Skip to main content

Enable Data Copilot

Prophecy Data Copilot is an AI-powered assistant that delivers intelligent suggestions and automates repetitive tasks for visual data transformations. You can read more about it at Data Copilot.

Data Copilot leverages OpenAI's generative AI models to understand user intent, and enriched by the organizations' knowledge graph, to automate repetitive data engineering tasks. By default, Data Copilot leverages Prophecy's managed OpenAI subscription and is entirely free for existing Prophecy customers. Prophecy uses user queries and metadata when communicating with OpenAI. Prophecy never sends any customer data to OpenAI.

However, for the most security conscious organizations, it is possible to configure Prophecy to use your own OpenAI endpoint. This page describes how to enable Prophecy Data Copilot for private VPC SaaS environments and configure it to use your own OpenAI or Azure OpenAI endpoint.

Prophecy Data Copilot & OpenAI Flow Architecture

note

Prophecy also supports running Data Copilot from locally hosted private models such as Llama or Mistral. Additionally, you can configure Prophecy Data Copilot to use custom LLM endpoints provided by Anthropic, AWS Bedrock, Google Vertex AI, and Databricks Model Serving.

Installation​

From the Kubernetes cluster where Prophecy services are running:

  1. Check that the prophecy-operator, edweb, metagraph, and other Prophecy services are running version 3.3.8.0 or greater.

    1. The above services should all be running the same release-version as the copilot service defined in the image below, for example 3.3.8.0.

  2. Check that the athena Prophecy service is running version 3.3.8.0 or greater. This is only required if using the pipeline approach to the upgrade.

  3. In the ProphecyCluster CR, do the following:

    • Add copilot to enabledapps.

  4. Once deployed, wait for the copilot-cm to be created. Then edit the default values in the configmap to match the ENVs required.

  5. Supply the API keys for the appropriate endpoint and ENV as below. Prophecy Copilot requires two types of models:

    1. Smart LLM for complex tasks

      • Recommended Model: gpt-4o
      • Supported Models: Any model from the gpt-4 family

    2. Fast LLM for easy and fast tasks

      • Recommended Model: gpt-4o-mini
      • Examples: gpt-4o-mini, gpt-3.5-turbo

    These models ensure optimal performance and efficiency for various tasks within Prophecy Copilot.

    OpenAI

    AI_MODEL_PROVIDERS_CREDS: "{ 'openai' : {'api_key': '___openai_api_key___'} }"
    AVAILABLE_AI_MODELS: "{ 'fast_model': {'provider': 'openai', 'model_name': 'gpt-4o-mini' }, 'smart_model': {'provider': 'openai', 'model_name': 'gpt-4o'} }"
    NUM_WORKERS: "< add value here >"
    PORT: "< add value here >"

    Azure's OpenAI

    AI_MODEL_PROVIDERS_CREDS: "{ 'azure_openai' : {'api_key': '___azure_openai_api_key___', 'api_endpoint': 'https://name.openai.azure.com/'} }"
    AVAILABLE_AI_MODELS: "{ 'fast_model': {'provider': 'azure_openai', 'model_name': 'gpt-4o-mini', 'deployment_name': '___4o_mini_deployment___' }, 'smart_model': {'provider': 'azure_openai', 'model_name': 'gpt-4o', 'deployment_name': '___4o_deployment___'} }"
    NUM_WORKERS: "< add value here >"
    PORT: "< add value here >"

  6. Enable the following variable on the common CP config map:

    COPILOT_ENABLED: "true"

  7. If the app pod, especially copilot, doesn't redeploy itself, restart the app pod.

note

To learn more about how to connect Prophecy Data Copilot to private OpenAI endpoints, speak to your Prophecy account owner.

How much will I pay for Prophecy Copilot?​

Copilot features are included with Prophecy’s Spark and SQL offerings. There would be an additional cost if you chose to use a private subscription LLM service. In this case, the cost depends on the following:

  • Number of word tokens sent to the LLM provider
  • Size of the datasets
  • Number of iterations Prophecy sends requests to the LLM provider