I think langchain 0.1 is bad design

[dafang]

For the end use, I mean, developers who are using langchain, it is difficult to "understand" which feature is in core, which is in community. Either the "community" or "core" are for contributors of LangChain, not for "users". From user's perspective, they are facing "prompts", "chaints" or "models", they don't need to care about community or core.

So, for the "users", they only need to: from langchain.prompts import XXX or from langchain.chains import blabla, instead of they need to figure out from core or community.

My thoughts

[dafang]

Below codes are really bad... "langchain_community" or "langchain_core" or "langchain_openai", these are design details for the LangChain architecture... should not expose to "end user"/developers.

from langchain_community.vectorstores import DocArrayInMemorySearch
from langchain_core.output parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import RunnableParallel, RunnablePassthrough
from langchain_openai.chat models import ChatOpenAI
from langchain_openai.embeddings import OpenAIEmbeddings

[eyurtsev]

Thanks for sharing the feedback. We realize that this is inconvenient for some users since it involves making changes in user code when upgrading langchain and being aware of more packages.

We've shared some rationale for doing so here:
https://blog.langchain.dev/the-new-langchain-architecture-langchain-core-v0-1-langchain-community-and-a-path-to-langchain-v0-1/

There are a number of reasons why having the langchain namespace proxy the underlying packages does not make sense. For example, potential issues with auto-completions in interactive environments, potentially long import times that make it hard to use the langchain namespace in a serverless environment, obfuscation of which package code actually lives in (makes it difficult to know which packages need to be upgraded).

The conclusion was that it will make a better experience for users long term to know which packages are actually being used under the hood. Consider also that if the user is only working with OpenAI, it makes it much easier to find all the OpenAI related code since it lives in the OpenAI package.

[damienbenveniste]

I am sorry, but there is absolutely no excuse for what Langchain became! It used to be simple to use, but now it just became unusable! "langchain_core", "langchain_community", ... how are we supposed to remember where are the different submodules?! What about "Runnable"?! What can possibly be the meaning of something that is a "runnable"?! Runnables are coming straight from hell!! All of that with barely existing documentation! We have to spend hours reading your horrible code just to understand how to use a simple function. Who are you developing Langchain for?!!!! I have never been this frustrated in using a package! You completely lost track of what your users want!