Custom LLM + Dial Digit?
# support
h
Hi there I wanted to know if it is possible to use a custom LLM that is somehow able to access the inbuilt dial keypad function.
v
You can continue this conversation by mentioning me in the message.
@Hugo Absolutely, you can integrate your custom LLM with Vapi to access the inbuilt dial keypad function. You would need to ensure that your LLM can handle the necessary API calls to interact with the dial keypad feature. If you need guidance on how to set this up or have any specific requirements, please let me know, and I'll be happy to assist you further!
h
Yes. What is the necessary api call to interact with the dial keyoad feature
@User Can someone give me an answer to this
s
@Hugo Yes, you may utilize the custom LLM and dial keypad function; however, please note that the dial keypad function has exhibited some instability. I encourage you to give it a try to determine if it meets your needs. Should you require any further assistance, please do not hesitate to let me know.
h
Hi @Shubham Bajaj thank you for the response. Glad to hear this is possible. the only question I really have is how would this work since with a custom LLM the function would be performed or when it is a dial keypad function do I send that response back to vapi?
s
@Hugo With a custom LLM, you are not required to manage the function on your server; instead, you can yield the response, allowing Vapi to handle the function on its end. For your reference, you may review this thread https://discord.com/channels/1211482211119796234/1328028159949537341/1329207905643270154
@Hugo Do let me know if further help is required.
h
Yes I got it to work which is huge already. Now you mentioned that is "exhibited some instability" now if one of my systems requires IVR navigation frequently say 60% of conversations and I want to push this system to prod would the function be reliable for this orrr.
@Shubham Bajaj
Any chance that I can pay for your hour @Shubham Bajaj so that you can help me with a few questions I have with this setup.
If not I'll quickly explain the problem essentialyl I got this working fine. Not too difficult however obviously with a custom LLM I might want a few more functions which need to be handled on server. Now I want to be able to differentiate these say during the stream if function_call = "dtmf" we stream to vapi if not system does its thing and then streams the content response. But I STRUGGLING TO ACCOMPLISH THIS. 😥
s
You don't need pay, I am here to help but I have to cater all so response time varries.
You have to do the IVR navigation by yourself first and then connect the calls with Vapi. I can share the code for IVR navigation just let me know.
@Hugo can you share your custom LLM code in an github repo so I can take a look?
a
This would be super helpful if you could share this
h
Yh if you could share that
g
Was it shared? @Shubham Bajaj
a
Never got shared with me
g
@Shubham Bajaj
s
@Hugo @adomic_ @gzqq Check this documentation on how to use inbuilt functions or inbuilt tools of WAPI with your custom LLM. https://docs.vapi.ai/customization/tool-calling-integration#b-vapi-attached-tools
v
I was out of office for a few days and was unable to respond to your message earlier. I apologize for any inconvenience this may have caused. 
4 Views