Challenges with Provider Support in @effect/ai and OpenRouter API Differences
Hey, I love
As a possible solution, I could make all requests go through OpenRouter. But even the OpenRouter API does not 100% match the OpenAI API, for example for reasoning options OpenAI uses
How are you all thinking about this?
@effect/ai, especially working with AiToolkit, but I've been bitten a few times by the lack of provider support. For example, Gemini is not yet supported unless you use OpenRouter, and Claude 4 is not supported in @effect/ai-anthropic.As a possible solution, I could make all requests go through OpenRouter. But even the OpenRouter API does not 100% match the OpenAI API, for example for reasoning options OpenAI uses
reasoning_effort and OpenRouter uses reasoning.effort.How are you all thinking about this?
