@DollarD Hello! I'm delighted to hear that you're enjoying your experience with Vapi. It's great to see users like you making the most of the platform and coming back to explore new features. For your use case, you can achieve the immediate invocation of a custom tool to bring in extra context about the caller by using the

assistantOptions

during the initialization of your Vapi assistant. You can set up a custom tool to be called as part of the assistant's startup sequence. Here's a general idea of how you can configure this: 1. Define a custom tool that fetches the caller's context from your API. 2. Include this custom tool in the

assistantOptions

when you start your Vapi assistant. If you need assistance with the specific code or setup, please provide more details about your custom tool and how you've structured your assistant, and I'll be happy to help you integrate it. If this is something that isn't currently supported directly through the Vapi Dashboard or API, I would recommend reaching out to the support team for more information on potential workarounds or upcoming features that might meet your needs.

@Vapi do you know where in the vapi dashboard if anywhere that i can configure that?

if you can trigger a tool when the call first starts, that would be great if you would tell me where to configure that in the vapi dashboard, instead of giving such attitude 🙂 and yes, if you dont want to help me now, i don't really care that much, as you did not seem like quite the helpful individual in the first place

are you daft mate ?

If you create a propmt to trigger a tool on the call start it would - if you have not tried it then you should.

Why would I give you an idea that purposely does not work? That makes no sense whatsoever.

And if you have a database with the information why are injecting it? Just use it in your prompt. I see you guys doing that and just making it way harder. Dashboard is dynamic whatsoever. Use the API and add that information. Does not matter how many people you are calling.

It sounds like you don't know how vector databases work

😂

the more you talk the more its clear you dont know much

Becuase you didnt clarify and still dont know how VAPi works? i love your thinking and logic.

I can see why you are struggling.

@User I managed to resolve this through some advanced prompting techniques involving some workarounds, but having the ability to invoke a web a webhook for initial context on call start would make this much more consistent, as my prompt based solution is not flawless and will sometimes not work as expected

awesome, i think i can get it working based off of that. thank you for your help @Shubham Bajaj Might i suggest making this behaviour more apparent for the future? the documentation on this is not so clear but at the same time it seems like something that a lot of applications would want to take advantage of

currently trying to forward an assistant from a call of

GET https://api.vapi.ai/assistant/:id

as the response body, but it doesn't seem to work appropriately at the moment (i have indeed wrapped it in the

assistant

object as shown in the documentation). will continue to debug in a bit, have a meeting to attend. but this is definitely the right track.

perhaps i need to do some transformation of the response from the assistant api. I want to get things working with the result of the response from the VAPI api, and then i will go ahead and add my custom context messages and so on.

I was hoping that feeding the result from GET /assistant/:id as an answer to

assistant-request

would work out of the box, but it seems not.

willing to provide you any ids or credentials if necessary.

python
def handle_assistant_request(request):
    url = "https://api.vapi.ai/assistant/<my assistant id>"

    headers = {"Authorization": "Bearer <token>"}

    assistant_response = requests.request("GET", url, headers=headers)

    response = {
        'assistant': json.loads(assistant_response.text),
    }
    del response['assistant']['orgId']
    del response['assistant']['id']
    del response['assistant']['createdAt']
    del response['assistant']['updatedAt']
    del response['assistant']['name']
    del response['assistant']['voice']
    del response['assistant']['endCallFunctionEnabled']
    del response['assistant']['transcriber']
    del response['assistant']['clientMessages']
    del response['assistant']['serverMessages']
    del response['assistant']['dialKeypadFunctionEnabled']
    del response['assistant']['serverUrl']
    del response['assistant']['maxDurationSeconds']
    del response['assistant']['backgroundSound']
    del response['assistant']['isServerUrlSecretSet']
    return JsonResponse(response, safe=False)

I went through and manually deleted every entry. other than what was in that sample response in the docs. seems to kind of work now, it at least accepts the assistant response. I am going to comment one by one and see which of these are causing the failure

definitely feels like something weird is going on here. i want to initiate my assistant with the exact settings that i have configured in the vapi UI, except i want to inject extra messages as context into the messages array of the model.

@Shubham Bajaj this is minimum code I was able to utilize to get things working. Things, are working, but obviously not an ideal developer experience 🙂

python
def handle_assistant_request(request):
    url = "https://api.vapi.ai/assistant/:id"

    headers = {"Authorization": "Bearer <token>"}

    assistant_response = requests.request("GET", url, headers=headers)

    response = {
        'assistant': json.loads(assistant_response.text),
    }

    # must delete
    del response['assistant']['orgId']
    del response['assistant']['id']

    del response['assistant']['isServerUrlSecretSet']

    del response['assistant']['createdAt']
    del response['assistant']['updatedAt']
    del response['assistant']['name']
    return JsonResponse(response, safe=False)

by deleting those keys from the response of the VAPI endpoint, things seem to work as expected.

if you include 1 or more of those fields, the api says "Could not get assistant".

as of right now i have things working, but if there is a way to simplify, i am all ears 🙂