Custom LLM/TTS/STT Integration with Vapi
# support
s
I'm looking to build a voice agent over telephony using Vapi. Currently, the telephony is handled via Twilio. I'm planning to use the TypeScript SDK to build the server. I want to know if it's possible to use custom LLM, TTS, or STT models instead of the default options listed in Vapi's documentation. If so, how can I integrate those custom models into the workflow? Thanks
k
You can easily implement custom LLM, STT, and TTS using the API calling the /assistant endpoint. Here is a link with example usage:
s
Thanks.
Is it possible to use just the Vapi TTS functionality? Our team really loved Neha's voice and we'd like to integrate it into our existing workflow.
k
We are an orchestration platform, and we don't support exclusive services to use our TTS. You have to use our AIA systems.
AI Assistants\*
s
Got it, Thanks
k
Please reach out to us if you have any other questions
s
Yeah
I was wondering if there’s an example repo or tutorial available that demonstrates how to handle inbound and outbound calls using the TypeScript server SDK and Twilio.
According to the documentation, I should be able to make an outbound call to any number using Twilio as the provider through the server SDK (TS) with
client.calls.create
, correct? What I’m unsure about, though, is how to handle an inbound call.
I also wanted to ask how I can obtain an Indian phone number. Since Twilio doesn’t support Indian numbers, do you support integration with Plivo or any other provider that offers Indian numbers?
k
Hey, You will need to get in touch with ([huzefa@vivadigitally.com](huzefa@vivadigitally.com)) [vivadigitally.com](http://vivadigitally.com), or any other telephony provider that offers a SIP trunk. They will be able to help with the number, and then you can read this guide about integrating it. Note: This guide uses the Twilio SIP trunk as an example, but the procedure is pretty much the same for other telephony providers as well.
s
Okay understood.
Could you guide me on how to handle inbound calls using my server?
k
Hey, can you please clarify what you mean by "to handle inbound calls using your server"? I couldn't understand.
s
Hi, I'm receiving the tool-calls webhook correctly on my server and sending back a properly formatted response with HTTP 200, as specified in the documentation. However, I’m not seeing any response content reflected in the logs—only that a tool call was triggered, but not the actual result from the response. I’ve also observed this behavior while interacting with the agent. For example, when I say, "I'm looking for a 3BHK flat in Bangalore," the tool gets invoked, and the model replies with, "I got a few options," but doesn't provide any further details—which should have come from the tool's response.
k
Hey! To help track down this issue, could you share: - The call ID - When exactly this happened (the timestamp) This would really help us figure out what went wrong!
s
- c8d2ed5e-83e7-43ef-9c06-1e195ecbf95f - 12:53:19:035
k
Hey, can you set async to false?