Advanced Bot Agent with different transitions - RAG agent not working properly

Questions
,
1

My plan is to create a chat-agent that has the ability to transition to different outcomes - but idk how to set that up. Also for somereason my tool agent has stopped using the RAG vector system to get an answer back to me, its started to create its own answers - im running this on my mac not docker.

This is my current setup -

2

It looks like your topic is missing some important information. Could you provide the following if applicable.

  • n8n version:
  • Database (default: SQLite):
  • n8n EXECUTIONS_PROCESS setting (default: own, main):
  • Running n8n via (Docker, npm, n8n cloud, desktop app):
  • Operating system:
3

so if i prompt it better and use a better llm i can get it to work better… but now idk what llm would be best to use a local one?

4

Hi @Maki99,

Also for somereason my tool agent has stopped using the RAG vector system to get an answer back to me, its started to create its own answers

^ For this, you can try adding another rule in your system prompt that instructs the Agent to not add any made-up answers, something like this:

You are a personal assistant designed to provide answers based on the documents in the school knowledge base.

Rule 1: Respond directly with the answer, without including prefatory phrases such as "Here's the answer formatted to your request:" or anything similar.

Rule 2: If the answer is not found in the school knowledge base, do not create or speculate an answer. Instead, state that you do not know the answer.

Which Chat model are you using at the moment?

6

So i fixed my issue - basically my airtable was giving out 9 records - and the issue is i didnt realize… when i put the schema into the agent ALL 9 records will go through… thats why it started to answer multiple times giving me a random final answer. so i fixed it by filtering to 1 record. i was using claude 3.5

But now im facing a new issue my Google drive Vector to DB is giving me an error - it was working before with the same document but now idk why it isnt - it also seems to have deleted the vectors that was already created and saved in my superbase - it fails on the superbase node where it says that the Data loader couldnt initialize - or sometimes it doesnt even say that it acts like it was successful but nothing happened, and no new vectors where created - was working before idk if the 1.65 update broke that or not. its not like i changed anything

7

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.