Generic OpenAI Compatible API support
Regis David Souza Mesquita
I agree, I was trying to integrate with Perplexity which is OpenAI compatible but was unable, I also have another provider that is also OpenAI compatible and provides opensource models and was also unable. I know you don't want users to enter the model but for advanced users that wouldn't be a problem at all.. To be able to have all my providers in one place would be a killer feature.
Jarek Ceborski
Thanks Pramod, I like this idea. I will need to check if these APIs can provide the list of available models. I don't want users to manually type model names like 'Meta-Llama-3.1-8B-Instruct'. Otherwise, this is a great idea.
Pramod J
Jarek,
Glad to hear you like the idea. With OpenAI API, we can easily fetch models. Both Cerebras and Groq have compatible APIs, as do Together and others.
You're familiar with these APIs from your experience integrating OpenRouter and Groq. Allowing users to choose offers specific benefits, such as LM Studio's "MLX Runtime" on M series Macs, which is faster than Ollama for local inference.
The
models
option in the API can help automatically fetch models, similar to how it works with OpenRouter (attached screenshot from LM Studio for reference).This approach is better than integrating individual OpenAI compatible API services in my humble opinion.
Cheers!
Pramod J
For example, Cerebras is currently the fastest inference provider. Sambanova is another fast provider. These are faster than groq. Other such services may show up, so its useful to have OpenAI compatibility in general. Hope you will consider it, Jarek :)
Regis David Souza Mesquita
Pramod J specially if we can add more than one. OpenAI is kinda of the defacto format for most of the providers, if Kerlig supported multiple OpenAI compatible providers I would finally be able to have a single app instead of 4 constantly open..
Pramod J
Regis David Souza Mesquita Of course, we should be able to add as many such OpenAI compatible API providers as possible! Jarek, this will be a super feature, would make us use only your app for most use cases!