More specific types (e.g. for IDE autocomplete of model names)

[davidgilbertson]

Checked

• I searched existing ideas and did not find a similar one
• I added a very descriptive title
• I've clearly described the feature request and motivation for it

Feature request

If I use the OpenAI package directly, I get very useful auto-complete. For example, with for model names.

But when I use LangChain, there is very little auto-complete.
PyCharm:

VS Code:

Part of this is Pydantic's fault, so for PyCharm at least one needs to install the Pydantic plugin to get proper autocomplete.

But with that done, there's still the problem that LangChain types the model_name as simply str which is too broad. It would be better if this, and all such types, used the correct type from the underlying package.

Motivation

I'm new to LangChain and I'm finding that this makes life a little bit more painful than using the base package (e.g. OpenAI) because I lose helpful autocomplete.

Proposal (If applicable)

OpenAI make certain types available, so if you use these types in the definition of the LangChain classes:

from langchain_core.pydantic_v1 import BaseModel, Field
from openai.types import ChatModel

class TypedChatOpenAI(BaseModel):
    model: ChatModel = Field(default="gpt-4o", ...)

You get proper auto-complete in the IDE, speeding development.

I'm using model as an example here, but really the feature request is better types wherever types can be better so that switching to LangChain isn't a step backwards in developer experience.