What is the most deterministic way to ask questions and gather information in the VoicePipelineAgent

To deterministically ask questions and gather information using the VoicePipelineAgent, you can utilize the

before_llm_cb

callback to modify the chat context before it is sent to the LLM. This allows you to control the flow of conversation by dynamically adjusting the context based on the user's responses. You can also use the

before_tts_cb

callback to alter the text before it is sent to the TTS model, ensuring that the agent's responses are clear and contextually appropriate. For example, you can implement a structured workflow by setting up a sequence of questions and using the

before_llm_cb

to check the user's responses and decide the next question or action. This approach allows you to maintain a clear and controlled conversation flow, ensuring that the agent gathers the necessary information in a structured manner. Would you like more details on implementing this approach? Sources: - VoicePipelineAgent 👍 0 button 👎 0 button 💡 Hint: Mention @runllm in the thread for followups.

can we use this method to integrate temporal.io workflow @millions-hair-24956

It seems likely that integrating Temporal workflows with the VoicePipelineAgent could be feasible by leveraging the

before_llm_cb

callback to manage state transitions and workflow logic. Temporal is designed for orchestrating complex workflows, and you could potentially use it to manage the sequence of interactions within the VoicePipelineAgent by invoking Temporal workflows at specific points in the conversation. To achieve this, you would need to set up a Temporal client within your LiveKit agent environment and use the

before_llm_cb

to trigger Temporal workflows based on the current state of the conversation. This setup would allow you to handle complex state management and ensure that the conversation follows a predefined workflow managed by Temporal. Would you like more detailed guidance on how to set this up? Sources: - https://livekit-users.slack.com/archives/C07FY8WHGPM/p1733314962.037489 👍 0 button 👎 0 button 💡 Hint: Mention @runllm in the thread for followups.