Unify

Unify is designed to optimize your large language model (LLM) applications by dynamically routing prompts to the best available endpoints. This service lets you use a single API key to tap into all LLMs from multiple providers, making your life easier when you're building your own applications.

Unify's main features include:

• All Models, All Providers, One API: Use a unified API to interact with multiple LLMs without needing separate API keys.
• Custom Routing: Set your own cost, latency, and output speed requirements, and define a custom quality metric that works for you.
• Live Benchmarks: Get the latest benchmark data for your region, updated every 10 minutes, to pick the fastest provider.

Customers can set their own cost, latency, and output speed requirements, and define a custom quality metric to customize their router for their specific needs. Unify automatically sends requests to the fastest provider based on the latest benchmark data. This approach is designed to produce the best possible output while balancing speed and cost.

The benefits of Unify include:

• Better Results: Get the best results by taking advantage of the strengths of each LLM.
• More Flexibility: Support a wide range of input types and complexity by routing requests to the best LLM for the job.
• Better Resource Efficiency: Route requests to the best LLM to minimize costs and maximize speed.
• Faster Development: Repurpose existing LLM capabilities to speed up your AI application development and deployment.

Unify charges customers based on a credits system, with 1 credit worth $1. Customers only pay what the endpoint providers charge themselves, with no markup. New customers get $50 in free credits.

Unify is geared for developers and companies building production LLM applications. It automates the complex process of optimizing LLM selection so you can focus on the important logic of your application. By publishing transparent, unbiased and up-to-date benchmarks, Unify gives you the information you need to make the best decisions about which models and providers are best for your use case.

Published on June 14, 2024