Passing stream: true to the message.create method will return an async iterator that yields response objects. 
const res = await lira.message.create({
  stream: true,
})

for await (const message of res) {
  console.log(message)
}

​
Response object

No matter the provider used the response object will stay consistent to the following schema 
{
  id?: string
  model?: string
  message: AssistantResponse | ToolUseResponse
  stop_reason?: StopReason
  stop_sequence?: string
  usage?: Usage
  logprobs?: Logprobs
  openai_options?: OpenAIOptions
}
The response object is the same as the one used in the static output. The only difference is that the id and model fields are optional.
This is because the streaming response object is not the final response object. It’s a partial response object that will be updated as the model generates more tokens. So the id and model fields might be generated at any step in the stream.
👉 Static response object