Content Generation

• Connection ID - The connection established in Integration Service. Access the dropdown menu to choose, add, or manage connections.

• Model name - The generative AI model or ID to use. This activity defaults to the model with the highest quality observed output. However, you can select a different model based on desired outputs and testing. Switching models can impact the output.
• Prompt - The user prompt for the chat completion request. This field supports String type input.
  Note: To attach an image to a custom prompt, use the Image Analysis activity.
• PII detection - Whether to detect PII from the input prompt. Boolean value. Default value is False.
  • PII filtering - If set to True, any detected PII/PHI is masked before sending to the LLM. The quality of the output may be impacted. If set to False, the detected PII is included in the prompt. In both cases, the detected PII is available in the output. This field is displayed if PII detection is set to True.
  • PII language - The language of the prompt input and output to scan for PII. Select a language from the available dropdown list. This field is displayed if PII detection is set to True.
  • PII/PHI category - The optional PII/PHI category or categories to analyze for. If not set, all categories are reviewed. This field is displayed if PII detection is set to True.
• System prompt - The system prompt or context instruction for the chat completion request. This field supports String type input.
• Context grounding (Public Preview) - Insert context into the prompt from an existing index (Orchestrator bucket) or from a file. Select one of the available options from the dropdown menu: None, Existing index, File resource.
  • Index (Public Preview) - The name of the index to reference. This field is displayed if Context grounding is set to Existing index. This field supports String type input.
  • File - Click to use variable. This field supports IResource type input. This field is displayed if Context grounding is set to File resource.
    Note: This field has a 30MB file size limit. For larger files, upload data to Orchestrator and create an index using the Index and Ingest (Public Preview) activity.

    Currently supported formats: PDF, JSON, CSV.

  • Number of results (Public Preview) - Indicates the number of results to be returned. This field supports Int64 type input.

Manage Properties

Additional properties

• PII confidence score - The minimum confidence score required in order to qualify as PII and be redacted. This field is displayed if PII detection is set to True.
• Maximum tokens count - The maximum number of tokens to generate in the completion. The token count of your prompt plus those from the result/completion cannot exceed the value provided for this field. It's best to set this value to be less than the model's maximum count so as to have some room for the prompt token count. Default value is 1024. If not set, the activity defaults to the necessary tokens to accommodate the request or the maximum tokens allowed by the model. This field supports Int64 type input.
• Temperature - The value of the creativity factor or sampling temperature to use. Higher values mean the model will take more risks. Try 0.9 for more creative responses or completions, or 0 (also called argmax sampling) for ones with a well-defined or more exact answer. The general recommendation is to alter, from the default value, this or the Nucleus Sample value, but not both values. Default value is 0.
• Frequency penalty - Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing frequency in the text, decreasing the model's likelihood to repeat the same line verbatim. Default value is 0.
• Presence penalty - Number between -2.0 and 2.0. Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. Default value is 0.
• Completion choices count - The number of completion choices to generate for the request. The higher the value of this field, the more the number of tokens that will get used. This results in a higher cost, so you need to be aware of that when setting the value of this field. Default value is 1.
• Stop sequence - Up to four sequences where the API will stop generating further tokens. The returned text does not contain the stop sequence. Default value is null.

Output

• Top generated text - The generated text.
• Masked test - The input prompt where any potential PII data has been replaced with masked placeholders.
• Content Generation - This output contains the full nested response object including additional details about the completion, model used, and PII detection results.
• Citations string (Public Preview) - The Context grounding citations string.

The Content Generation activity offers flexibility in how you interact with and insert LLM-generated responses to custom prompts in Studio, Studio Web, or Apps. By inserting arguments and variables into the prompt, you can accomplish a dynamic prompt template that pulls in data from popular business applications via connectors, Orchestrator queue items, etc. This helps you build meaningful, scalable automations that fit unique use cases.

• Existing index: Reference the Index name of an index created using the Index and Ingest (Public Preview) activity. This performs RAG on the dataset within that index. This is the typical use case: querying over multiple documents or files.
• File resource: Use a file uploaded just-in-time to the activity and Context grounding enables a just-in-time or in-memory RAG to query that specific document. You can only upload one document here, good for summarization-based use cases.