n8n for AI Agents: LangChain, OpenAI & LLM Orchestration

n8n’s AI Agent node orchestrates large language models from OpenAI, Anthropic, Google, and open‑source providers, chaining prompts, tools, and memory to build autonomous agents. It can retrieve context from vector stores, execute API calls via tools, and maintain conversation memory across sessions—all within a visual workflow. This guide covers implementing RAG, LangChain‑style chains, function calling, and multi‑model agents. [1]

How do you build an LLM chain with the n8n AI Agent node?

Connect an AI Agent node to a chat model sub‑node like OpenAI Chat Model. The agent receives a system prompt and user message, then runs the specified model. Chain this output into another AI Agent node with a follow‑up prompt to emulate LangChain‑style sequential chains. Each node access can be conditioned with an IF branch. [2]

For retrieval‑augmented generation, insert a Vector Store Tool sub‑node between the agent and the model to pull documents from a vector database before generating an answer. Learn how vector stores fit into the pipeline in our n8n node techniques guide.

How do you integrate OpenAI’s GPT-4 into an n8n workflow?

Add an OpenAI Chat Model sub‑node to the AI Agent node, authenticate with your API key, and select the model (e.g., gpt-4-turbo). Set the temperature, max tokens, and system message in the node’s parameters. The agent sends the prompt, and the raw response is returned for further processing by downstream nodes. [3]

Combine with a Switch node to route different prompts based on input, or use a Code node to format multi‑modal messages with images. You can also call OpenAI’s endpoints directly via the HTTP Request node, as shown in the use cases for AI-driven automations.

How do you set up a RAG pipeline with n8n and a vector store?

Load documents via a file or HTTP node, split them with the Recursive Character Text Splitter node, embed them with an Embeddings sub‑node (OpenAI, Cohere, or Hugging Face), and store them in a vector store like Pinecone or Qdrant. The AI Agent then uses a Vector Store Tool to query these embeddings at runtime. [4]

For high‑throughput RAG, consider batching the embedding generation using SplitInBatches to loop over chunks while respecting API rate limits. The final answer can be enriched with citations from the vector store metadata.

How do you add tools (function calls) and memory to an n8n AI agent?

Tools are sub‑nodes like Read/Write File, HTTP Request Tool, or custom Code Tool that the AI Agent can call based on user intent. Add them to the agent node, describe their function in natural language, and the model decides when to use them. Memory is enabled via a Window Buffer Memory sub‑node. [5]

The agent combines tools with memory to simulate a stateful assistant— it remembers previous turns and can execute multi‑step tasks. For production, pair this with the error workflow to retry failed tool calls and log the agent’s decisions.

How does n8n support Anthropic’s Claude for generative AI tasks?

n8n provides an Anthropic Chat Model sub‑node (for Claude 3 Opus, Sonnet, or Haiku) that works identically to the OpenAI node within the AI Agent. Select the model, set a system prompt and temperature, and handle the output. For image understanding, pass base64‑encoded images in a multi‑modal prompt block. [6]

Combine Claude with a vector store and memory for a complete RAG assistant that uses Anthropic’s constitutional AI safety features. For multi‑model fallback, place an HTTP Request node in a sub‑workflow that tries Claude first, then OpenAI on failure.