RAG Agent Looping Issue with Supabase Vector Store – No Output, High Token Usage

I'm building a RAG Agent using Supabase as my vector store for document retrieval. The workflow involves a RAG Agent > Chat Model > Chat Memory > Tool (Supabase Vector Store under "Retrieve Documents") > Embedded Reranker > Reranker > and then loops back to the RAG Agent.

The problem: it's stuck in this cycle, consuming AI credits rapidly without ever producing a final answer. The model just keeps passing control between components and re-entering the loop.

Has anyone encountered this issue with Supabase + RAG Agent? Is it a problem with tool definition, vector dimensions, reranker configuration, or memory setup? Any suggestions on how to break this cycle and ensure the agent actually responds would be much appreciated.

3 comments