role

This is optional and defaults to "assistant".
When role=assistant, content is said out loud.
When role=system, content is passed to the model in a system message. Example: system: default one assistant: user: assistant: user: assistant: user: assistant: tool called tool: your server response <--- system prompt as hint ---> model generates response which is spoken This is useful when you want to provide a hint to the model about what to say next.