Input Guardrails for Realtime SDK

[Skyline-9]

Please read this first

• Have you read the docs?Agents SDK docs

Yes

• Have you searched for related issues? Others may have had similar requests

Describe the feature

What is the feature you're requesting? How would it work? Please provide examples and details if possible.

Add input guardrails to the realtime API, bringing it closer to parity with the non-realtime Agent/Runner, which already supports input_guardrails. Realtime today only supports output guardrails (RealtimeAgent.output_guardrails / RealtimeRunConfig["output_guardrails"]); there is no first-class way to screen the user's transcribed input.

#4048

[seratch]

Thanks for filing this. Realtime input validation cannot be added as direct parity with the regular Runner because completed transcription is asynchronous with Response creation, so post-hoc cancellation cannot reliably provide pre-response blocking.

Strict gating is already supported by sending session.update with turn_detection.create_response=false and turn_detection.interrupt_response=false, validating the completed transcript, and then sending response.create. A first-class input guardrail would either duplicate this workflow or expose weaker semantics under the same API name, so I am going to close this issue.