Query Tool Transparency & error
# supportj
Jimmy
12/09/2025, 11:48 AMHi team,
Thanks in advance, this one's a bit long
---
Running into a critical issue where the agent is prompted to use the query tool to look up details for a procedure. When using the query tool I can see it's querying the right knowledge base
However, there are either no results returned or the wrong knowledge base is used. I have another agent with the exact same setup and it consistently gets it right, while the one in production consistently fails.
We don't have transparency into HOW the agent is querying, only that it queries a specific KB:
Query:
{"knowledgeBaseNames":["clinic_appointment_type"]}{"knowledgeBaseNames": ["clinic_appointment_type"]}v
Vapi Support Bot
12/09/2025, 11:49 AMTo resolve inconsistent query tool results, ensure your assistant's system prompt explicitly instructs when to use the query tool and references the correct tool name. For example:
Copy code
json
{
"role": "system",
"content": "When users ask about appointment types, use the 'clinic-appointment-type-query' tool to search the knowledge base. Always call this tool before answering."
}j
Jimmy
12/09/2025, 12:26 PMI have done this ^ The only difference I can think of currently is that the broken agent was created in October while the working agent was more recent. Is there some sort of caching that could be resolved by recreating the broken agent?
Jimmy
12/09/2025, 12:28 PMIt looks like what the broken tool call is doing is returning parts of the system prompt as the response. eg
`[
{
"name": "qr_j573wcg4bba9gb6gfeef2dt8s97tcprf",
"toolCallId": "call_c4ORQGfemxDm9yVSLyDOY4Bo",
"type": "query",
"result": "```text\n2. Ask: \"What days and times work best for you?\"\n```"
}
]
`
Call ID: 019b0311-bcbc-7cc2-8acf-38e64476d01d
Jimmy
12/09/2025, 12:38 PMAnother example: 019b031b-9f8b-799d-8c2f-9b28fda721b6
`[
{
"name": "qr_j573wcg4bba9gb6gfeef2dt8s97tcprf",
"toolCallId": "call_FGjEHGNLrtTtJBGj8dNwcFqb",
"type": "query",
"result": "```text\nMonday: 08:00 - 18:00\n```"
}
]`
Jimmy
12/09/2025, 1:45 PMAnother example: 019b0358-d3dd-7114-be07-13e17073fe5e
It's worth noting that the two broken queries are both in the same org, while the working query is in a separate org
`[
{
"name": "qr_j578hcrrz0c94ycfpnpg5pa0gs7wxf2p",
"toolCallId": "call_3LgESgCRo01ibsGnK4IkOPVg",
"type": "query",
"result": "```text\nFriday: 09:00 - 17:00\n```"
}
]`
Jimmy
12/10/2025, 11:17 AMExample here where the query returns part of the recent context
call id: 019b07f5-2005-7441-b28e-57dfc5fdff44
[
{
"name": "qr_j573wcg4bba9gb6gfeef2dt8s97tcprf",
"toolCallId": "call_1VNMdm5QvM0ec8CNFx4mlQRJ",
"type": "query",
"result": "```text\nOur clinic is open Monday to Friday. We slightly shorter hours on Wednesdays.\n```"
}
]
Jimmy
12/10/2025, 11:17 AM@Vapi sorry to tag
Jimmy
12/10/2025, 11:39 AM@User
c
Chiranjeet Mishra
12/10/2025, 12:51 PMHi bobbychick,
It seems you're experiencing inconsistent query tool results across different agents, despite having similar configurations. Here are a few steps to help resolve this:
1. **Confirm System Prompts**: Ensure the system prompts for both agents clearly specify when to use the query tool and reference the correct tool name. Consistency here can significantly impact tool invocation.
2. **Check Knowledge Base Descriptions**: Verify that the descriptions accurately reflect the intended use, as they help the assistant understand when to query.
3. **Logs Analysis**: Review the logs to ensure the queries reference the correct knowledge base. For example, make sure that
{"knowledgeBaseNames":["clinic_appointment_type"]}