用 Dify 搭建自主数据分析 Agent

AI Agent 接受自然语言提问，自动写 SQL、执行查询、生成图表和报告。

Data teams that get ad-hoc analysis requests from non-technical stakeholders and want to automate 60-80% of routine queries.

Dify with code execution enabled, read-only database access, OpenAI/Claude API key

工作流图

Database Connection

Agent Configuration

Query Safety Layer

Visualization Generation

Report Output

1
Database Connection

In Dify, add a database tool connection. Provide your PostgreSQL/MySQL connection string. Configure read-only access to prevent accidental writes. Set a row limit of 10,000 to prevent memory issues.
2
Agent Configuration

Create a new Agent app with ReAct (Reasoning + Acting) mode. Add the database tool and a code execution tool. Set the system prompt: 'You are a data analyst. When given a question, write SQL, execute it, analyze the results, and provide insights. Always explain your SQL logic before running it.'
3
Query Safety Layer

Add a pre-processing Code node that validates the SQL query before execution. Block DROP, DELETE, UPDATE, INSERT statements. Add a timeout of 30 seconds per query. Log all queries for audit.
4
Visualization Generation

After the agent gets query results, use a Code node to generate a chart using a JavaScript charting library (e.g., Chart.js). Convert the chart to a base64 image and embed it in the response.
5
Report Output

Format the final output as a structured report: 1) Executive summary, 2) SQL query used, 3) Data table (top 20 rows), 4) Chart visualization, 5) Key insights (3-5 bullet points), 6) Recommended actions. Save to your knowledge base for future reference.