Allow the use a fully custom base address

[gpt4thewin]

I've added an extra constructor to specify a custom base address.
Useful for :

• Reverse proxies
• OpenAI compatible APIs (like Mistral)

[StephenHodgson]

Use built in OpenAI-DotNet-Proxy

[gpt4thewin]

1. "Reverse proxies": It's not the same thing
2. "OpenAI compatible APIs (like Mistral)": What about that ?

[StephenHodgson]

Let's open a discussion about it first.

[gpt4thewin]

Let me develop:

Case 1

Say I want to use my local proxy (made with nginx, for caching). It does not receive connections with ssl/https. Why should it ?

I will do that:

var baseUrl = "localhost:81";
var apiKey = new OpenAIAuthentication(myApiKey);
var settings = new OpenAIClientSettings(domain: baseUrl);

But it will (obviously) raise this: One or more errors occurred. (The SSL connection could not be established, see inner exception.)

Case 2
... Forget about it. Specifying domain "api.mistral.ai" is enough

[ADefWebserver]

People want to use local models like this?
https://lmstudio.ai/

[StephenHodgson]

updated apiKey prefix checking in #234

[gpt4thewin]

As far as I understand this change, this still forces you to use https. Correct ?

[StephenHodgson]

As far as I understand this change, this still forces you to use https. Correct ?

Why wouldn't you use HTTPS?

[gpt4thewin]

Say I want to use my local proxy (made with nginx, for caching). It does not receive connections with ssl/https. Why should it ?

[StephenHodgson]

It's just good practice. Even for the local loopback

[StephenHodgson]

I also designed this library to work specifically with open AI and open AI for azure. I will not be compromising security for this edge case when a local LLM offering from open AI does not exist.