原文要点

Documentation Index Fetch the complete documentation index at: /llms.txt Use this file to discover all available pages before exploring further. Skip to main content Dify Documentation Dify is an open-source platform for building AI applications. Create agents, agentic workflows, and chatbots that draw on your own data, then publish them as web apps or integrate them through APIs. Build Your First App New to Dify? Start here: build a working AI application and pick up the core concepts along the way. Choose How You Run Dify Dify Cloud Start building right away on the managed platform. No installation, free Sandbox plan included. Self-Host Run the open-source Community Edition on your own infrastructure. Deploy with Docker Compose in minutes. Explore More API Reference Integrate Dify applications into your own product through the REST API. CLI Run your Dify apps from terminals, scripts, CI, and AI agents with difyctl. Develop Plugins Extend Dify with your own tools, models, and integrat

原文完整内容

Documentation Index

Fetch the complete documentation index at:

/llms.txt

Use this file to discover all available pages before exploring further.

Skip to main content

Dify Documentation

Dify is an open-source platform for building AI applications. Create agents, agentic workflows, and chatbots that draw on your own data, then publish them as web apps or integrate them through APIs.

Build Your First App

New to Dify? Start here: build a working AI application and pick up the core concepts along the way.

Choose How You Run Dify

Dify Cloud

Start building right away on the managed platform. No installation, free Sandbox plan included.

Self-Host

Run the open-source Community Edition on your own infrastructure. Deploy with Docker Compose in minutes.

Explore More

API Reference

Integrate Dify applications into your own product through the REST API.

CLI

Run your Dify apps from terminals, scripts, CI, and AI agents with difyctl.

Develop Plugins

Extend Dify with your own tools, models, and integrations.

Tutorials

Step-by-step lessons, from your first workflow to complete example apps.

Additional Resources

六、MVP 范围

先做下面三个能力，不要一开始就做复杂 Agent：

  1. 文档导入与切块。
  2. 检索 + 生成回答。
  3. 回答里给引用。

等你看见真实查询分布以后，再去优化 reranker、文档加权或图谱。

七、最小验收

  • 20 个真实问题里至少 12 个能给出正确来源。
  • 错误回答不是凭空编造，而是能定位到检索失败或模型理解失败。
  • 一脸数据时命中知识点 > 60%。
  • 每次失败都能记录原因。

八、常见坑

  • 所有文档用同一个切块大小，但实际上公式、代码、长表格需要不同处理。
  • 一开始就做复杂多轮 Agent，导致排查困难。
  • 没有统一引用来源，用户无法信任回答。
  • 检索和生成同时失败时，没有区分责任。
  • 把向量库和知识图谱都堆在一起，导致维护成本失控。

先用简单方法跑通，再让数据决定下一步该优化哪一块。

来源：https://docs.dify.ai/