The Birth of a Sales Data Analysis Bot: How Dify Achieves Automated Closed-Loop with DMS

Background & Challenges

Dify is a low-code AI application development platform that leverages intuitive visual workflow orchestration to lower barriers to large model application development.

However, during enterprise-level deployments, two major limitations emerge:

Key Bottlenecks

• Limited Code Execution
• The built-in Sandbox in Dify supports basic Python execution but cannot install custom packages.
• This prevents implementing complex logic, data processing, or integrating enterprise SDKs.
• No Automated Scheduling
• Lacks support for scheduled triggers, periodic runs, or orchestrated dependencies for Agents/Workflows.
• Limits integration into enterprise automation systems.

These restrictions hinder Dify’s deployment in production, especially for intelligent Agents requiring a fully closed-loop Perception → Decision → Execution → Feedback workflow.

---

Proposed Enhanced Architecture: Dify + DMS Notebook + DMS Airflow

Components

• ✅ DMS Notebook – Full Python environment, custom pip installs, environment variables, async services, secure VPC communication.
• ✅ DMS Airflow – Robust scheduler to trigger workflows/notebook scripts with dependency management.
• ✅ DMS Integration – End-to-end development, debugging, deployment, scheduling, and monitoring.

We’ll demonstrate this by building a Sales Data Analysis Bot that is schedulable, extensible, and maintainable.

---

Extending Dify with DMS Notebook

Why Use Notebook?

Sandbox is insufficient for:

• Pandas data cleaning
• Prophet forecasting
• Enterprise SDK integration

DMS Notebook offers:

• Custom pip install support
• Environment variables (API keys, etc.)
• Async services via FastAPI
• Secure VPC communications

---

Step-by-Step Guide

1. Create a Notebook Session

• Navigate to: `DMS Console → Notebook Session → Create Session`
• Define Parameters:
• Select Python version:
• Edit Settings:
• Dependencies:
• Add to `requirements.txt`:

   pandas
   fastapi
   uvicorn
   nest-asyncio

Environment Variables:

• `ALIBABA_CLOUD_ACCESS_KEY_ID`
• `ALIBABA_CLOUD_ACCESS_KEY_SECRET`
• Large model `API_KEY`

Configuration Tips:

• Install `fastapi`, `uvicorn`, `nest-asyncio`
• Resource release time = `0`
• Set `DMS_KERNEL_IDLE_TIMEOUT=0` → prevents kernel being killed

💡 Tip: Without setting `DMS_KERNEL_IDLE_TIMEOUT=0`, long-running APIs may fail due to idle kernel termination.

---

2. Write & Start FastAPI Service

• Create new notebook under default library
• Reference official FastAPI docs: fastapi.tiangolo.com

Example FastAPI template:

import os
from fastapi import FastAPI, HTTPException, Request
from fastapi.staticfiles import StaticFiles
import nest_asyncio

nest_asyncio.apply()
app = FastAPI(title="Service Name")

# Static files
static_dir = "static"
os.makedirs(static_dir, exist_ok=True)
app.mount("/static", StaticFiles(directory=static_dir), name="static")

@app.get("/")
async def root():
    return {"message": "Service is running"}

Refer to subsequent provided code for handling path/query/header/form/file uploads.

---

3. Service Deployment

• Run the notebook cell to start Uvicorn within the Jupyter async loop.
• Verify service start:
• !image
• !image
• Fetch Notebook session’s VPC IP:
• Example: `172.16.0.252`

  !ifconfig

---

4. Integrate with Dify Workflow

• Add HTTP Request node with API URL `http://:/endpoint`
• Pass parameters in request body.
• Test:

---

Example: Daily Sales Analysis API

Provided code:

• File Processing: CSV validation, parsing with Pandas
• Metrics: Total Sales, Orders, Quantity, Average Order Value
• Analysis: Top products, regions, price ranges
• Insights: Sales performance commentary

Key endpoints:

• `/analysis/daily_sale_analysis` – Upload CSV → JSON analysis output
• `/upload-file` – Multi-part form file uploads
• `/status` – API health check

---

Using DMS Airflow for Scheduling

Steps:

• Create Airflow Instance
• Docs:
• Purchase Airflow Resources
• Create & Manage Instance
• Airflow Official Docs
• Write Airflow DAG – Example provided in code:
• Reads CSV file
• Uploads via `/files/upload`
• Triggers Dify workflow `/workflows/run`
• Scheduled daily at 08:00
• View Scheduled Tasks:
• Result:
• DingTalk bot pushes daily report

---

Summary & Best Practices

Architecture:

`Dify + DMS Notebook + DMS Airflow` enables:

• Agile low-code orchestration
• Flexible code execution
• Reliable scheduled automation

Core Concept:

An enterprise AI Agent should be extensible, schedulable, and maintainable — blending low-code development with robust engineering practices.

---

Join Community:

• DingTalk code: `96015019923` → Dify on DMS User Group
• Dify app link: Apply here

---

Tip for Content Creators:

For AI-driven analysis that needs cross-platform publishing, consider AiToEarn官网.

An open-source AI content monetization platform enabling creators to:

• Generate AI content
• Publish simultaneously to Douyin, Kwai, WeChat, Bilibili, Facebook, Instagram, LinkedIn, YouTube, Pinterest, X, etc.
• Analyze engagement and model rankings

Integrating this with Dify + DMS automation empowers both internal intelligence workflows and external content monetization.