Launching today

VernLLM

Launching today

The production reliability layer for LLM applications

3 followers

The production reliability layer for LLM applications

3 followers

Visit website

What happens when your LLM call hits a rate limit, times out, or returns malformed JSON? With VernLLM, it just works. One typed call() wraps retries, timeouts, cancellation, and a self-healing circuit breaker, skipping retries on errors that won't fix themselves, like bad JSON or bad key. Get typed, Zod validated output via native JSON Schema mode. Cache calls, track cost per request, plug in your own logger. One interface across OpenAI, Groq, Anthropic, Gemini, Bedrock, and more. MIT licensed.

Free

Launch tags:Open Source•Developer Tools•Artificial Intelligence

Launch Team / Built With

Customer.ioAutomate Messaging Everywhere — Startups Get 12 Months Free

Promoted

OpenAI

OpenAI Day

What did GPT-5.6 Sol, Terra and Luna unlock that made your launch possible?

Maker

GPT-5.6 Sol, Terra, and Luna accelerated what developers can build with LLMs, making sophisticated AI applications accessible to small teams and individual developers. However, as LLM applications become more powerful, reliability becomes a critical challenge. Developers need consistent outputs, failure recovery, and production safeguards around increasingly capable models. These advancements created the need for VernLLM: a lightweight reliability layer between applications and LLM providers. VernLLM adds retries, timeouts, circuit breakers, caching, structured outputs, and validation through a simple developer focused interface. GPT-5.6 also helped accelerate development and iteration, allowing me to build and validate the product with a battle tested suite of 110+ unit and integration tests covering reliability features, provider adapters, and failure scenarios. The shift from "can we build this with AI?" to "can we run it reliably at scale?" is the problem VernLLM solves.

Report

15h ago

Maker

📌

Hey hunters! VernLLM is a lightweight resilience layer for OpenAI compatible LLM chat completion APIs. It wraps your existing LLM client and adds production reliability features like retries, timeouts, circuit breakers, caching, structured outputs, schema validation, usage tracking, and observability through one typed interface. I built VernLLM because while calling an LLM API is simple, running LLM applications reliably in production is not. Developers end up rebuilding the same infrastructure around every project: retry logic for transient failures, timeout handling, provider fallbacks, response validation, caching, and usage monitoring. The goal was to create a small, dependency light layer that keeps the simplicity of OpenAI's API style while adding the reliability features needed for real applications. During development, the focus evolved from solving individual reliability problems into building a complete reliability foundation that stays lightweight and flexible. Instead of locking developers into one provider or framework, VernLLM works across OpenAI, Groq, Mistral, DeepSeek, Cerebras, Together AI, Fireworks, Ollama, Anthropic, Gemini, AWS Bedrock, and custom HTTP endpoints. Key advantages: • Production reliability without a heavy framework • Automatic retries with exponential backoff and jitter • Timeout protection and circuit breaking • Structured outputs with provider native JSON Schema support • Typed schema validation for safer AI responses • Caching with custom adapters • Usage tracking and observability hooks • Zero bundled dependencies and TypeScript first design VernLLM is open source, MIT licensed, and tested with 110+ unit and integration tests covering core reliability features and provider adapters. If you are building with OpenAI or other LLM providers, I would love to hear what reliability challenges you have faced moving from prototype to production!

Report

1d ago

Maker

Il be active throughout the week to answer some questions!

Report

9h ago