.be/Z_l_T22px3U

Who it's for

This workflow is for developers and AI builders who want to deploy a context-aware chat assistant powered by a private knowledge base. It suits teams that need persistent conversation memory and retrieval-augmented generation (RAG) on their own infrastructure.

How it works

1. The Chat Trigger listens for incoming user messages and forwards them to the AI Agent.
2. The AI Agent (Claude Sonnet 4) orchestrates the response generation.
3. Postgres Chat Memory stores and retrieves the conversation history, enabling multi-turn dialogue.
4. The Supabase Vector Store is queried as a tool to retrieve relevant documents from the knowledge base.
5. OpenAI Embeddings convert the user query into vectors used for the Supabase similarity search.
6. Claude synthesizes context from memory and retrieved documents to produce a final response.

How to set up

• [ ] Add your Anthropic API key to the Claude Sonnet 4 Model node
• [ ] Add your OpenAI API key to the Generate OpenAI Embeddings node
• [ ] Configure your Supabase project URL, API key, and target table name
• [ ] Configure the Postgres Chat Memory node with your PostgreSQL database credentials
• [ ] Optionally set authentication on the Chat Trigger node

Requirements

• Anthropic account (Claude Sonnet 4)
• OpenAI account (text-embedding model)
• Supabase project with a vector-enabled table
• PostgreSQL database for session memory

How to customize

• Swap Claude for OpenAI GPT-4 or another LLM supported by n8n
• Replace Supabase Vector Store with Pinecone, Qdrant, or another vector database
• Use a unique session key per user to support isolated multi-user conversations