Reference for Generative AI Controller

  • Release version: Yokohama
  • Updated January 30, 2025
  • 2 minutes to read
    • Summarize
    Summarized using AI
    This content was generated using new OpenAI-powered functionality. Results are provided on an as is basis and are not guaranteed to be accurate or complete.

    Summary of Generative AI Controller

    The Generative AI Controller in the Yokohama release (updated January 30, 2025) provides essential tables, properties, and configurations to manage and utilize generative AI capabilities within ServiceNow. It supports key actions such as summarization, content generation, and prompt handling through Workflow Studio and Virtual Agent Designer integrations.

    Show full answer Show less

    Key Features

    • Capability Management: Tables like sysoneextendcapability, sysoneextendcapabilitydefinition, and sysoneextenddefinitionattribute define generative AI capabilities and their input/output variables used in workflows and subflows. Variable names are locked when a capability is active to maintain consistency.
    • Builder Integration: The sysoneextendbuilderconfig and sysoneextendbuildercapability tables map capabilities and providers to components used in Workflow Studio and Virtual Agent Designer, enabling seamless AI-driven automation design.
    • Usage Tracking and Logging: The sysoneextendusage table tracks each instance of capability use across workflows, scripts, and agents. Logs such as sysgenailogmetadata capture request details, errors, user feedback, and support audit retention beyond six months when configured.
    • Performance Metrics: The sysgenerativeaimetric table records metrics like edit scores and guardrail activities to evaluate the accuracy and quality of AI model responses.
    • Model Version Mapping: sysgenaimodelversionmapping maintains relationships between AI models, providers, and configuration metadata, facilitating model management and selection.

    Key Properties

    • Provider Selection: com.sn.generative.ai.provider sets the default AI provider if none is specified in the capability definition.
    • Error and Moderation Messages: Customizable messages are provided for AI search failures (com.sn.generative.ai.ais.message) and for content flagged by moderation tools (com.sn.generative.ai.moderation.message).
    • Logging Control: com.sn.generative.ai.logprompt enables or disables logging of generative AI API calls, supporting compliance and troubleshooting.
    • Token Management: com.glide.one.extend.token.buffer defines the buffer for token counts to prevent exceeding provider API limits when executing AI capabilities.

    Practical Benefits for ServiceNow Customers

    • Enable integration of advanced generative AI functions directly into workflows and virtual agents, enhancing automation and user experience.
    • Maintain control and visibility over AI usage, performance, and compliance through detailed logs and metrics.
    • Configure default providers and messages to tailor AI interactions to organizational policies and user expectations.
    • Manage AI model versions and capabilities effectively to optimize AI-driven processes.

    Reference topics provide information about Generative AI Controller flow actions, tables, and properties.

    Tables installed

    Name Table Description
    OneExtend Capability sys_one_extend_capability Generative AI Controller capabilities that include Summarize, Record Summarization, Generate Content, and Generic Prompt.
    OneExtend Capability Definition sys_one_extend_capability_definition Attribute configuration for input and output variables for Workflow Studio subflows.
    OneExtend Capability Definition Attribute sys_one_extend_definition_attribute Input and output variables for Workflow Studio subflows. Variable names can't be changed if the capability is active and used on the instance. You can check whether a capability is used by going to the OneExtend Usages table.
    OneExtend Builder Config sys_one_extend_builder_config Determines which capability and provider is related to each builder component for Workflow Studio and Virtual Agent Designer.
    OneExtend Builder Capability sys_one_extend_builder_capability Definitions for a capacity and its provider for builder components.
    OneExtend Usage sys_one_extend_usage Each usage of a capability in a Workflow Studio or Virtual Agent Designer topic, as well as any scripts such as business rules or UI actions.
    Gen AI Log Metadatas sys_gen_ai_log_metadata Log data about requests to the LLMs, including information about definition, errors, user, and feedback provided.

    AI-generated content can be tracked for a duration beyond six months with Now Assist configuration option. You can export historical data by writing a script to copy it into a different table without deleting the information.
    Generative AI Metric sys_generative_ai_metric Logs various metrics to help evaluate the performance and accuracy of LLM responses, such as edit score, edit distance, and guardrail activity.
    Gen AI Model Version Mapping sys_gen_ai_model_version_mapping Maintains mappings between AI models, their providers, and configuration details. It helps to show how source models map to target models along with metadata like skill type, model type, resource associations, and provider information.

    Properties

    Table 1. Generative AI Controller properties
    Property Description
    com.sn.generative.ai.provider Default provider when capability definition has no default.

    Type: choice list

    No default value
    com.sn.generative.ai.ais.message Message that is displayed when AI Search fails to find an answer to a query.

    Type: string

    Default value: No answer found.
    com.sn.generative.ai.log_prompt Prompt that determines whether generative AI API calls are logged.

    Type: true | false

    Default value: true
    com.sn.generative.ai.moderation.message Message that is displayed if the OpenAI or Azure OpenAI moderation tools identify the content that goes against their terms of service.

    Type: string

    Default value: The response cannot be displayed as it’s deemed inappropriate by OpenAI.
    com.glide.one.extend.token.buffer Buffer that checks the request for the number of tokens before a OneExtend capability is executed. The maximum allowed request tokens are calculated based on the maximum tokens that are permitted by the AI provider's API minus the response token and buffer value that is specified in this system property.

    Type: integer

    Default value: 250

    External links

    Provider Data policy Usage policy
    Amazon Bedrock Data protection AWS Service Terms
    Aleph Alpha Data privacy Terms and Conditions
    Google Cloud Google Cloud Platform Terms of Service Google Cloud Platform Terms of Service
    IBM watsonx Keeping your data secure and compliant Foundation model terms of use in watsonx.ai
    Microsoft Azure OpenAI Data, privacy, and security for Azure OpenAI Service Code of conduct for Azure OpenAI Service
    OpenAI API data usage policies Usage policies