Hey PH 👋 I'm Varad — 20, engineering student, built ARGUS because I lost 6 hours debugging a pipeline that never threw a single error. The node succeeded. The data was wrong. Four steps later — garbage output.LangSmith showed green the whole time. That's the problem nobody talks about. AI agents don't crash. They quietly lie. ARGUS sits at every node boundary and catches it the moment it happens. Wrong field, missing data, type mismatch: right there, not after your user notices. What makes it different from LangSmith/Langfuse: Those show you traces. ARGUS validates whether what each node produced was actually correct. Different layer entirely. Also built deterministic replay — freeze the LLM responses from a failed run, re-execute from any specific node without burning API costs again. This one alone has saved hours. Open source, AGPL, free forever for core features. Would love brutal feedback — especially if you've hit this problem yourself. And if you find something broken, tell me directly. I'll fix it same day. If you want to follow the build and talk directly : https://discord.gg/dMWU3r92e2

ARGUS

Catch Silent Failures in your AI Agent Pipelines

Catch Silent Failures in your AI Agent Pipelines