Build a powerful Retrieval-Augmented Generation (RAG) AI agent that intelligently answers questions by retrieving and processing your own data sources. This project leverages no-code workflow builder n8n and Supabase vector database for easy and scalable implementation.
This project automates the process of vectorizing documents, embedding them into a Supabase vector database, and setting up an AI agent that can chat and respond with context-aware answers by querying the vector store.
- Automated document ingestion: Break down PDFs and other documents into text chunks via n8n workflows.
- Embedding and vector storage: Use OpenAI or Google Gemini embeddings stored in Supabase for ultra-fast retrieval.
- AI agent Q&A: Natural language conversation powered by GPT models, augmented with vector database search results.
- Persistent chat memory: Keep user session history for dynamic and meaningful conversations.
- Documents (e.g., PDFs) are loaded from Google Drive and chunked into smaller parts.
- Each chunk is embedded into a high-dimensional vector using an LLM API (OpenAI or Gemini).
- Chunks are indexed and stored in Supabase’s vector database.
- A user query triggers the RAG Agent, which uses vector search to find relevant chunks.
- The query plus retrieved chunks are sent to the LLM to generate an informed response.
- Conversation context is maintained via PostgreSQL chat memory for multi-turn dialogues.
- Fully no-code RAG agent workflow in n8n
- Integration with Google Drive, Supabase, and OpenAI/Gemini
- Persistent memory for conversational context
- Supports PDF and document ingestion at scale
- Real-time client-agent chat with vector enhanced understanding
- Clone this repo and import the n8n workflow JSON.
- Configure your Supabase project, OpenAI/Gemini API keys, Google Drive access, and PostgreSQL memory.
- Use the UI to run and test your RAG AI agent interactions.
- Extend with additional data sources and agent triggers as needed.
Below are key screenshots showing major stages of the RAG AI Agent lifecycle as automated in n8n:
-
PDF Ingestion
documents are loaded from Google Drive or user uploads as part of the automation flow.
-
RAG Agent Process
The central view of the RAG agent orchestration—showing tools, nodes, and integration points within n8n.
Example of an agent node connecting to embedding models and vector search.
-
Vectorization Pipeline
A schematic (or n8n node screenshot) illustrating the flow: document chunking → embedding → vector DB upload.
-
Automation Workflow
High-level automation showing document retrieval, agent triggers, and query/response path.
This automation leverages n8n to connect document data sources, process content into searchable vectors, and trigger a retrieval-augmented AI agent to answer questions using your private knowledge base.
- n8n (workflow automation)
- Supabase (vector database & Postgres chat memory)
- Google Drive (document source)
- OpenAI / Google Gemini (embeddings and chat)
- PostgreSQL (chat history memory)