My function tool requires three inputs: shop_id, customer_phone, and reason. The reason field is the field that requires ai to extract info from the conversation. However the shop_id and customer_phone fields don't need ai to extract info from the conversation. Instead values for these fields are passed in as variables from my code. Is there a way to do this using vapi tool calling?

Hi there! 👋 I understand you're looking to pass

shop_id

and

customer_phone

directly from your code while having AI extract the

reason

field from the conversation. Good news - this is definitely possible with Vapi tool calling! Yes, Vapi tool calling can handle this mixed approach where some variables come directly from your code and others are extracted by AI from the conversation. Here's how you can do this: 1. **Define your function tool with all three parameters**:

json
   {
     "type": "function",
     "functionName": "your_function_name",
     "parameters": {
       "type": "object",
       "properties": {
         "shop_id": {
           "type": "string",
           "description": "The ID of the shop"
         },
         "customer_phone": {
           "type": "string",
           "description": "The customer's phone number"
         },
         "reason": {
           "type": "string",
           "description": "The reason extracted from the conversation"
         }
       },
       "required": ["shop_id", "customer_phone", "reason"]
     }
   }

2. Pass your variables directly from code using

assistantOverrides

with

variableValues

when starting a call:

javascript
   vapiInstance = vapiSDK.run({
     apiKey: apiKey,
     assistant: assistant,
     config: config,
     assistantOverrides: {
       variableValues: {
         shop_id: "your-shop-id-here",
         customer_phone: "customer-phone-here"
       }
     }
   });

3. Handle the tool call in your code: ```javascript vapi.on("tool-call", async (toolCall) => { if (toolCall.name === "your_function_name") { // The AI will have extracted the reason from the conversation // while shop_id and customer_phone come from your variables const { shop_id, customer_phone, reason } = toolCall.parameters; // Use all three parameters to call your function

can you show me how to implement in python instead?

You just need to define those function parameters and add the description to those fields so that when LLM does a similarity search, it can find out these two terms are similar and pass the values according to the conversation. It's all about prompting and tool creation - nothing else.