Edit

Share via

Run Python code that processes files in agent workflows with Azure Logic Apps (Preview)

Applies to: Azure Logic Apps (Standard)

Note

This capability is in preview and is subject to the Supplemental Terms of Use for Microsoft Azure Previews.

In Standard logic apps, not only can agent workflows handle natural language requests through chat interfaces by using agents, they can also process files when you create agent tools with the Code Interpreter (Python Container Apps session) built-in connector operations. These operations make it possible for the agent workflow to generate and run Python code, upload, download, and delete files. From a scenario perspective, the agent can receive instructions, accept files, generate and run code in an isolated environment, and return usable results, such as summaries, forecasts, or other requested data transformations - all within the same workflow.

Note

This article assumes that you're working with an agent workflow and use the Code Interpreter operations inside agent tools so that you can have the agent generate the necessary Python code.

The following example shows a workflow with an agent action named Sales Report Agent. The agent can use a tool named Upload sales data to run the Code Interpreter operations named Upload file and Execute Python code.

Screenshot shows workflow with agent and tool, which uses Code Interpreter actions that upload files to Container Apps session pool and runs Python code for file processing.

When the workflow receives an HTTPS request for a specified file through the Request trigger, the HTTP action gets the requested file. The agent has instructions to upload the file to a Container Apps session pool, generate Python code to process the file, and return results based on the received instructions:

Note

You can use the Code Interpreter (Python Container Apps session) operations outside an agent action or in non-agent workflows in a Standard logic app. However, you must provide your own Python code for the Execute Python code action. In these cases, the Code Interpreter operations don't have access to agent capabilities provided by the agent-linked AI model, which include natural language interactions and model-generated code.

Your agent workflow can process files that contain diverse or fragmented data and have formats such as CSV, Excel, or JSON, which can include thousands to millions of rows. However, raw data often needs extra work before you can extract useful information and insights, for example:

  • Cleaning and transformation
  • Custom logic to extract insights
  • Visualizations or summaries that make the data actionable

These tasks often require manually performed steps that are error-prone and time consuming, especially for people who don't have data science or engineering expertise. Your workflow can use the Code Interpreter (Python Container Apps session) operations to automate these tasks and bring code interpreter capabilities natively and directly to the Azure Logic Apps runtime. The operation that executes Python code is powered by Azure Container Apps session pool. Dynamic sessions in Azure Container Apps provide fast and scalable access to a code interpreter. Each code interpreter session is fully isolated by a Hyper-V boundary and is designed to run untrusted code. When you enable network isolation on your app container, your data never leaves the defined network boundaries.

These capabilities make scenarios like the following examples possible:

Scenario Use case
Data exploration Business users can ask questions about the data using natural language and without writing a single line of code or manually manipulating spreadsheets and use natural language, for example:

- "Find the top 5 products by revenue."
- "Forecast demand for the next quarter by region."
- "Highlight customer segments based on purchase patterns."
Operations Clean large log files, surface exceptions, and generate insights to improve reliability.
Finance Analyze expense reports, detect anomalies, or generate quarterly breakdowns from Excel exports.
Sales and marketing Upload raw sales data and get on-demand summaries, forecasts, or regional comparisons.

This article shows how to set up and use various Code Interpreter (Python Container Apps session) operations in your agent workflow.

Prerequisites

  • An Azure account and subscription. If you don't have an Azure subscription, sign up for a free Azure account.

  • A container app and a code interpreter session pool in Azure Container Apps.

  • A Standard logic app resource and an agent workflow. Make sure that you set up a connection between the Default Agent action and an AI model to use for code generation.

    To create this workflow, see Create workflows that use AI agents and models to complete tasks in Azure Logic Apps.

    Note

    Agent workflows must start with the Request trigger. The Code Interpreter (Python Container Apps session) connector provides actions, but not triggers.

Connector technical reference

For technical information about this connector's operations, see the connector's reference documentation.

Upload a file

  1. In the Azure portal, open your Standard logic app resource.

  2. On the resource sidebar, under Workflows, select Workflows, and then select your agent workflow.

  3. On the workflow sidebar, under Tools, select the designer to open the workflow.

  4. In the Default Agent action, follow the general steps to create a tool and add the Code Interpreter (Python Container Apps session) action named Upload file.

  5. If you're prompted to create a connection, on the Create connection pane, provide the connection information for the session pool.

  6. In the action information pane, expand Input files to upload, and provide the following values:

    Parameter Value Description
    Input file name <file-name> The name to assign for the file.
    Input file content <file-content> The path for the file content to upload.

    In the following example, the workflow uses an HTTP action, which makes a REST call to get the file from its source location. The agent action has a tool that uses the Upload file action with the file name and content as parameters. The file name is manual input, while the file content references output from the preceding HTTP action:

    Screenshot shows agent workflow with HTTP action that gets a file and Upload file action that adds file to session pool.

  7. If you have more files to upload, select Add new item.

  8. When you're done, save your workflow. On the designer toolbar, select Save.

Run Python code

  1. In the Azure portal, open your Standard logic app resource.

  2. On the resource sidebar, under Workflows, select Workflows, and then select your agent workflow.

  3. On the workflow sidebar, under Tools, select the designer to open the workflow.

  4. For the Default Agent action, rename the agent for your scenario, and follow these steps:

    1. Based on whether you created a tool with the Upload file action from the preceding section, choose either option:

    2. If you're prompted to create a connection, on the Create connection pane, provide the connection information for the session pool.

    3. On the new tool, follow the general steps to create an agent parameter for model-generated outputs.

      This agent parameter passes the Python code generated from the agent-linked AI model at runtime to the Execute Python code action.

      After you're done, the Execute Python code action now shows the following code reference in the Python code box: @{agentParameters('python_code')}

      Screenshot shows agent workflow and code reference inside action for Execute Python Code.

    4. For the Execute Python code action to use the file from the Upload file action, set the Session ID parameter value to the session ID for the Upload file action by following these steps:

      1. In the Execute Python code action, from the Advanced parameters list, select Session ID to add this parameter to the action.

      2. Select inside the Session ID parameter, which shows input options, and select the dynamic content option (lightning icon).

      3. From the dynamic content list, under Upload file, select Session ID.

      The following example shows the Session ID parameter set to the session ID for the Upload file action:

      Screenshot shows advanced parameters and session ID parameter for Execute Python Code action.

    5. Rename the tool to describe its purpose or task.

  5. When you're done, save your workflow. On the designer toolbar, select Save.

Connection information

The following table describes the information required to connect to your session in Azure Container Apps:

Parameter Required Value Description
Connection Name Yes <connection-name> A name to identify the connection.
Connection Type Yes Managed identity The authentication to use for the connection.

Note: To use Managed identity authentication, you must set up the managed identity on your Standard logic app with access to the resources you want. On the session pool resource, make sure to set up the managed identity with the role named Azure ContainerApps Session Executor. For more information, see Authenticate access and connections to Azure resources with managed identities in Azure Logic Apps.
Subscription Yes <Azure-subscription> The Azure subscription to use.
Session Pool Yes <session-pool> The session pool in Azure Container Apps, for example, fabrikam-session-pool (/fabrikam-session-pool).

Related content

  • Last updated on