Force AI Agent to use the tool

Questions
1

Hey community,

I am developing an AI agent which has a large Supabase PGVector knowledge base. Now it often happens that the AI agent doesn’t use the tool, actually most of the time. I have specified in the prompt to use the tool called (name of the Supabase Vector Store node) “CompanyKnowledgeBase” but no result.

Could it be that this is because I am using an Output parser? I am having other issues as well with the output parser.

System prompt:
You are AI Coach and starting a test XYZ.

Explain what this test is about, searching the CompanyKnowledgeBase.

Then present EXACTLY the following first next statement and ask the user to give a rating from 1–5: ${$input.first().json.Item}

Follow the provided JSON schema for your response: {“type”:“object”,“properties”:{“output”:{“type”:“object”,“properties”:{“canProceedWithNextQuestion”:{“type”:“boolean”,“description”:“true in this case”},“previousScore”:{“type”:[“number”,“null”],“description”:“null in this case”},“response”:{“type”:“string”,“description”:“Your response”}},
“required”:[“canProceedWithNextQuestion”,“previousScore”,“response”],“additionalProperties”:false}},“required”:[“output”],“additionalProperties”:false}}`

This is how the output parser schema looks like:
{
“canProceedWithNextQuestion”: “Whether we can proceed with the next question, if the last one has been answered with a score (true or false)”,
“previousScore”: “The user’s numeric answer (1–5) to the question, or null if invalid”,
“response”: “Your statement and next question or user query”
}

Can anyone help?

2

Would you mind updating your question with your workflow to gain a better understanding of the problem?
It could be just a name mismatch or something simple, but if you want to improve the usage of your tools, you can use the MCP client/server for better flexibility and tool descriptions.

3

If you want to force your agent to retrieve content from the vector store, then use the Question and Answer node like below. If you have a chat bot / agent doing tool calls, then wrap this example in its own worflow and call the workflow as a tool from the main agent

4

Basically I am trying to implement some kind of quiz. Here there are couple of cases:

  1. The quiz starts → AI Agent should provide the initial message and the first question
  2. Normal procedure → AI Agent should answer queries or save answer and propose next question
  3. After every X question → AI Agent should give a score
  4. After all of the questions → AI Agent should give a final score and talk about the results

Why am I using two agents?
This is also due to another issue since I need the “score interpretation” output. When I was having both in one agent, due to the output parser I was getting interpretations when I shouldn’t. Also the output parser doesn’t work right, like sometimes I would get jsons from the other agent…