GPT Router – Open-Source API Gateway for LLMs (OpenAI, Anthropic, etc.)

Hacker News

Share card

GPT Router – Open-Source API Gateway for LLMs (OpenAI, Anthropic, etc.)

Hi HN, Today, we are excited to be open-sourcing GPTRouter, an LLMOps tool we have been using internally at Writesonic for handling millions of monthly requests for our users. Universal API for 30+ LLMs, Vision and Image Models Smart Fallbacks based on latency and uptime Automatic Retries Supports streaming Since embracing OpenAI GPT-3 in production in 2020, we at Writesonic have been serving millions of users and faced the typical scaling pains with generative AI models: 1. Dependency on a single model risked total downtime. 2. Latency issues with models like GPT-4 affected user experience. 3. Integrating various models was tough due to different APIs and SDKs. Early this year at Writesonic, we set out with a clear vision: to become model agnostic. Faced with single-model limitations and diverse AI challenges, we began building GPTRouter - our bespoke solution to navigate and thrive in a multi-model AI world. With GPTRouter's Universal API, you're the master of AI models. Swap between OpenAI, Azure, Anthropic, Replicate, Cohere & more with just one line of code. It simplifies model management to a great extent. Downtime isn't an option. GPTRouter's Smart Fallbacks mean your service is always on. You can define a hierarchy of models for each use case. GPTRouter will constantly check for uptime/downtime, latency and other factors, and automatically fallback to the next best model with zero interruption. Say goodbye to manual retries. GPTRouter does the heavy lifting with Automatic Retries for failed requests, keeping your AI services sharp and consistent. GPTRouter's Edge: Universal API for seamless model switching. Smart, automatic fallbacks for continuous service. Reduced latencies for quick interactions. Additionally, we will also be open sourcing our frontend LLMOps layer that provides a playground to test multiple models in parallel, keep a tab on the latencies for each model, track tokens and costs for each model and user all in one place. We are looking forward to seeing how developers leverage GPTRouter in their own use cases. Thank you!