KnowledgeBase doesn't work with Llama-Sharp on local LLM

I have configured Llama-Sharp and I am using the llama3.1-8b-instruct-q6_k.gguf model on my cpu. I have installed the knowledgebase plugin and I have also installed and configured qdrant. The knowledge that I have installed in qdrant is of the Q&A type. When I ask a question in the chat, Llama Sharp does not use knowledge and gives the user the answer through LLM. I have put the agent configuration below.
{
"id": "121a6dab-afa9-469f-bbe6-8df111b77700",
"name": "Q&A bot",
"description": "This is a Q&A bot",
"type": "task",
"mode": "eager",
"createdDateTime": "2025-05-10T12:45:57.3500389Z",
"updatedDateTime": "2025-07-02T09:52:29.17704Z",
"llmConfig": {
"is_inherit": false,
"provider": "llama-sharp",
"model": "llama3.1-8b-instruct-q6_k.gguf",
"max_recursion_depth": 16,
"max_output_tokens": 4096
},
"isPublic": true,
"disabled": false,
"iconUrl": null,
"profiles": [],
"labels": [],
"mergeUtility": false,
"utilities": [
{
"name": "kg.knowledge-base",
"disabled": false,
"functions": [
{
"name": "util-kg-knowledge_retrieval"
}
],
"templates": [
{
"name": "util-kg-knowledge_retrieval.fn"
}
]
}
],
"mcpTools": [],
"rules": [],
"knowledgeBases": [
{
"name": "BotSharp",
"type": "question-answer",
"disabled": false,
"confidence": 0.1
}
],
"inheritAgentId": null,
"maxMessageCount": 4096,
"routingRules": []
}

And the Instruct was as follows:
"
You are a helpful Q&A assistant. When the user asks a question, search the knowledge base using the 'util-kg-knowledge_retrieval' function to retrieve relevant information, then respond clearly and helpfully based on the retrieved knowledge. If no relevant knowledge is found, politely inform the user. Always keep answers accurate and easy to understand.

Use this function for retrieving knowledge "util-kg-knowledge_retrieval"
"

I ask the experts to help me solve this problem.

Start by reproducing the configured Q&A agent with Llama-Sharp, Qdrant, and the llama3.1-8b-instruct-q6_k.gguf model, then trace the util-kg-knowledge_retrieval function named in the configuration. Done means a question invokes knowledge retrieval and the response uses retrieved knowledge, or clearly reports when none is found.