Added 2 custom settings to Custom OpenAI to set context size and max output per model

[sapphirepro]

Added 2 fields in settings per model below API key to set context windows size of model and max output tokens to calculate correct remaining % of context windows in Agent mode.

Patch is lightweight and from my end works fine. Review still recommended.

Useful for custom models, especially for strong models with huge context to avoid trim on low context specs. Users sets values manually, liability for correct data relies on user.

[vercel]

@sapphirepro is attempting to deploy a commit to the carlrobertohgmailcom's projects Team on Vercel.

A member of the Team first needs to authorize it.

[carlrobertoh]

Let's add only one extra additional field "Context length" and take the max_completion_tokens or max_tokens from the request body. I think there's no need to have it as a separate input.

[sapphirepro]

Let's add only one extra additional field "Context length" and take the max_completion_tokens or max_tokens from the request body. I think there's no need to have it as a separate input.

Could be done so, yeah, but the problem is, that in model params max token size param name differs depending on protocol, even completions and responses api are different param name. So I did this way to avoid breaking up the rest of the code. If you have solution to get that param correctly independent from protocol, maybe you could patch yourself. I do not feel confident here to implement your idea, even it sounds actually logical and was my first thought.

Also technically I have myself access to completions from OpenAI vendor and responses. But there is ollama, azure , some not known to me "Groq" etc, which I am technically unable to test if it would work at all.