https://livekit.io logo
Join Slack
Powered by
How can I get customer's last message in "before_l...
# ask-ai
a
How can I get customer's last message in "before_llm_cb". Give me a simple function example where i can get last user's message for which LLM's response is expected
t
Thanks for reaching out, Hiren. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.
Hi there! To help you capture the customer's last message in the 
before_llm_cb
callback, here’s a friendly example function you can use:
Copy code
async def before_llm_cb(assistant: VoicePipelineAgent, chat_ctx: llm.ChatContext):
    # First, let's check if there are any messages at all
    if not chat_ctx.messages:
        print("No messages in chat context.")
        return  # We exit early if there’s nothing to process

    # Now, we’ll look through the messages starting from the most recent to find the last user message
    last_user_message = None
    for msg in reversed(chat_ctx.messages):
        if msg.role == "user":
            last_user_message = msg.content
            break

    if last_user_message:
        print(f"Last customer message: {last_user_message}")
        # Here you can also tweak the chat context or do other useful things
    else:
        print("No user messages found in chat context.")
This function is designed thoughtfully to make sure you safely access the last user message by: • First checking if 
chat_ctx.messages
actually contains any messages. • Going backward from the newest message to find the latest one from the user. • Gracefully handling situations where no user message is present. When you set up your 
VoicePipelineAgent
, you can register this callback like this:
Copy code
agent = VoicePipelineAgent(
    vad=silero.VAD.load(),
    stt=deepgram.STT(model="nova-2-general"),
    llm=openai.LLM(),
    tts=cartesia.TTS(),
    chat_ctx=initial_ctx,
    before_llm_cb=before_llm_cb,
)
This way, you’ll have a reliable and smooth way to manage your conversation history dynamically. If you have any questions or need further help, feel free to ask—I’m here to support you! Sources: chain-of-thought-tts/README.mdoc
a
like I want to control, is this required to go to LLM to get the response.. OR stop it here itself.. like if user's message is "hello" then I dont want to generate the response
t
Let me guide you through how you can control whether a message gets sent to the LLM for a response in the 
before_llm_cb
callback. You have a few options when returning a value: • 
False
to skip processing the message altogether. • 
None
to continue with the normal processing flow. • 
True
to make sure the message is processed. Here’s a simple example function that stops processing if the user’s message is "hello":
Copy code
async def before_llm_cb(assistant: VoicePipelineAgent, chat_ctx: llm.ChatContext):
    # First, check if there are any messages
    if not chat_ctx.messages:
        return False  # Skip if no messages

    # Find the last user message
    last_user_message = None
    for msg in reversed(chat_ctx.messages):
        if msg.role == "user":
            last_user_message = msg.content
            break

    # If the last message is "hello", stop processing
    if last_user_message and last_user_message.lower() == "hello":
        print("Skipping LLM response for 'hello'")
        return False

    return None  # Otherwise, continue normal processing
When you’re setting up your 
VoicePipelineAgent
, you can register this callback like this:
Copy code
agent = VoicePipelineAgent(
    vad=silero.VAD.load(),
    stt=deepgram.STT(model="nova-2-general"),
    llm=openai.LLM(),
    tts=cartesia.TTS(),
    chat_ctx=initial_ctx,
    before_llm_cb=before_llm_cb,
)
This way, you have a handy way to filter out specific messages or add your own custom logic before anything gets sent to the LLM. If you have any questions or want to explore more, feel free to ask—I’m here to help! Sources: chain-of-thought-tts/README.mdoc | Vision Agent Quickstart | LiveKit Docs | VoicePipelineAgent | LiveKit Docs | Agents 0.x migration guide | LiveKit Docs
Open in Slack
PreviousNext
Powered by