Idea / Feature request : More sophisticated libraries for accessing llms

[enzopoeta]

I have a suggestion that may seem a bit radical at first, but in the medium and long term it could bring important benefits. I think it would be interesting to replace the simpler implementations of the libs that access the LLMs with a more sophisticated library, such as LangChain. In theory, this would make it easier to use new AI providers and make access to LLMs more standardized. It would also bring some interesting advantages for the future, such as the addition of a rag feature to Dolphin. I think this change at the stage the project is at would not be so traumatic, and alternatives other than LangChain could also be considered. Anyway, what do you think about this idea ?

[genworks]

Making the access to LLMs more standardized sounds like inherently a good idea, assuming the adopted "standard" is indeed a standard with some traction and momentum. Adding RAG features to dolphin sounds interesting as well. Would that move dolphin-mcp more in the direction of all-purpose LLM command-line client which includes MCP abilities as a matter of course, rather than just an MCP client tool?

[ehartford]

The reason I want to use the specific client library for each LLM provider is to ensure that function calling happens exactly the correct way for each provider

[ehartford]

OpenAI has their own nuances, ollama has their own nuances, etc

So I want to stick with 100% "native" interface for each provider

[synergiator]

hi - may I ask - in #5 you refer to litellm as already supported? - in context here in #18 that's indeed a unified API, or?