A deployment is a version of an application running in an particular environment, e.g stage, production, canary.


An operation is any function, event or action that is traced by Graphsignal, e.g. model API call, inference, agent action.


Session groups multiple traces together to represent a sequence of operations, e.g. an AI agent workflow execution.


Scores are numerical values and metadata assigned to events or objects identified by tags, e.g. generations, runs, sessions, or users to evaluate their behavior, importance, performance or impact. For example, scores are useful to analyze and act on negative user interactions.


Metrics show different aspects of execution and data in time. Graphsignal provides performance, system, payload, and cost metrics.


Tracing is the recording of operation execution in context of user request, AI agent run or model inference, with details such as latency, errors, and payload. Tracing also automatically records metrics, such as latency histograms, token usage, etc.


Graphsignal tracer is a module that is integrated into applications and scripts. It automatically traces operation execution for natively supported libraries and exposes tracing API for manual tracing.