I have an AI agent connected to a pinecone vector database. When I talk to it within the n8n instance using the chat node and I ask a question, it gives me the perfect answer. But when I use the chat widget on my website and I ask the same question, it does not answer correctly. What can be causing this?
In executions I saw it go through and asked the correct question within pinecone vector store, the data coming through pinecone contained the appropriate information, but the agent disregarded the information.
Information on your n8n setup
- n8n version: 1.81.4
- Database (default: SQLite):
- n8n EXECUTIONS_PROCESS setting (default: own, main):
- Running n8n via (Docker, npm, n8n cloud, desktop app):
- Operating system:
Could you possible ctl a ctl c your entire workflow and insert it into the field when clicking the ‘</>’ button on the forums here?
Sorry i didn’t understand the instructions, i’m relatively new to this whole thing. Could you explain it please?
Yeah sure.
When you are in n8n, on the workflow. You can to ctl + a (or cmd + a for mac) and then ctl + c (or cmd + c) to copy the entire workflow.
Then when creating a new forum reply, above the chat there is a ‘</>’ button, and you can paste it into the field that appears.
I am not confident, I can check deeper later. But could potentially be because there are two variables within the agent user message. Could you screenshot what it looks like on a run inside the ai gent node? With inputs?
Now for some reason it won’t even receive the chat input from the website link. Although when i did a checkup of the data being received and the output data of each node, everything is the same up to the answer of the AI agent. Meaning both chat inputs trigger the pinecone database and give the right answer, but when using the chat from the website, the answer is limited as if there was no vector database. What can be the issue? I thought it might have to do with the website’s chat not having access to pinecone so the answers are being limited/restricted?
Could you show some more input/output run details?
Did you investigate the network requests in chrome/browser developer tools? A common issue is n8n not connecting to its websocket, so bad data could be sent, or none at all.
