latitude-dev/latitude-llm ? reverse-engineered prompt

Build me an open source AI monitoring app that helps me see when my agents are failing and makes it easy to fix them.

I want a web dashboard where I can send in traces from my LLM app, view full conversations and tool calls, and spot failing runs over time. Group similar failures together so I can see what is breaking most often, then let me inspect the raw trace, add notes or tags, and track the issue until it is fixed. I also want a way to push the problem context back into my coding agent so it can make a small fix and, if possible, verify the fix against the same failing traces. Make it work with common LLM providers and let me connect it from a TypeScript app. If you need to check current setup details or docs online, do that. Keep it self hosted friendly, clean, and practical for real teams.