n8n: A Practical Guide to Building AI Agents for Automation and RAG

n8n offers two primary ways to use its platform: via its web interface, which requires an account and a paid subscription (though a 14-day free trial is available), or by running it locally or self-hosting. The open-source nature of n8n allows for free local execution, granting access to most features. For local setup, the simplest method involves downloading and installing Node.js from its official website. Once Node.js is installed, users can open a terminal and execute the command `npx n8n`. This command initiates the n8n instance, and users can access the interface by pressing 'o' in the terminal or by navigating to the provided localhost URL (e.g., `http://localhost:5678`).

This section details the construction of a practical workflow to automate invoice processing from emails. The scenario involves a specific email address where invoices are received. The content of these emails is then sent to ChatGPT to extract key information such as the invoice ID and the total amount due. This extracted data is subsequently added to a shared spreadsheet. This automation is particularly useful for managing shared expenses, like rent for a shared property, where invoices need to be divided and tracked.

Following the Gmail trigger, the next node in the workflow is an OpenAI node, specifically the 'Message to a model' action. This node is responsible for sending the email content to a Large Language Model (LLM) for processing. Similar to Gmail, an OpenAI API key is required to authenticate. Users can create this key from the OpenAI platform. The configuration involves selecting an AI model, such as GPT 4.1, and crafting a prompt that instructs the model to extract the invoice ID and the total amount due. The prompt can dynamically include data from previous nodes, like the email snippet, using n8n's variable syntax (e.g., `{{ $json.snippet }}`). To refine the output, the LLM's response can be set to JSON, allowing for the extraction of specific fields directly, eliminating the need for further string processing.

This section introduces the construction of a more complex AI agent using the Retrieval-Augmented Generation (RAG) technique. RAG combines information retrieval from a data source with an LLM's generation capabilities to produce responses based on specific knowledge. This is ideal for creating agents that can answer questions about a defined knowledge base, such as a long document. The example used is building an agent based on the rules of a card game to resolve rule disputes during gameplay. The process involves two main workflows: one to load data into a Pinecone vector database and another to power the RAG agent that queries this database.

The final workflow constructs the RAG agent itself. It begins with an 'On Chat Message' trigger node, designed for chat-based workflows. This is connected to an 'AI Agent' node, which can be configured with an AI model (e.g., OpenAI Chat Model with 'gpt-4.1'), memory (e.g., 'Simple Memory' with a context window of 5 interactions), and tools. The Pinecone vector database is provided as a tool, enabling the agent to access the rules document. A description field within the tool configuration is crucial for the agent to determine when to use this tool. The embedding model used by the vector store is also configured, again using an OpenAI embedding node. Once complete, users can interact with the agent via chat, and the agent will utilize the Pinecone database to retrieve relevant information for generating responses, as demonstrated in the example where the agent answers a rule-based question.

n8n provides a vast ecosystem of integrations, enabling users to connect over a thousand services and tools to build sophisticated AI agents. This tutorial has only scratched the surface of n8n's capabilities. By exploring how to use n8n for automating everyday tasks and building AI agents, users can begin to harness its full potential. The platform's visual workflow builder, extensive integrations, and open-source nature make it a powerful tool for both beginners and experienced developers looking to streamline processes and enhance productivity with AI.