n8n Workflow: All-Source AI Knowledge Agent (Claude, RAG, Drive)

Turn chat into answers from everything you know—Claude + RAG + Google Drive

This n8n workflow, All-Source AI Knowledge Agent (Claude, RAG, Drive), automatically takes incoming chat messages and responds using an orchestrated Anthropic Claude agent that can recall prior context (via Postgres), retrieve relevant knowledge (RAG), query structured data, and search documents in Google Drive.

What this workflow does

• Receives messages: A “When chat message received” trigger captures each user message and sends it to the Knowledge Agent.
• Orchestrates the AI response: The Knowledge Agent uses the Anthropic Chat Model with claude-sonnet-4-20250514 to generate the final reply.
• Remembers conversations: Postgres Chat Memory stores and recalls conversation context across sessions.
• Uses tool-assisted reasoning: A “Think” tool is included to force internal reasoning before actions are taken.
• Finds relevant knowledge (RAG): It searches a General knowledge vector store using OpenAI embeddings (1536-dimensional) and Cohere reranking to improve retrieval quality.
• Answers with data: It provides a structured data Postgres tool to run queries against relational tables.
• Searches your docs: It includes search about any doc in google drive, then calls the Read File From GDrive sub-workflow to fetch and process files for use in responses.
• Optionally pulls current info: It offers Message a model in Perplexity to access up-to-date web information.

Use cases

• Customer support agents answering questions from internal knowledge bases and policy docs stored in Google Drive.
• Operations teams using Postgres-backed structured data (FAQs, status tables, KPIs) combined with RAG retrieval.
• SaaS operators building an n8n knowledge copilot for research + documentation workflows.

Technical details

• n8n nodes/tools: set, switch, sticky note, Google Drive, Postgres tool, manual trigger.
• Integrations: Google Drive file search + Read File From GDrive, Postgres Chat Memory, vector search with OpenAI embeddings + Cohere reranking, and Anthropic Chat Model (claude-sonnet-4-20250514).
• Agent orchestration: “When chat message received” → Knowledge Agent with reasoning via Think.