Chat with PostgreSQL Database via AI Agent

Overview

This workflow enables you to interact with any PostgreSQL database using natural language. An AI agent (powered by OpenAI) translates user questions into SQL queries, executes them against your database, and returns the results in a conversational chat interface. It is ideal for non-technical team members who need to query data without writing SQL, or for developers who want a quick, interactive way to explore their database.

How It Works

The workflow consists of a chat trigger that receives messages, an AI agent that orchestrates the conversation, an OpenAI model for language understanding, and three PostgreSQL tools that the agent can call to inspect the database schema and run queries. A memory buffer keeps the conversation context for multi-turn interactions.

Node-by-Node Breakdown

• When chat message received (@n8n/n8n-nodes-langchain.chatTrigger) — No auth. This node creates a webhook URL that n8n’s built-in chat interface uses. When a user sends a message, it triggers the workflow. Parameters: no additional options.

• AI Agent (@n8n/n8n-nodes-langchain.agent) — No auth. This is the brain of the workflow. It uses the OpenAI Functions agent pattern and receives a system message instructing it to act as a database assistant. It can call the three PostgreSQL tools defined below. Parameters: agent = openAiFunctionsAgent, options.systemMessage = detailed instructions.

• OpenAI Chat Model (@n8n/n8n-nodes-langchain.lmChatOpenAi) — API Key auth (requires an OpenAI API key). This node provides the language model (gpt-4o-mini) that powers the AI agent. Parameters: model = gpt-4o-mini.

• Chat History (@n8n/n8n-nodes-langchain.memoryBufferWindow) — No auth. Stores the last few messages (default 5) to maintain conversation context. Parameters: none.

• Execute SQL Query (n8n-nodes-base.postgresTool) — Database credentials (PostgreSQL). This tool executes any SQL query generated by the AI agent. The query is dynamically filled using $fromAI("sql_query"). Parameters: operation = executeQuery, toolDescription = instructs the agent to append schema prefixes.

• Get DB Schema and Tables List (n8n-nodes-base.postgresTool) — Database credentials (PostgreSQL). This tool returns a list of all tables and their schemas by querying information_schema.tables. Parameters: static query, operation = executeQuery, toolDescription = describes the output.

• Get Table Definition (n8n-nodes-base.postgresTool) — Database credentials (PostgreSQL). This tool returns column details, types, foreign keys, etc., for a given table and schema. The parameters use $fromAI("table_name") and $fromAI("schema_name") to get the table name from the AI. Parameters: query with joins, operation = executeQuery, toolDescription = describes the output.

Setup Instructions

1. PostgreSQL Credentials: You need a PostgreSQL database accessible from n8n. Create a database credential in n8n with the host, port, database name, username, and password. The user must have read access to information_schema and to the tables you want to query.

2. OpenAI API Key: Obtain an API key from OpenAI. Create an OpenAI credential in n8n and paste the key.

3. Activate the Workflow: After adding credentials, save and activate the workflow. A chat button will appear in the n8n editor. Click it to start a conversation.

4. Make Public (Optional): The workflow can be made publicly accessible by sharing the webhook URL from the chat trigger node. Ensure your database is secure before exposing it publicly.

Use Cases & Customization

• Analytics for non-technical users: Let business analysts ask questions like “Show me total sales by region last quarter” without writing SQL.
• Database exploration: Quickly understand your schema by asking “What tables are in the public schema?”.
• Custom model: Swap the OpenAI model with another provider (e.g., Anthropic, Ollama) by replacing the OpenAI Chat Model node.
• Security: You can restrict the AI agent by limiting the tools or modifying the system prompt to only allow SELECT queries.
• Multi-turn conversations: The memory buffer keeps context, so follow-up questions like “And for the previous month?” work seamlessly.

This workflow turns your database into a conversational assistant, saving time and lowering the barrier to data access.

{
  "id": "eOUewYsEzJmQixI6",
  "meta": {
    "instanceId": "77c4feba8f41570ef06dc76ece9a6ded0f0d44f7f1477a64c2d71a8508c11faa",
    "templateCredsSetupCompleted": true
  },
  "name": "Chat with Postgresql Database",
  "tags": [],
  "nodes": [
    {
      "id": "6501a54f-a68c-452d-b353-d7e871ca3780",
      "name": "When chat message received",
      "type": "@n8n/n8n-nodes-langchain.chatTrigger",
      "position": [
        -300,
        -80
      ],
      "webhookId": "cf1de04f-3e38-426c-89f0-3bdb110a5dcf",
      "parameters": {
        "options": {}
      },
      "typeVersion": 1.1
    },
    {
      "id": "cd32221b-2a36-408d-b57e-8115fcd810c9",
      "name": "AI Agent",
      "type": "@n8n/n8n-nodes-langchain.agent",
      "position": [
        0,
        -80
      ],
      "parameters": {
        "agent": "openAiFunctionsAgent",
        "options": {
          "systemMessage": "You are DB assistant. You need to run queries in DB aligned with user requests.\n\nRun custom SQL query to aggregate data and response to user. Make sure every table has schema prefix to it in sql query which you can get from `Get DB Schema and Tables List` tool.\n\nFetch all data to analyse it for response if needed.\n\n## Tools\n\n- Execute SQL query - Executes any sql query generated by AI\n- Get DB Schema and Tables List - Lists all the tables in database with its schema name\n- Get Table Definition - Gets the table definition from db using table name and schema name"
        }
      },
      "typeVersion": 1.7
    },
    {
      "id": "8accbeeb-7eaf-4e9e-aabc-de8ab3a0459b",
      "name": "OpenAI Chat Model",
      "type": "@n8n/n8n-nodes-langchain.lmChatOpenAi",
      "position": [
        -60,
        160
      ],
      "parameters": {
        "model": {
          "__rl": true,
// ... truncated (copy to see full JSON)