N8n RAG Configuration: Agent Not Processing Vector Store Outputs

[Alessio_Lanzillotta]

Hello everyone,
I’m trying to make a RAG (Retrieval-Augmented Generation) system work. It queries a PDF that has been vectorized in Qdrant. Apparently, the response returned from the vector store tool created by Ollama 3.1 locally is always perfect, but it seems the agent doesn’t read the output from the vector store and says it doesn’t have enough information.
The chat model I’m using is the same for both the vector store and the agent.
Can anyone help me understand why this is happening or suggest a solution?
Thank you in advance for your assistance.

vector store perfect answer

chat model error

Information on your n8n setup

• n8n version:
• 1.70.3
• Database (default: SQLite):
• default
• n8n EXECUTIONS_PROCESS setting (default: own, main):
• default (docker compose installation)
• Running n8n via (Docker, npm, n8n cloud, desktop app):
• docker
• Operating system:
• w11 24h2

[Alessio_Lanzillotta]

@Alessio_Lanzillotta

“systemMessage”: “tienes que enviar como respuesta el texto che da como output “Ollama Chat Model1””

The original system message was:
Purpose: To interact directly with the user and coordinate the execution of tools to provide relevant answers.
Key functions:
Process user inputs, understand the query’s intention, and decide how to handle it.
Call specific tools (such as vector databases or memory systems) when necessary.
Generate responses based on information obtained from the tools.
Facilitate a smooth and dynamic conversation experience.
However, it still doesn’t work.

[Alessio_Lanzillotta]

resuelto!
el problema es llama3.1 GGUF, con llama 3.1:latest funciona todo!

[ihortom]

Thaks for sharing @Alessio_Lanzillotta !