I switched from openai to google models in my voice agent and now it seems liek it's not getting the system message.

Thanks for reaching out, Tigran. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

No error, we pass the system prompt by appending to _chat_ctx_ in before_llm_callback, no other changes

Thank you so much for reaching out! I totally get how tricky system prompts and chat contexts can be. Let me break down how they work with Google models in LiveKit's framework in a way that hopefully makes things crystal clear: • System prompts are super easy to set up - you just append them to the

chat_ctx

parameter when you're initializing a

VoicePipelineAgent

. Here's a quick example that might help:

initial_ctx = llm.ChatContext().append(
    role="system",
    text="You are a helpful voice AI assistant."
)
agent = VoicePipelineAgent(
    vad=silero.VAD.load(),
    stt=deepgram.STT(),
    llm=google.LLM(),
    tts=cartesia.TTS(),
    chat_ctx=initial_ctx,
)

• Wanna modify your chat context before it hits the Google LLM? The

before_llm_cb

callback is your friend! Check out this neat trick for keeping your context tidy by trimming it down to the last 15 messages:

async def truncate_context(assistant: VoicePipelineAgent, chat_ctx: llm.ChatContext):
    if len(chat_ctx.messages) > 15:
        chat_ctx.messages = chat_ctx.messages[-15:]

agent = VoicePipelineAgent(
    ...
    llm=google.LLM(),
    before_llm_cb=truncate_context,
)

• Heads up - in version 1.0 of the framework,

before_llm_cb

got an upgrade to

llm_node

, which gives you even more flexibility for tweaking your chat context. If you're already doing everything right but still not seeing the results you expect, here are a few things to double-check: • Make sure your system message is formatted just right when you're adding it to

chat_ctx

. • Take a peek at your

before_llm_cb

or

llm_node

(if you're on version 1.0) to make sure it's not doing anything sneaky with your context. • Verify that the Google LLM is playing nice with your

chat_ctx

parameter. I'm totally here if you need any more help or want to chat through this! Don't hesitate to reach out. 😊 Sources: Agents 0.x migration guide | LiveKit Docs | VoicePipelineAgent | LiveKit Docs

What is AFC why does it say AFC enabled

Hey there! I totally get that you're looking for more clarity about this "AFC enabled" situation. Would you mind sharing a bit more about where you're seeing this message? • Are you spotting it in a specific area of the LiveKit platform or maybe in some logs? • Is this popping up when you're working with a particular feature or trying something out? The more details you can give me, the better I can help you figure this out. I'm here to help and want to make sure I understand exactly what you're experiencing!