Hey HN, I'm 16 and built this after an agent loop cost me $500. WatchLLM auto-terminates agents when they hit cost limits and alerts you in real-time via Slack/Discord. Also has semantic caching (70% savings) and streaming response caching (industry first). Built-in agent debugger shows you exactly where loops happen. Just launched cost kill switch + real-time alerts this week.

Would love feedback from the HN community!