Side project feedback: is there a gap after LLM budgets and guardrails?

I’m Sungtae at OraXLabs, researching Boundary, an early commercial side project. AI helped draft this post. Before building further, I’d like feedback from US teams of roughly 3–20 people already paying for LLM APIs, especially through LiteLLM or OpenRouter. What happened the last time one workflow incurred an unwanted API cost? What budget, limit or guardrail was already configured, and did your subsequent fix solve it? If your current controls work well and another tool would add no value, that is useful feedback too. I’m trying to understand whether a separate project is warranted, not offering a finished beta or promising free access or savings. Please leave out credentials, customer data and private logs. Public replies are welcome. If you prefer a voluntary, unpaid 20-minute conversation, mention that here; any private contact would happen only with your agreement.