hi, is it possible to use LLM+TTS or Realtime OOB responses to play out thinking phrases during long tool calls executions? this would be like running in parallel to the tool call, a bg job that makes this OOB calls every once in a while (can be fixed delays), so we outputs some thinking phrases instead of only playing the thinking sound. i manage to make it work directly with OpenAI Realtime but in Agent implemetation is hidden in generate reply

Interesting, we have an example that shows a similar use case: https://github.com/livekit-examples/python-agents-examples/tree/main/complex-agents/exa-deep-researcher but the implementation is more complex.

mmm pretty interesting, regarding the agent state, during research it transitions back and forth between thinking and speaking or you can continue talking with the agent? for our use case, we don't want the user (elderly people) to be able to keep talking as it may be confusing for them. so we want to keep it in thinking loop but with occasional phrases to notify agent is still working on the request