Core concepts

Here are some core concepts and definitions to help you understand how Trubrics works:

• Events: any action within an application that is performed by a user, or an AI assistant. Examples may include user prompts, user button clicks, LLM generations, AI agent database queries, etc. Events can be collected from the backend or frontend of your app.

• Properties: information that is tracked as part of an event. An example of a property for an LLM generation event is the number of tokens generated. Properties are completely custom, allowing analysis of data that is relevant to your product. Some properties are recognised by Trubrics.

• Text attributes: when the property text is set in an event, Trubrics automatically extracts attributes from the text that can be used for analysis. Default text attributes extracted by Trubrics include sentiment and topic extraction. Custom text attributes (e.g. AI generation quality / evaluation metrics) can be configured to be extracted from any events.

• Users: a human that is performing actions within your app.

• Assistants: an AI assistant that is performing actions within your app.

• Threads: any event associated to an interaction between a user and an assistant (a conversation between an assistant and a user, for example).

• Sessions: a user session, that can contain many events, and any number of threads with the same or different AI assistants. Two events are considered to be in the same session if they are within 30 minutes of each other.

• Generated insights: insights that are automatically generated by Trubrics from your event data. These may be sent daily or weekly, directly to your inbox or slack channel.