Build custom AI agent with LangChain & Gemini (self-hosted)

Overview

This workflow leverages the LangChain code node to implement a fully customizable conversational agent. Ideal for users who need granular control over their agent's prompts while reducing unnecessary token consumption from reserved tool-calling functionality (compared to n8n's built-in Conversation Agent).

Setup Instructions

1. Configure Gemini Credentials: Set up your Google Gemini API key (Get API key here if needed). Alternatively, you may use other AI provider nodes.
2. Interaction Methods:
   • Test directly in the workflow editor using the "Chat" button
   • Activate the workflow and access the chat interface via the URL provided by the When Chat Message Received node

Customization Options

1. Interface Settings: Configure chat UI elements (e.g., title) in the When Chat Message Received node
2. Prompt Engineering:
   • Define agent personality and conversation structure in the Construct & Execute LLM Prompt node's template variable
   • ⚠️ Template must preserve {chat_history} and {input} placeholders for proper LangChain operation
3. Model Selection: Swap language models through the language model input field in Construct & Execute LLM Prompt
4. Memory Control: Adjust conversation history length in the Store Conversation History node

Requirements:

⚠️ This workflow uses the LangChain Code node, which only works on self-hosted n8n.
(Refer to LangChain Code node docs)