How to build a Snowflake Analytics Agent

Workflow Overview

This flow is designed to answer user questions about a Snowflake data warehouse, using both a data dictionary and live database queries, and then provides a C-suite-friendly, well-formatted answer.

1. User Input

• Node: Input

• Purpose: The user enters a question or request (e.g., “Which customers have placed the most orders over the past 30 years?”).

2. Data Dictionary Lookup

• Node: Data Dictionary (Knowledge Base)

• Purpose:

  • The user’s question is sent to a knowledge base (in this case, a data dictionary stored in OneDrive).

  • This node helps provide context about the database structure, common analysis patterns, and available fields/tables.

  • It’s useful for both the AI and the user to understand what kinds of questions are possible and how the data is organized.

3. Snowflake Database Query

• Node: Query Snowflake (Action Node)

• Purpose:

  • To tell you more about this tool, let me inspect its exact properties.

(Fetching details about the "Query Snowflake" action...)Here’s how the Query Snowflake action node works in your flow:

• Node: Query Snowflake (Action Node)

• Purpose:

  • This node takes the user’s question (in plain English or SQL) and the list of available tables.

  • It translates the question into a Snowflake SQL query and runs it against your Snowflake database.

  • The output includes:

    • The actual SQL query that was executed.

    • The results of that query (as a table of data).

4. AI-Powered Analytics Explanation

• Node: Anthropic (LLM Node)

• Purpose:

  • The results from the Snowflake query, along with the original user request, are sent to an advanced AI model (Anthropic Claude 4.5 Sonnet).

  • The AI is instructed to:

    • Read the query results and the user’s question.

    • Respond in clear, concise language suitable for C-suite executives.

    • Use Markdown formatting for tables and headers.

    • If the query is incorrect or cannot be answered, the AI will inform the user.

5. Output to User

• Node: Output

• Purpose:

  • The AI’s response is displayed to the user as the final output.

6. Sticky Note (for Builder Reference)

• Node: Sticky Note

• Purpose:

  • This is a builder’s note reminding you that the data is from the 1990s, so questions about recent years (like “last 12 months”) are not appropriate.

  • It also gives example questions you can ask.

    
    

Summary Table

In short:
This flow lets a user ask a business question, checks the data dictionary for context, runs a live query on Snowflake, and then uses AI to explain the results in a way that’s easy for executives to understand.