Vapi is the Voice AI platform for developers.

Vapi AI

image

im building a dual-input chatbot using the websdk (chat + audio) . the chat modality would be useful for typing in inputs like email ids or copy pasting text  etc..


my question is referencing the attached image: I send "my name is sangy" via vapi.send(). 
So why could the bot be mis-reading my name from the transcript when there's no transcription involved?

Hi <@939930793755676753>, I'm an AI assistant for Vapi. While a team member reviews this question, I'll look through relevant sources and see if I can provide suggestions. Please tag <@1145576664366055464> with any follow-up questions.

-----------------------------
**While I generate an answer, here are some sources that might be helpful**:
* [Vapi Web SDK](<https://github.com/VapiAI/web/blob/main/README.md>)
* [Persistent Assistants](<https://docs.vapi.ai/persistent_assistants>)
* [Set up your Server URL](<https://docs.vapi.ai/server_url>)
* [Swagger](<https://docs.vapi.ai/api-reference/swagger>)
-----------------------------

### Answer
When you use `vapi.send()` to input text directly into the chat, the message is processed without needing speech-to-text transcription. If there's an issue with the bot misreading the input from `vapi.send()`, it might not be related to transcription errors but rather how the text input is being handled or processed within your chatbot's logic or the LLM's interpretation of the text. Ensure your assistant's configuration and handling logic for text-based inputs are correctly set up to interpret and respond to these inputs accurately [(1)](<https://github.com/VapiAI/web/blob/main/README.md>) .

Got it. Tagging <@&1211491065807507517> for additional help.

there's no additional logic apart from sending and receiving messages to vapi. 
i was able to reproduce this a few times as well.

call ids: 
8d42abb0-4aca-449d-aaed-c454e6d9f97c
5f2877a2-1c4f-4f1f-b556-e33081f4acde

@nikhil but all the messages in the chat were typed and sent via vapi.send(). 
there was no transcription involved?

<@&1211491065807507517> OH confirmed this was a bug and offered to look deeper. So following up here

yes we're working on it. could you follow up thursday?

<@1192736571812753462> followin up on this

nikhil let us have a chatbot too, to debug and stuff, and see what the llm is doing. Calling to test it all is just not fast or clear.

<@&1211491065807507517> Following up from today's OH with <@1192736571812753462>  and last weeks OH with <@760495257375866940> 

latest update from <@760495257375866940> was that this *"the text was still being processed, which caused the issue. We have an upcoming fix which will fix the issue."
*

Can you try again <@760495257375866940> just pushed an update yesterday.

<@&1211491065807507517>  This issue is still not fixed. 
<@760495257375866940> please check call id 984cc2f6-52dd-4b97-a304-0e10b1f630ba

Transcript
AI: Hi there. I'm Jen, a demo AI assistant. May I know your name?
User: sangy
AI: Nice to meet you, Sammy. This is a practice call.

Why can the bot not know the name just from message history without messing with transcribers? 

isn't this core issue for all users?

Can you bring it up in today's Vapi Office Hour?