Pinecone Vector Database not being used, using openAI Chat model instead

Questions
1

I am currently trying to make my workflow use the Pinecone Vector Database.

For specific questions I would like to access my Data in Pinecone (txt file). But it is not being used. Instead it is using the openAI chat model. To fix this quicker and you are willing to help me live, my discord username is: bertl19 . you can add it and if you are willing to do so we could try fixing this live.

What is the error message (if any)?

There is no real “error message” being displayed, but when looking at the Logs, the Pinecone Vector Database does not show up. Therefore it was not used. I’ve tried editing the system prompt of my orchestrator but no success. This is what my workflow looks like:

grafik
grafik1485×999 169 KB

Workflow

Information on my n8n setup

  • n8n version:
  • Database: Pinecone
  • Running n8n via (Docker, npm, n8n cloud, desktop app): Cloud
  • Operating system: Ubuntu
2

I’ve just completed a workflow where I access data from a Pinecone database. I followed this tutorial: https://www.youtube.com/watch?v=UeFi5oV9UpY

Your workflow is more complex but the main difference I’m seeing is your Orchestrator connects directly to your Pinecone Vector Store as a tool. I have mine connected to a “Answer questions with a Vector Store“ node. This node takes in a Vector Store and a Model. This is where I connect my Pinecone Vector Store. The above tutorial was super helpful for me.

3

hey, sorry to bother you again! I’ve tried building the tutorial step by step. Still with the same kind of result. My pinecone is not being acessed here either. ‘I don’t know is the default response, when something is wrong’. As you can see in the logs, it is not using the vector store. (double- and triple-checked with setting up pinecone etc.).

Thanks a lot for your help already!

grafik
grafik1764×1398 288 KB

4

Ai agent system prompt with clear instructions referring to the tool name, and parameters eventually.

Remove the simple memory for test.

1 Like
5

No bother at all. I’m very much a beginner with all of this but I also had a little trouble. In my Pinecone Vector Store I have the following Parameters:

  1. My Pinecone Credential
  2. Retrieve Documents (As Vector Store for Chain/Tool)
  3. Pincone Index - The Pinecone index holding my data. This was available via a dropdown based off of the credential connection in number 1.
  4. Pinecone Namespace - The specific namespace that I created in Pinecone.

I hope this helps. :slight_smile:

6

Hey! Pinecone Developer Advocate here!

When using Pinecone Vector Store node as a tool in AI Agent node (your Orchestrator), set the System Message and give the Pinecone Vector Store node a very descriptive description as to what it does.

Something like this for System Message:

Use the Pinecone Vector Store tool to retrieve data about XYZ

And for the tool description:

Contains data about XYZ

Be as descriptive as you can as these are instructions to the model to decide if/when/how to use the tools.

Also, you probably don’t need the Answer questions with a Vector Store node as this means you’ll be doing chat generation twice (extra tokens usage, extra cost, possible degradation in accuracy). Using the Pinecone Vector Store node as a tool directly will return the search results, which will get sent to the model (attached to your Orchestrator) which will then generate a natural language output from them.

1 Like