Hey Product Hunt! 👋 I'm building Lumni to solve a problem that's becoming increasingly important as AI agents start taking real-world actions. An agent saying "done" doesn't mean the job was actually done. Imagine a support agent telling a customer their refund was processed, even though the payment API returned an error. No obvious crash. No alert. Just an unhappy customer. Lumni helps teams catch these silent failures, understand what went wrong, replay incidents, and turn failures into regression checks. You can start by pasting an agent trace, without installing an SDK. We're especially interested in agents handling refunds, cancellations, orders, and other business-critical actions. I'd love your feedback on two questions: 1. What's the hardest AI agent failure you've had to debug? 2. How do you currently verify that an agent actually completed its task? Excited to hear what you're building and where Lumni could help!

Lumni

Catch AI agent failures that look like success

Catch AI agent failures that look like success