The problem is that the AI is trying to call a tool, but that tool’s definition is broken—its schema is missing, so the AI doesn’t know what parameters to expect.
In short: I'm building is a system that uses AI and a Pinecone vector database to answer questions, like finding email addresses. But it seems nnothing is working.
We tried a few solutions along the way. First, we ensured the Pinecone vector store node was set up correctly, double-checking credentials, index, and namespace. Next, we adjusted the operation mode from "Insert" to "Retrieve" to ensure we were querying, not uploading. We also simplified the tool name to avoid any truncation issues. We added a concise description and ensured the AI agent was explicitly told to use the vector tool. However, none of these resolved the issue. The problem persisted because the AI tool definition was missing a proper schema.
Please help me
yeah that’s the issue — if the tool schema is missing or malformed, the ai literally has no idea what parameters to send. we ran into exactly this with vector tools before (schema had type: null instead of type: object). the fix is right on track — just make sure when you recreate the tool node, the schema has all your expected parameters fully defined. also worth testing the tool output separately before connecting it to the agent, that usually catches schema mismatches fast
Can you be a bit more specifiv? What exactly solved the problem? I tried to rewrite te prompts in Vector Store Question Answer Tool node but none of this helped.
And also don’t know to make sure to fully define the parameters while all tha data is in Pinecone, and there is no problem with credentials.
Hi @Andras_Turnai Welcome!
Consider removing that middle node mail-research and directly connect Pinecone tool to the AI agent, make sure you give good system prompt on when to use it and also give a nice tool description , and that would work.
Thank you, thank you, thank you. I’ve been searching for the solution for a long time. I bought a udemy course, and this was one of the tasks. I did everything like the guy in the video but nothing worked. Asked all the LLM-s but none of them could find the solution. Finally there is a solution:D
