Hey Product Hunt! I built VernLLM after watching the same pattern repeat across every LLM project I touched. Someone ships a naive wrapper around the provider's chat completion call, it works fine in the demo, then a provider hiccups in production and the whole pipeline goes down because there was no retry, no timeout, no fallback. VernLLM is my answer to that. It's a resilience layer, not a new SDK to learn. Bring your existing OpenAI, Anthropic, or any other client, wrap it, and get retries, circuit breaking, provider fallback, rate limiting, caching, and structured output validation for free. A few things I'm especially proud of. Cache coalescing that actually prevents stampedes on concurrent misses, not just a wrapper around a key value store. Circuit breakers that are isolated per target, so one provider tripping doesn't affect another. And zero magic, every feature is opt in, fully typed, and documented with real API shapes, not marketing pseudocode. I'm one engineer who got tired of writing this from scratch on every project, so I made it something the community can rely on instead. Would genuinely love to hear what you're running into with LLM reliability in your own stack, and what would make VernLLM a fit for it.

VernLLM

The resilience layer for LLM calls

The resilience layer for LLM calls