Adaptive RAG strategy with query classification & retrieval (Gemini & Qdrant)
Needs changesNeeds changes. This template uses 1 thing(s) that n8n 3.0 removes: AI Agent v1 (old agent modes). 1 breaking item(s) found. Static check, last run 2026-10-19.
What to replace before upgrading
| Removed in 3.0 | Replace with |
|---|---|
| AI Agent v1 (old agent modes) Query Classification (v1.8), Factual Strategy - Focus on Precision (v1.7), Analytical Strategy - Comprehensive Coverage (v1.7), Opinion Strategy - Diverse Perspectives (v1.7), Contextual Strategy - User Context Integration (v1.7), Answer (v1.8) | Latest AI Agent node (Tools Agent). For SQL Agent, use a Postgres or MySQL tool with a recent AI Agent. |
Nodes used (11)
- AI Agent
@n8n/n8n-nodes-langchain.agent - Switch
n8n-nodes-base.switch - Chat Trigger
@n8n/n8n-nodes-langchain.chatTrigger - Edit Fields (Set)
n8n-nodes-base.set - Google Gemini Chat Model
@n8n/n8n-nodes-langchain.lmChatGoogleGemini - Simple Memory
@n8n/n8n-nodes-langchain.memoryBufferWindow - Embeddings Google Gemini
@n8n/n8n-nodes-langchain.embeddingsGoogleGemini - Summarize
n8n-nodes-base.summarize - Qdrant Vector Store
@n8n/n8n-nodes-langchain.vectorStoreQdrant - Respond to Webhook
n8n-nodes-base.respondToWebhook - Execute Workflow Trigger
n8n-nodes-base.executeWorkflowTrigger
What it does
This n8n workflow implements a version of the Adaptive Retrieval-Augmented Generation (RAG) framework. It recognizes that the best way to retrieve information often depends on the type of question asked. Instead of a one-size-fits-all approach, this workflow adapts its strategy based on the user's query intent. 🌟 How it Works 1. Receive Query: Takes a user query as input (along with context like a chat session ID and Vector Store collection ID if used as sub-workflow). 2. Classify Query: First, the workflow classifies the query into a predefined category. This template uses four examples: Factual: For specific facts. Analytical: For deeper explanations or comparisons. Opinion: For subjective viewpoints. Contextual: For questions relying on specific background. 3. Select & Adapt Strategy: Based on the classification, it selects a corresponding strategy to prepare for information retrieval. The example strategies aim to: Factual: Refine the query for precision. Analytical: Break the query into sub-questions for broad coverage. Opinion: Identify different viewpoints to look for. Contextual: Incorporate implied or user-specific context. 4. Retrieve Info: Uses the output of the selected strategy to search the specified knowledge base (Qdrant vector store - change as needed) for relevant documents. 5. Generate Response: Constructs a response using the retrieved documents, guided by a prompt tailored to the original query type. By adapting the retrieval strategy, this workflow aims to provide more relevant results tailored to the user's intent. ⚙️ Usage & Flexibility Sub-Workflow: Designed to be called from other n8n workflows, passing userquery, chatmemorykey, and vectorstoreid as inputs. Chat Testing: Can also be triggered directly via the n8n Chat interface for easy testing and interaction. Customizable Framework: The query categories (Factual, Analytical, etc.) and the associated retrieval strategies are examples. You can modify or replace them entirely to fit your specific domain or requirements. 🛠️ Requirements Credentials: You will need API credentials configured in your n8n instance for: Google Gemini (AI Models) Qdrant (Vector Store)