Call id: 7ccb1e13-b4b8-4c95-9c9e-9d9b7eeb54c2 As you can see, my assistant wrongly said the "request-complete" message even though the request failed. we ended up getting a tool call result like this: { "name": "place_order", "role": "tool_call_result", "time": 1747689324398, "result": "Error: Failed to add order (forced failure for testing)", "metadata": {}, "toolCallId": "call_PkVjPULo9vPHH5Q8tIfNfLyg", "secondsFromStart": 22.109 } Is there a way to mark the request as failed? Is this tool call result not the right way to do it?

To ensure your assistant says the correct failure message, update your tool's configuration to include a

request-failed

message type, like this:

json
"messages": [
  { "type": "request-complete", "content": "Order placed successfully." },
  { "type": "request-failed", "content": "Sorry, there was a problem placing your order." }
]

Then, adjust your server logic so that when an error occurs, the result is structured in a way that triggers the

request-failed

path (e.g., by returning a recognizable error string or object). This helps the assistant pick up failures correctly. [See documentation for details](https://docs.vapi.ai/tools/custom-tools#adapting-the-payload-for-your-needs) Source: - [Custom Tools – Adapting the Payload for Your Needs](https://docs.vapi.ai/tools/custom-tools#adapting-the-payload-for-your-needs)

I did that already. request complete, request start both work. but it doenst recognize request failed. is that something my tool must return? "messages": [ { "type": "request-start", "content": "One moment while I place your order..." }, { "type": "request-complete", "content": "I've sent your order to the kitchen. It will be ready in around 30 minutes." }, { "type": "request-failed", "content": "I apologize, but I'm having trouble placing your order right now. Would you like me to try again?" }, { "type": "request-response-delayed", "content": "It's taking a moment to process your order. Please hold on...", "timingMilliseconds": 2000 } ] } ],

Remove the request-complete part and then set async = false. It will work.

What do you mean

Hi Liamdu, For your failed tool call to properly trigger the "request-failed" message, return this structure: { "results": \[{ "name": "place_order", "toolCallId": "call_PkVjPULo9vPHH5Q8tIfNfLyg", "error": "Failed to add order (forced failure for testing)" }\] } The key difference from your current implementation is: 1\. Wrap your tool result in a

results

array 2\. Use the

error

field instead of

result

for error messages Your message configuration is correct - this change in response format will ensure the appropriate "request-failed" message is triggered.