用 Dify 搭建知识库客服机器人

上传支持文档到向量知识库，创建能检索答案的客服机器人，复杂问题转人工。

SaaS companies with 1000+ support tickets per month looking to deflect 40-60% of repetitive questions.

Dify account (cloud or self-hosted), knowledge base documents, OpenAI/Anthropic API key

工作流图

Upload Knowledge Base

Create Chatbot App

Configure Model Parameters

Add Escalation Logic

Deploy and Test

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
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
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
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
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.