Andrew Ng's team drops aisuite to unify LLM APIs

aisuite is a lightweight open-source Python library from Andrew Ng’s team designed to fix one of GenAI's most annoying developer headaches: fragmented APIs. It lets you swap LLMs across providers by changing a single string in your code.

Anyone who has built a multi-model app knows the pain. OpenAI has its own client format, Anthropic requires different conventions, and local tools like Ollama bring yet another interface. Comparing them usually means writing ugly glue code and rewriting request handlers.

How it works

The idea behind aisuite is simple and clean. It abstracts the underlying API complexities into a single, standardized framework that handles different request and response formats uniformly.

Instead of juggling different SDKs, you define models using a simple "provider:model" format. You can call "openai:gpt-4o", swap it for "anthropic:claude-3-5-sonnet-20241022", or pivot to a local setup with "ollama:llama3.1:8b".

Your core application logic stays completely untouched. As long as the environment keys or endpoints are configured, switching models takes literally two seconds.

What you can do with it

This unified interface makes multi-model workflows much easier to build and maintain.

For example, your app could route open-ended creative tasks to OpenAI's GPT-4, and then route strict factual extraction to an Anthropic model. Instead of maintaining multiple SDK clients and formatting differences, you just pass different model strings.

The library is also built for extensibility, making it straightforward to plug in new models and providers as they appear. The project is completely open source and available directly on GitHub.

Why it matters

We think aisuite hits a sweet spot for rapid prototyping. Model performance changes by the week, and tying an entire codebase to one provider's proprietary SDK is risky and unnecessary.

If you are an engineer or builder experimenting with multiple models—or balancing local inference via Ollama with proprietary cloud APIs—aisuite is well worth testing. It is minimal, fast to adopt, and spares you from writing boilerplate code, even if power users might still need native SDKs for provider-specific edge cases.