Use Copilot

This tutorial provides step-by-step instructions for using Copilot within a Snoweaver project.

Prerequisites

Before continuing with this tutorial, ensure you have completed the following prerequisites:

Note

For detailed information on the models supported by Cortex’s COMPLETE function in each region, please refer to this link: Snowflake Cortex availability

Enable Copilot

  1. Access the Admin Console with the owner role of Snoweaver or a custom role with the SNOWEAVER.APP_ADMIN application role. The role must also have the necessary privileges to grant Snoweaver Import Privileges on the Snowflake database.

  2. On the Home page, enable Copilot and click the Grant Privileges button if prompted. This is required because Copilot needs access to the Cortex language models in the Snowflake database.

  3. Configure Copilot with the following settings:

    Allowed Models:

    llama3.1-8b
    llama3.1-405b
    mistral-large2
    

    LLM Settings:

    {"temperature": 0}
    
    ../_images/24.png
  4. Save the configuration.

Use Copilot within a job

  1. Open the LEARN_DEV project in Snoweaver with the LEARN_DEV_DEVELOPER role.

  2. Open sfdc_upsert_object_record on the Jobs page.

  3. Update the Job Endpoint template to the following specification:

    https://snoweaver3-dev-ed.develop.my.salesforce.com/services/data/v61.0/sobjects/Account/customExtIdField__c/11999
    
  4. Open Copilot and select llama3.1-8b from the options

  5. Enter the prompt below and review the generated response.

    parameterize the template with _proj and _vars
    
    ../_images/27.png
  6. Test different language models and evaluate the responses they produce.

  7. Click Undo to revert the change.

  8. Go to the Response Handler section.

  9. Under the Preview Request button, click Open Copilot, then choose llama3.1-405b from the options.

  10. Enter the prompt below and review the generated response.

    What other properties can be updated using this request?
    
    ../_images/47.png
  11. Test different language models and evaluate the responses they produce.