Hey Product Hunt! 👋 I'm Yasser, the founder of Currai. I've spent the last 6+ years as a software engineer, and a lot of that lately has been building on top of LLMs. Every time I shipped something with AI in it, I hit the same wall: I had no idea what the model actually did. Why did it give that answer? Which prompt did it really see? Why did one little feature quietly cost more than the rest of my app combined? I was debugging surprise token bills with print() statements. So I built Currai — the observability layer I wished I had. Drop in one SDK and you can finally see inside your LLM app. Here's how it works: 🎯 Traces every prompt, token, and tool call in one view — install the SDK, paste your keys, wrap a call, and your first trace shows up in ~5 minutes. No agent, no collector, no infrastructure to run. 💸 Rolls up token cost for you — per trace, model, user, and day — so you stop guessing what an AI feature actually costs. 🧪 A/B tests your prompts in production — split live traffic across prompt versions by weight, then compare cost and quality right in your traces. Stop arguing about wording; measure it. And if you're already on Langfuse or OpenTelemetry, you migrate by changing a single line — your existing trace code, spans, and exporters just keep working. Currai isn't another print() statement, and it isn't a self-hosted cluster you have to babysit (no ClickHouse to keep alive). It's hosted observability you drop in and own — you watch your data, not your infrastructure. It's free to start — first traces in about five minutes, no credit card needed. 👉 currai.app print() → production. 🧵

Currai

Observability & A/B Testing for LLM Apps

Observability & A/B Testing for LLM Apps