n8n RAG Chat Agent: Gemini Embeddings + Supabase Vector Store

Build a Gemini-powered RAG chatbot in n8n—powered by Supabase vector search

This n8n RAG Chat Agent workflow ingests a README.md from GitHub, generates Google Gemini embeddings, stores them in a Supabase Vector Store, and then answers questions in an n8n Chat interface using retrieved, grounded context.

What this workflow does

• Ingest content from GitHub: Runs manually to fetch a README.md file from a GitHub raw URL using an HTTP Request node.
• Chunk and enrich documents: Splits the fetched Markdown into document chunks and adds metadata for each chunk.
• Create embeddings with Google Gemini: Uses Gemini for embeddings, then inserts embedded documents into a Supabase vector-enabled table via the Supabase vector store.
• Enable chat with retrieval-augmented answers: Triggers when a chat message is received in n8n Chat, uses a Google Gemini chat model with buffer memory, and a Supabase retrieval tool to pull relevant chunks.
• Grounded responses: Returns answers grounded in the retrieved Supabase documents.

Use cases

• SaaS operator support: Answer internal questions using docs stored in a Supabase vector table.
• n8n workflow onboarding: Build a chat agent that explains how your GitHub-hosted README works.
• Knowledge base Q&A: Retrieve the most relevant document chunks from Supabase for accurate, source-based responses.

Technical details

• Integrations: GitHub (raw README via HTTP Request), Google Gemini (embeddings + chat), Supabase Vector Store.
• Workflow triggers: Manual trigger for ingestion; n8nnodes-langchainchat for chat messages in n8n Chat.
• Key nodes/tools: n8nnodes-langchainagent, n8nnodes-langchainchat trigger, n8nnodes-langchainlm (Google Gemini), plus vector store retrieval.

Setup note: Create a Supabase vector table (e.g., documents) and ensure the vector size matches the embedding model; then add Supabase credentials and Google Gemini (PaLM) API credentials for both embeddings and the chat model.