Build a Knowledge-Base Customer Support Bot with Dify

This Dify workflow uploads your support docs into a vector knowledge base, creates a chatbot that retrieves relevant answers, and routes complex questions to a human agent queue with full conversation context.

Workflow Diagram

Step-by-Step Guide

1. 1

   Upload Knowledge Base

   In Dify, create a new Knowledge Base. Upload PDF, Markdown, or text files of your FAQ, product docs, and support tickets. Configure chunking strategy: 500 tokens per chunk, 10% overlap.

2. 2

   Create Chatbot App

   Create a new Chat App in Dify. Select your knowledge base as the context. Set the system prompt: 'You are a support agent for [COMPANY]. Answer questions using only the provided knowledge base. If you cannot find an answer, say you will connect them to a human agent.'

3. 3

   Configure Model Parameters

   Set temperature to 0.3 for factual responses. Set max tokens to 500. Enable citation display so the bot can show which doc section the answer came from.

4. 4

   Add Escalation Logic

   Use Dify's workflow mode to add a conditional node: if the bot confidence score is below threshold (or user asks for human), trigger a webhook to your ticketing system (Zendesk/Intercom) with the full conversation.

5. 5

   Deploy and Test

   Embed the chatbot widget on your website or connect via API. Run 20 test conversations covering common, edge, and unknown questions. Adjust the knowledge base based on gaps found.

Tags

Back to Workflow Library