Use Pydantic discriminator so we can easily parse webhooks

[emhagman]

Hi!

We use the Vapi SDK for both the client but also for type checking in our webhooks we received. Currently, we have to do some reflection with Pydantic to more easily/programatically parse the events.

However, Pydantic supports discriminators https://docs.pydantic.dev/latest/concepts/fields/#discriminator where a field can determine the model type. It would be really nice to be able to do something like:

@router.post("/vapi/events")
async def webhook_vapi_events(request: Request):
    # make sure the data coming in is from vapi
    if not authorize_vapi_webhook_request(request):
        logger.warning("Invalid secret supplied via webhook")
        return JSONResponse(status_code=401, content={"status": "error", "message": "Invalid secret"})

    # continue processing
    json_data = await request.json()

    # parse our event
    event_type = VapiEvent.model_validate(json_data)

    # pydantic automatically maps/casts to the model
    # event_type: ServerMessageStatusUpdate in the case of `status-update`
    ...

Currently, we have to have a mapping from type to each model like so:

    message = json_data.get("message")
    if not message:
        logger.warning("No message found in request")
        return {"status": "missing-message"}

    raw_type = message.get("type")
    if not raw_type:
        logger.warning("No message type found in request")
        return {"status": "missing-message-type"}

    # we have to manually parse models
    event = mapping[raw_type]
    

Where mapping is a generated via reflection from ServerMessageStatusUpdate etc based on the type field which used to be a default string but currently uses literals.

Are there any future plans to support discriminators in the future to more easily parse these? I believe you use fern to generate your SDK and they do support this: https://buildwithfern.com/learn/api-definitions/ferndef/types#discriminated-unions