AiIntegration

Create a new feature instance.

Determine whether AI tracing is enabled and the optional package is present.

Register the feature in the environment.

Register only the enabled AI listener families.

Set up the feature in the environment in an inactive state (when no DSN was set).

Initialize the feature.

Initialize the feature in an inactive state (when no DSN was set).

Retrieve the user configuration.

Determine if default PII should be sent.

Determine whether AI inputs may be collected.

Determine whether AI outputs may be collected.

Indicate if the given feature is enabled for tracing.

Indicate if the given feature is enabled for breadcrumbs.

Determine if the active SDK client can record spans.

Determine if the active SDK client can record breadcrumbs.

Start an invocation under the current sampled span.

Complete an invocation with response metadata.

Close a failed invocation after failed-turn persistence.

Close the failed attempt before another provider reuses the invocation ID.

Close an invocation whose consumer stopped before completion.

Capture resolved metadata and start the exact model step.

Record the model response and finish its step span.

Finish a failed model step, including failures during stream consumption.

Start a tool span without resolving its schema again.

Record a tool result and restore its caller's span.

Close a tool span when execution throws or is canceled.

Start an embeddings span under the current sampled span.

Complete embeddings with the returned model and input-token count.

Close a failed embeddings attempt by its exact invocation ID.

Method used so that other integrations that produce gen_ai spans can submit their spans here which will be populated with the conversation ID once it becomes available.

Set the conversation of the gen_ai spans started from now on in the current trace, replacing the current one.

Passing null forgets the current conversation.