Tracing

AgentTrace

A trace that can be exported to JSON or printed to the console.

Main trace object containing execution spans and final output.

Field	Type	Description
`spans`	`list[AgentSpan]`	A list of [`AgentSpan`][AgentSpan] that form the trace.
`final_output`	`str \| dict[str, Any] \| BaseModel \| None`	Contains the final output message returned by the agent.

Add an AgentSpan to the trace and clear the tokens_and_cost cache if present.

def add_span(
    self,
    span: AgentSpan | Span,
) -> None

Add a list of AgentSpans to the trace and clear the tokens_and_cost cache if present.

def add_spans(
    self,
    spans: list[AgentSpan],
) -> None

Convert spans to standard message format.

def spans_to_messages(
    self,
) -> list[AgentMessage]

Serializable representation of an OpenTelemetry span.

Check whether this span is an agent invocation (the very first span).

def is_agent_invocation(
    self,
) -> bool

Check whether this span is a call to an LLM.

def is_llm_call(
    self,
) -> bool

Check whether this span is an execution of a tool.

def is_tool_execution(
    self,
) -> bool

Extract input messages from an LLM call span.

def get_input_messages(
    self,
) -> list[AgentMessage] | None

Extract output content from an LLM call or tool execution span.

def get_output_content(
    self,
) -> str | None

Field	Type	Description
`input_cost`	`float`
`output_cost`	`float`	Cost associated to the output tokens.

Properties: total_cost - float: Total cost (input + output).

Field	Type	Description
`input_tokens`	`int`	Number of input tokens.
`output_tokens`	`int`	Number of output tokens.

Properties: total_tokens - int: Total tokens (input + output).

Constants for accessing span attributes following OpenTelemetry semantic conventions for generative AI systems.