Chat with SQLite Database using AI Agent (OpenAI)

Overview

This workflow turns any SQLite database into a conversational AI assistant. By combining a Chat Trigger (webhook) with an AI Agent powered by OpenAI's GPT-4 Turbo, you can ask natural language questions about your data and get answers without writing a single SQL query. The agent automatically generates and executes SQL, interprets results, and maintains conversation history via Window Buffer Memory.

This is especially useful for data analysts, product managers, or anyone who needs to explore a database quickly. The workflow includes a one-time setup to download and save a sample SQLite database (Chinook), so you can test it immediately.

Node-by-Node Breakdown

Setup Section (Run Once)

• When clicking "Test workflow" (Manual Trigger) — No auth. Starts the one-time download process when you click "Test Workflow" in the editor.
• Get chinook.zip example (HTTP Request) — No auth. Downloads the sample Chinook database as a ZIP file from https://www.sqlitetutorial.net/wp-content/uploads/2018/03/chinook.zip. No authentication required.
• Extract zip file (Compression) — No auth. Decompresses the downloaded ZIP archive to extract the chinook.db file.
• Save chinook.db locally (Read/Write File) — No auth. Writes the extracted binary file to ./chinook.db on the n8n server's file system. This file is used later by the agent.

Chat Section (On Every Message)

• Chat Trigger (Chat Trigger) — No auth. Listens for incoming chat messages via a webhook URL. When a user sends a message, this node triggers the workflow.
• Load local chinook.db (Read/Write File) — No auth. Reads the chinook.db file from the local file system and passes it as binary data to the next node.
• Combine chat input with the binary (Set) — No auth. Merges the user's chat message (JSON from Chat Trigger) with the binary SQLite file data. This ensures the AI agent has both the question and the database content.
• AI Agent (Agent) — No auth (but uses OpenAI credentials). Configured as a SQL Agent with dataSource: sqlite. It receives the combined input and uses the connected language model and memory to generate SQL queries, execute them against the SQLite database, and produce a natural language answer.
• OpenAI Chat Model (Language Model) — API Key auth. Uses gpt-4-turbo with a temperature of 0.3 for deterministic outputs. You must provide an OpenAI API key in n8n credentials.
• Window Buffer Memory (Memory Buffer Window) — No auth. Stores the last 10 conversation turns (context window length = 10) so the agent can recall previous questions and answers.

Sticky Notes (Informational)

• Sticky Note — Explains the one-time setup steps.
• Sticky Note1 — Explains that on each chat message, the SQLite file is loaded and combined with chat input.
• Sticky Note2 — Provides example queries and notes that the agent may run multiple SQL queries before answering.

Setup Instructions

1. OpenAI Account: You need an OpenAI API key. Create one at platform.openai.com and add it as a credential in n8n (type: OpenAI).
2. n8n Instance: This workflow runs on n8n (self-hosted or cloud). Ensure file system access is enabled (for reading/writing local files).
3. One-Time Setup: Activate the workflow, then click "Test Workflow" to download and save the Chinook database. After that, you can deactivate or keep the setup section; it only needs to run once.
4. Chat: Copy the webhook URL from the Chat Trigger node. Use any HTTP client (e.g., Postman, curl) or embed it in a chat UI. Send a JSON payload like {"message": "What are the top 5 genres by revenue?"}. The agent will respond with the answer.

Use Cases & Variations

• Query your own SQLite database: Replace the download/save steps with a node that reads your own .db file (e.g., from a cloud storage or upload).
• Switch to PostgreSQL or MySQL: Change the AI Agent's dataSource to postgres or mysql and provide connection credentials. You'll also need to replace the file-loading nodes with a database connection node.
• Add authentication to the chat: Use an HTTP node with basic auth or an API key before the Chat Trigger to secure the endpoint.
• Integrate with Slack or Teams: Replace the Chat Trigger with a Slack/Teams webhook node to let users query the database directly from their messaging app.
• Use a different LLM: Swap the OpenAI Chat Model for another provider (e.g., Anthropic, Ollama) by changing the language model node and credentials.

{
  "id": "AQJ6QnF2yVdCWMnx",
  "meta": {
    "instanceId": "fb924c73af8f703905bc09c9ee8076f48c17b596ed05b18c0ff86915ef8a7c4a",
    "templateCredsSetupCompleted": true
  },
  "name": "SQL agent with memory",
  "tags": [],
  "nodes": [
    {
      "id": "3544950e-4d8e-46ca-8f56-61c152a5cae3",
      "name": "Window Buffer Memory",
      "type": "@n8n/n8n-nodes-langchain.memoryBufferWindow",
      "position": [
        1220,
        500
      ],
      "parameters": {
        "contextWindowLength": 10
      },
      "typeVersion": 1.2
    },
    {
      "id": "743cc4e7-5f24-4adc-b872-7241ee775bd0",
      "name": "OpenAI Chat Model",
      "type": "@n8n/n8n-nodes-langchain.lmChatOpenAi",
      "position": [
        1000,
        500
      ],
      "parameters": {
        "model": "gpt-4-turbo",
        "options": {
          "temperature": 0.3
        }
      },
      "credentials": {
        "openAiApi": {
          "id": "rveqdSfp7pCRON1T",
          "name": "Ted's Tech Talks OpenAi"
        }
      },
      "typeVersion": 1
    },
    {
      "id": "cc30066c-ad2c-4729-82c1-a6b0f4214dee",
      "name": "When clicking \"Test workflow\"",
      "type": "n8n-nodes-base.manualTrigger",
      "position": [
        500,
// ... truncated (copy to see full JSON)