Business Central Code Interpreter

Overview

This project provides a natural language data analysis solution for Microsoft Dynamics 365 Business Central. It enables users to ask business questions in plain language and receive insightful answers with visualizations, powered by Azure OpenAI and Azure Functions.

The system works by:

1. Taking natural language questions from users
2. Generating Python code to analyze the requested data
3. Executing that code securely in the cloud
4. Processing Business Central data through its API
5. Presenting user-friendly results back to the business user

Demo

Watch the showcase video to see the Business Central Code Interpreter in action:

Full.Process.-.No.Retry.-.Original.mp4

For more details, read the full blog post.

Repository Structure

This repository contains the following components:

• Business Central Extension: The AL extension that integrates with Business Central and provides the user interface for interacting with the Code Interpreter
• Azure Function: The secure Python execution environment that processes data from Business Central

Architecture

sequenceDiagram
    participant User
    participant BC Extension
    participant Azure Function
    participant LLM
    participant BC API

    User->>BC Extension: Ask question (e.g. "What were my top 5 customers in Q1?")
    BC Extension->>LLM: Generate Python code (system prompt)
    LLM-->>BC Extension: Return Python code
    
    BC Extension->>Azure Function: Execute Python code via HTTP POST
    Azure Function->>BC API: Call Business Central API with token
    BC API-->>Azure Function: Return API data
    Azure Function-->>BC Extension: Return JSON result or error
    
    rect rgb(85, 85, 85)
        Note over BC Extension,Azure Function: Retry Logic (up to 3 attempts)
        BC Extension->>LLM: Analyze error, review data & generate improved code
        LLM-->>BC Extension: Return improved Python code
        BC Extension->>Azure Function: Retry execution with improved code
    end
    
    BC Extension->>LLM: Final summary prompt with result
    LLM-->>BC Extension: Return natural language answer
    BC Extension-->>User: Show answer with visualizations

Loading

Prerequisites

• Microsoft Dynamics 365 Business Central (2024 Wave 2/v26.0 or higher)
• Azure subscription with:
  • Azure OpenAI service (GPT-4.1 model with 50,000+ token rate limit recommended; see note below)
  • Azure Function App (Python 3.10+ runtime)
• Business Central environment with API access configured
• App registration in Azure Active Directory with appropriate permissions

Note: For best results, deploy a GPT-4.1 or GPT 4o model with a high token rate limit (e.g., 50,000 tokens/minute). The model with much lower rate limit (e.g., 8,000 tokens/minute) can cause rate limit errors. If you encounter rate limit issues, consider requesting a quota increase or deploying a GPT-4.1 model. See Azure OpenAI quota increase for more information.

Getting Started

1. Azure OpenAI Setup

   • Create an Azure OpenAI resource
   • Deploy a model (GPT-4 recommended)
   • Note the endpoint and API key

2. Azure Function Setup

   • Create a Function App with Python 3.10+ runtime
   • Register an app in Azure AD with Business Central API permissions (API.ReadWrite.All)
   • Configure environment variables (BC_TENANT_ID, BC_CLIENT_ID, BC_CLIENT_SECRET)
   • Deploy the function code from the azure-function directory

3. Business Central Extension Setup

   • Deploy the AL extension to your Business Central environment
   • Configure OpenAI and Function details in the setup page
   • Activate the Copilot capability in BC settings

4. Start asking business questions in natural language!

Sample Questions

Here are some examples of questions you can ask:

• "What were my top 5 customers by sales last quarter?"
• "Show me the inventory items with the highest turnover rate"
• "Analyze the trend of overdue invoices over the past 6 months"
• "Which products have the highest profit margin?"
• "Compare sales performance across regions for the last fiscal year"
• "Identify customers with declining purchase patterns"

How It Works

1. User Query: The user submits a natural language question through the Business Central interface
2. Code Generation: Azure OpenAI generates Python code based on the question
3. Code Execution: The code is executed securely in an Azure Function
4. Data Retrieval: The function retrieves data from Business Central using the API
5. Data Processing: Python libraries (pandas, numpy, matplotlib) analyze and visualize the data
6. Smart Retry: If errors occur, the system analyzes them and automatically improves the code
7. Answer Formation: Results are passed back to Azure OpenAI to create a human-friendly answer
8. Presentation: The user receives an answer with text and visualizations

Security and Privacy

This solution prioritizes security and privacy:

• All data processing happens within your Azure subscription
• No business data is sent to external services outside your control
• Azure OpenAI is only used for code generation and result interpretation
• The Azure Function execution environment is securely sandboxed
• Authentication to Business Central API uses secure OAuth2 with minimal permissions
• Dangerous Python modules and functions are blocked in the execution environment

For detailed security information, see the Azure Function README.

Development and Customization

You can extend this solution by:

• Adding new helper functions to the Azure Function
• Enhancing the prompts used for code generation and result interpretation
• Creating specialized UI elements for specific types of questions
• Adding support for additional visualizations and data presentation options
• Implementing custom data transformations for specific business needs

For local development instructions, see the component-specific README files.

Troubleshooting

If you encounter issues:

• Check Azure OpenAI and Function settings in the Business Central setup page
• Verify Copilot capability is activated in Business Central
• Ensure Function URL format is correct (should include /api/execute)
• Check Application Insights logs for detailed error messages
• Verify environment variables are correctly set in Azure Function
• For complex operations, consider upgrading to a premium Function plan

License

This project is provided under the MIT License.

Support

For support, please contact dmitry@katson.com.

---

Empowering Business Central users with AI-driven data analysis