1. Use streaming endpoint of the LLM of your choice.
  2. Split text stream into stream of sentences. Similar approach is described here.
  3. Send the sentences one by one to Avaturn.Live endpoint.