ive been using
n8n
to connect api calls directly to a vector database, which keeps the context stable. its much more reliable than trying to manage long-running loops in a single chat window where the
context window eventually gets cluttered and hallucinates