Here’s what I learned about calling LLM APIs from the browser when building AI chat functionality in my note taking app Edna.

The API I care about is getting LLM response to a question in a streaming way.

OpenAI pioneered this and created https://api.openai.com/v1/chat/completions POST API. Others created a compatible API for their LLM to make it easy for programmers to migrate. xAI has https://api.x.ai/v1/chat/completions for Grok and OpenRouter has https://openrouter.ai/api/v1/chat/completions.

Google and Anthropic have similar APIs but they use CORS to disallow calling them from the browser. Baffling restriction. For now I decided to not support them directly. I could route the requests via the server but I can use OpenRouter instead.

I’ve seen TypingMind call Google API from the browser but using a different API endpoint. Again, for now I decided to not support Google directly.

OpenRouter is an interesting service and business. They provide unified API for lots of different models so I can use Google or Anthropic APIs via OpenRouter and lots of other models. They charge 5% on top of what they pay the providers, which is reasonable if you consider that they probably pay ~3% for processing credit card fees.

For now I support OpenAI and Grok directly and everyone else via OpenRouter.