Configuration Reference

This page contains the list of all available Airflow configurations for the apache-airflow-providers-common-ai provider that can be set in the airflow.cfg file or using environment variables.

Note

For more information see Setting Configuration Options.

[common.ai]

Options for the apache-airflow-providers-common-ai provider.

capture_content

Added in version 0.4.0.

Capture prompt, completion, and tool-call content on the emitted GenAI spans (gen_ai.input.messages / gen_ai.output.messages).

Off by default: only token counts, model id, latency, tool names, and finish reason are recorded, never message text. Turning this on exports model inputs and outputs to your tracing backend without redaction. Airflow’s secret masking applies to logs and rendered template fields, not to OpenTelemetry span attributes, so it does not scrub this content. Enable it only for debugging in a trusted environment. Has no effect unless otel_export_enabled is True.

Type:

boolean

Default:

False

Environment Variable:

AIRFLOW__COMMON_AI__CAPTURE_CONTENT

Example:

False

durable_cache_path

Added in version 0.1.0.

ObjectStorage URI used to persist per-step caches when running AgentOperator / @task.agent with durable=True. Each task execution writes a single JSON file under this path containing its cached model responses and tool results, so that on retry the agent can replay completed steps instead of re-issuing LLM calls and tool invocations. The file is deleted on successful task completion.

Required when durable=True is used. Any scheme supported by airflow.sdk.ObjectStoragePath is accepted (file://, s3://, gs://, azure://, …).

Type:

string

Default:

''

Environment Variable:

AIRFLOW__COMMON_AI__DURABLE_CACHE_PATH

Example:

file:///tmp/airflow_durable_cache

otel_export_enabled

Added in version 0.4.0.

Attach pydantic-ai OpenTelemetry instrumentation to agents created by this provider and emit GenAI spans (agent run, model call, tool call, token usage) for AgentOperator / @task.agent / @task.llm and the other LLM operators.

Spans are emitted through Airflow’s existing OpenTelemetry exporter, configured under [traces] / the standard OTEL_EXPORTER_OTLP_* environment variables, and nest under the task span so they are attributable to the originating DAG run and task instance. The provider does not configure an exporter of its own: if core tracing ([traces] otel_on) is not enabled in the worker process, no spans are emitted. Off by default so installing the provider never starts shipping spans without opt-in.

Type:

boolean

Default:

False

Environment Variable:

AIRFLOW__COMMON_AI__OTEL_EXPORT_ENABLED

Example:

True

Was this entry helpful?