ChatGPT Assistants API will max out context window and run up costs in a loop

I thought there was some conspiracy afoot myself until I realized I had a silent exception in my handler for a specific requires_action call. It looked like OAI was trying to clean me out by re-running the same command over and over.

Every support ticket I submitted so far regarding the Assistant API was ultimately my fault.

To date, I have not had an assistant API run that actually went off rails on its own. We're about $100 deep in spend since the announcement.