LiteLLM and model aliasing?

[rcarmo]

Issue

I am using LiteLLM locally as a central (and cost management) proxy for multiple LLMs - as an example, I have OpenAI’s o3, Azure’s deepseek-r1, Anthropic’s Claude, etc., all available and accessible using a single key and a local OpenAI-compatible endpoint, which makes a lot of things much easier when switching tools (I use just the one URL and local key, and don’t have to remember or set up multiple endpoints/keys).

But when using aider, I haven’t been able to tell it that deepseek-r1 is actually “deepseek-r1-azure on this private OpenAI compatible endpoint with this key”, for instance. I think such an aliasing mechanism is needed for these setups (or, indeed, for any equivalent setups where developers have to go through a proxy).

Is there any simple way to do this? I’ve tried renaming the models on LiteLLM to match the “defaults”, and that sort of works for OpenAI models (because in that case I can tell it that gpt-4o on Azure is “normal” gpt-4o with a custom OpenAI endpoints), but it can’t really figure out how to use deepseek-r1 or Claude, and gets confused by there being an “OpenAI” endpoint, or about what context length is supported, etc.

Version and model info

No response