I have programmed an assistant that uses knowledge base to lookup information that is not included in the prompt. If I upload the files directly to the assistant it DOES lookups in the knowledge base and then speaks the answers out. (See screenshot 1) If I instead create a custom query tool, upload the same files there and then link the assistant with that tool it DOES lookups but it DOES NOT speak out the answer out after retrieving it. Only after nudging it several times sometimes it will actually decide to speak out the answer. (See screenshot 2) Do you know what I might be doing wrong? https://cdn.discordapp.com/attachments/1404780697356472391/1404780697587028098/Screenshot_2025-08-12_at_12.56.16.png?ex=689c6f79&is=689b1df9&hm=9db9538d768cadcb8287f19610bfa2797b03a6ec873fd63e3d41cef0c8ad173c& https://cdn.discordapp.com/attachments/1404780697356472391/1404780698132545576/Screenshot_2025-08-12_at_12.54.25.png?ex=689c6f79&is=689b1df9&hm=7cd035d634dbd0704ce45f99573dddc182ff8c9ae387c23e2df1d13ff4fd5754&

Suggestion: When attaching your custom query tool, ensure you include the entire model object in your PATCH request—omitting fields like

temperature

or

messages

can cause the assistant to lose its ability to respond properly. Here’s a sample PATCH request:

bash
curl --location --request PATCH 'https://api.vapi.ai/assistant/ASSISTANT_ID' \
  --header 'Authorization: Bearer <YOUR_API_KEY>' \
  --data '{
    "model": {
      "temperature": 0.2,
      "provider": "openai",
      "model": "gpt-4o",
      "toolIds": ["YOUR_QUERY_TOOL_ID"]
    }
  }'

Double-check that all relevant model fields are present to avoid overwriting essential configuration. [See documentation for details.](https://docs.vapi.ai/knowledge-base/using-query-tool) Source: - [Using the Query Tool for Knowledge Bases](https://docs.vapi.ai/knowledge-base/using-query-tool)

Ensure your custom query tool uses the correct synchronous mode, response format, system prompt, and message settings so the assistant both retrieves and speaks answers automatically. If issue persist share call_id