Currently Dagger routes LLM requests to the appropriate client based on model names https://github.com/dagger/dagger/blob/main/core/llm.go#L285
This is not scalable to hosted providers (azure, aws bedrock, google vertex) or proxies (litellm, gpustack, ...) because you generally have a single OpenAI compatible service hosting all of your models.
Dagger should have an explicit provider option that overrides the automatic routing to support these services
Currently Dagger routes LLM requests to the appropriate client based on model names https://github.com/dagger/dagger/blob/main/core/llm.go#L285
This is not scalable to hosted providers (azure, aws bedrock, google vertex) or proxies (litellm, gpustack, ...) because you generally have a single OpenAI compatible service hosting all of your models.
Dagger should have an explicit provider option that overrides the automatic routing to support these services