Skip to main content

​
UnsafeTopic

Defines categories of potentially unsafe or sensitive content topics that can be detected and filtered. Used by safety systems to identify and control access to different types of sensitive content.
  • Politics: Politics and political discussions.
  • SubstanceUse: Content related to substance use and drugs.
  • Religion: Religious content and discussions.
  • UnconditionalSelfHarm: Content related to self-harm, suicide, or self-injury.
  • UnconditionalSexualMinors: Inappropriate content involving minors.
  • UnconditionalHateGroup: Content promoting hate groups or extremist organizations.
  • AdultSexual: Adult sexual content.
  • Violence: Violent content, descriptions of harm, or threatening language.
  • Alcohol: Content related to alcohol consumption.
  • Profanity: Profane language, swear words, and obscenities.

​
ModelType

Specifies the type of AI model deployment to use for processing. Determines whether to use cloud-based services or local computation resources.
  • Remote: Use remote cloud-based AI services for processing.
  • LocalGPU: Use local GPU acceleration for AI model processing.
  • LocalCPU: Use local CPU-only processing for AI models.

​
DeviceType

Specifies the type of computing device to use for local AI model processing. Affects performance and compatibility based on available hardware.
  • CPU: Use CPU for processing (compatible with all systems but slower).
  • CUDA: Use NVIDIA CUDA-compatible GPU for processing (requires NVIDIA GPU).
  • Metal: Use Apple Metal for GPU acceleration (macOS/iOS only).

​
Role

Defines the role of a participant in a conversation or interaction. Used to identify who is speaking or providing input in multi-party conversations.
  • User: Human user providing input or asking questions.
  • Assistant: AI assistant providing responses and assistance.
  • Tool: Automated tool or function providing specific capabilities.
  • System: System-level messages for configuration and control.
  • Developer: Developer or administrator with special privileges.

​
ResponseFormat

  • Text: Text response format.
  • Json: JSON response format.
  • JsonSchema: JSON schema response format.

​
StatusCode

Defines standard status codes for operations and API responses. Based on gRPC status codes, used to indicate the result of operations within the Inworld framework.
  • Ok: The operation completed successfully.
  • Cancelled: The operation was cancelled, typically by the caller.
  • Unknown: An unknown error occurred.
  • InvalidArgument: The client specified an invalid argument.
  • DeadlineExceeded: The deadline expired before the operation could complete.
  • NotFound: The requested entity was not found.
  • AlreadyExists: The entity already exists and cannot be created.
  • PermissionDenied: The caller does not have permission to execute the specified operation.
  • ResourceExhausted: Some resource has been exhausted (e.g., quota, disk space).
  • FailedPrecondition: The operation was rejected because the system is not in a state required for execution.
  • Aborted: The operation was aborted, typically due to a concurrency issue.
  • OutOfRange: The operation was attempted past the valid range.
  • Unimplemented: The operation is not implemented or supported.
  • Internal: An internal error occurred.
  • Unavailable: The service is currently unavailable.
  • DataLoss: Unrecoverable data loss or corruption occurred.
  • Unauthenticated: The request does not have valid authentication credentials.
  • Others: Other unspecified error conditions.

​
ResponseType

Defines the types of responses that can be associated with a goal.
  • Instruction: Response that provides instructions or guidance.
  • Verbatim: Response that should be delivered exactly as written.