Hey Product Hunt 👋 Maker here. I'm CPO at an AI fintech in my day job. We ship multi-agent AI for treasury automation to tier-1 banks. I've watched smaller versions of the OpenClaw $1.3M burn hit production 3 times in 18 months. The thing that kept bugging me: spend caps don't actually stop the bleeding. They just cap it. The agent still loops 50 times before hitting your $50 ceiling. You lose $50 instead of $200, but you've lost the work in flight. The fix isn't a tighter cap. It's catching the loop pattern at iteration 3, before the dollar ceiling matters. Tokenwall does three things: 1. Reserve-then-commit budgets per action (treasury pattern). 2. Loop detection at iteration 3 (repeated tool calls, similar args). 3. Per-user / per-task / per-agent attribution. This is validation week. The signup form goes to a real KV backend. If 30+ builders join the beta by Friday, I build the proxy over 4 weeks. If not, I kill it and write the post-mortem. Honest feedback welcome, especially from anyone who's had the "what just happened to my API bill" moment. — Atul (linkedin.com/in/atulydv)

Tokenwall

Circuit breaker for AI agents. Catch loops at iteration 3

Circuit breaker for AI agents. Catch loops at iteration 3