Snowflake

Author: hankookncompany Version: 0.0.1 Type: tool

Github Repository: https://github.com/hankookncompany/dify-plugin-snowflake

Overview

Snowflake for Dify is a plugin for Dify enables AI agents and workflows interact with data warehouse. It handles both generation and execution of SQL statements, via Snowflake's Cortex Analyst engine.

Configuration

1. Prerequisite

   This plugin supports the following two types of authentication method for Snowflake. Since the query generation cannot be performed with password authentication, we recommend creating a dedicated programmatic user on Snowflake, and providing both password and access token.

   • Password Auth : sql
   • Programmatic Access Token(PAT) : sql, cortex_analyst

2. Get Snowflake tools from Dify Plugin Marketplace

   The Snowflake plugin could be found at Dify Plugin Marketplace, please be patient while Dify Plugin Daemon preparing environment and installing requirements.

3. Fill Credentials for Snowflake

   On the Dify, go to Tools > Snowflake > Authentication to fill the credentials.

   Field	Description	Required
   account	Your Snowflake account identifier (e.g. AB123456 or AB123456.us-east-1.privatelink)	✅
   user	Your Snowflake user ID	✅
   password	Your Snowflake password	✅
   pat	Programmatic access token	❌
   warehouse	Snowflake warehouse to use (e.g. SMALL_WH)	❌
   role	Role to assume when connecting (e.g. CORTEX_USER)	❌

Usage

Execute SQL Statements

Simply put your SQL statements into sql block. The response will be a json with columnar structure.

/* input */
SELECT * FROM my_db.my_schema.my_table LIMIT 5;

/* response */
{
    "COLUMN_A" : [
        "1",
        "2",
        "3",
        "4",
        "5",
    ],
    "COLUMN_B" : [
        ...
    ],
    ...
}

Generate SQL Statements

Snowflake Cortex Analyst is a powerful assistant to retrieve data from database with natural language. This plugin provide a tool that enables using Cortex Analyst in your Dify workflows and agents.

To use Cortex Analyst, a detailed description for data, Semantic Model and Semantic View, should be pre-defined. Please reffer to the related quickstart from Snowflake.

Once your first semantic model has been built, browse into the stage, and click dropdown menu right on your semantic model, and then click copy path. The path should be in the format such as @my_db.my_schema.my_stage/my_model.yaml. Paste it into Cortex Analyst block.

/* question */
What are the top 10 sales by revenue for each product line?

/* response */
    {
        "text": "This is our interpretation of your question:\n\nWhat are the top 10 sales by revenue for each product line over the entire available time period?",
        "confidence": {
            "verified_query_used": null
        },
        "sql": "WITH __daily_revenue_by_product_line AS (\n  SELECT\n    product_line,\n    date,\n    revenue AS daily_revenue_per_product_line\n  FROM my_db.revenue_timeseries.daily_revenue_by_product\n), product_line_revenue AS (\n  SELECT\n    product_line,\n    date,\n    SUM(daily_revenue_per_product_line) AS total_revenue\n  FROM __daily_revenue_by_product_line\n  GROUP BY\n    product_line,\n    date\n), ranked_revenue AS (\n  SELECT\n    product_line,\n    date,\n    total_revenue,\n    RANK() OVER (PARTITION BY product_line ORDER BY total_revenue DESC NULLS LAST) AS rnk\n  FROM product_line_revenue\n)\nSELECT\n  product_line,\n  date,\n  total_revenue\nFROM ranked_revenue\nWHERE\n  rnk <= 10\nORDER BY\n  product_line,\n  rnk,\n  date DESC NULLS LAST\n -- Generated by Cortex Analyst\n;",
        
      "request_id": "ba60f783-dacc-47b1-9cac-xxxxxxxxxx",
      "metadata": {
        "cortex_search_retrieval": [],
        "model_names": [
          "gpt-4o",
          "claude-3-5-sonnet"
        ],

      "semantic_model_selection": null,
        "question_category": "CLEAR_SQL"
      },
      "warnings": []
    }

The response could contain additional inforamtions more than SQL statements, such as suggestions, text, or some others. To separate SQL statements from the response, please use the TEMPLATE block which is a build-in components of Dify.

/* Jinja2 */
{{ json.sql }}

License

Apache License 2.0

Privacy

This plugin does not collect or transmit data to third-party services.