AI Customer Support Chatbot with Elasticsearch Knowledge Base

Overview

This workflow transforms a webhook endpoint into an intelligent customer support chatbot. It receives a customer question, searches a knowledge base stored in Elasticsearch, and generates a helpful, context-aware reply using OpenAI's GPT-4. The bot is designed to reduce human workload by answering common queries instantly, and gracefully escalates to a human agent when the answer is not found. It's ideal for embedding into a website, messaging app, or any frontend that can send HTTP POST requests.

Node-by-Node Breakdown

1. Chatbot Webhook (Frontend Trigger) — Type: Webhook (No auth)

   • Listens for POST requests at the path /support-chatbot. The webhook ID is customer-support-chat. This is the entry point where your frontend sends the customer's question as JSON body (e.g., {"question": "How do I reset my password?"}).

2. Extract Customer Query — Type: Function (No auth)

   • Extracts question from the incoming JSON body using a simple JavaScript code snippet: const question = $json.body.question; return [{ json: { question } }];. This prepares the query for the next steps.

3. Search Knowledge Base (Elasticsearch) — Type: Elasticsearch (Requires Elasticsearch credentials – basic auth / API key)

   • Performs a search on the support-kb index using the extracted question as the query. The query parameter is set to ={{$json.question}}. It returns matching documents from the knowledge base.

4. Generate AI Response with GPT-4 — Type: OpenAI (API Key auth)

   • Uses the gpt-4 model to generate a response. The system message instructs the assistant to use the provided knowledge base content and to politely escalate if the answer is not found. The user message combines the customer question and the top knowledge base result ({{$json.hits.hits[0]._source.content}}).

5. Format Chatbot Reply — Type: Function (No auth)

   • Extracts the final reply from the OpenAI response: return [{ json: { reply: $json.choices[0].message.content } }];. This ensures the output is properly structured.

6. Return Chatbot Reply — Type: Respond to Webhook (No auth)

   • Sends the formatted reply back to the caller (the frontend) as a JSON response.

Setup Instructions

1. Elasticsearch: You need a running Elasticsearch instance (cloud or self-hosted) with an index named support-kb. Populate it with documents containing your knowledge base articles (e.g., each document should have a content field). Configure the Elasticsearch node in n8n with your host, port, and authentication credentials (API key or username/password).
2. OpenAI: Obtain an API key from OpenAI. Add it as a credential in n8n (type: OpenAI API).
3. Webhook URL: After activating the workflow, n8n will generate a unique webhook URL (e.g., https://your-n8n-host/webhook/support-chatbot). Use this URL in your frontend (website, chatbot widget, etc.) to send POST requests.
4. Activate the workflow: Toggle the workflow to active in n8n for the webhook to start listening.

Use Cases & Adaptations

• Customer Support Portal: Embed the webhook URL in a live chat widget on your website to answer FAQs instantly.
• Slack or Discord Bot: Replace the webhook trigger with a Slack/Discord trigger and adjust the response format to send messages back to the channel.
• Multiple Knowledge Bases: Modify the Elasticsearch query to search across multiple indices or filter by category.
• Fallback to Human: Customize the GPT-4 system message to trigger a notification (e.g., email or Slack message) when the answer is not found.
• Model Upgrade: Swap gpt-4 for gpt-3.5-turbo to reduce cost or use a custom fine-tuned model.
• Language Support: Add a language detection step before the query to route to different knowledge bases or translate the response.

{
  "name": "ai-chatbot-customer-support-kb",
  "nodes": [
    {
      "parameters": {
        "httpMethod": "POST",
        "path": "support-chatbot"
      },
      "id": "1",
      "name": "Chatbot Webhook (Frontend Trigger)",
      "type": "n8n-nodes-base.webhook",
      "typeVersion": 1,
      "position": [
        100,
        300
      ],
      "webhookId": "customer-support-chat"
    },
    {
      "parameters": {
        "functionCode": "const question = $json.body.question;\nreturn [{ json: { question } }];"
      },
      "id": "2",
      "name": "Extract Customer Query",
      "type": "n8n-nodes-base.function",
      "typeVersion": 1,
      "position": [
        300,
        300
      ]
    },
    {
      "parameters": {
        "operation": "search",
        "index": "support-kb",
        "query": "={{$json.question}}"
      },
      "id": "3",
      "name": "Search Knowledge Base (Elasticsearch)",
      "type": "n8n-nodes-base.elasticSearch",
      "typeVersion": 1,
      "position": [
        500,
        300
      ],
      "credentials": {
        "elasticSearchApi": {
          "id": "elasticsearch_credential",
          "name": "ElasticSearch"
        }
// ... truncated (copy to see full JSON)