Build a Modular GraphQL Backend for LLMs with Cosmo Plugins

We were trying to make API orchestration easier. Instead, we built a backend framework for LLMs. And it's actually good.

If you like the idea of vibe-coding backend services with LLMs, keep reading. If not, skip to the part where we reduce 20 API calls per page to just 1.

What Are Cosmo Plugins?

A plugin is a module that:

• Has a schema
• Gets a gRPC definition
• Uses Cursor to generate the adapter
• Runs as a subprocess inside the Cosmo Router (thanks to go-plugin)

Every plugin maps to a piece of your Supergraph. You can build modular monoliths or microservices. You can mix both. It’s up to you.

Why We Built This

We were working with a customer who had thousands of REST-ish APIs. Rewriting them all to GraphQL? Not gonna happen. But they wanted to use a Supergraph.

So we built a way to use LLMs to do the translation. One plugin per service. One query per page. That’s it.

Quickstart: Your First Plugin

npx wgc@latest router plugin init hello-world

This scaffolds a full setup. Open Cursor in the plugin dir. Use the built-in prompts to generate code. Modify the schema if needed. Done.

Add another module? Run the init command again. Update the Router config to compose it.

Handling Dependencies

Handled. That’s what Federation does best. Declare your needs. The Router composes and resolves it.

Learn More and Join Us

• Plugin docs
• Router docs
• Discord

We’re just scratching the surface. Let us know what you build.

Originally published on WunderGraph’s blog