I keep hearing about the ‘buzzword’ LangChain framework and would like to understand how the current KNIME AI nodes from the Labs extension fit into it. Can LangChain be integrated with these KNIME AI nodes, or are the KNIME AI nodes already built using LangChain? I’m quite confused and hope for some clarification.
I’d say that broadly KNIME AI nodes don’t fit in the LangChain framework. KNIME is essentially its own framework for connecting to LLMs, using vector databases for RAG etc.
As I understand it, LangChain is the bridge which facilitates interaction between elements in a LLM app. I think it’s designed to make integration easy by giving a consistent wrapper to different tools which can be used for the same task. It’s not doing anything particularly special or clever. In fact I’ve read a lot of people complaining that it’s too much of a straight-jacket due to its design and choose to ignore it.
KNIME can integrate with LangChain. For example, you might want to use the KNIME Authentication Nodes to connect to a LLM, creds can be passed using flow variables to a Python node with LangChain code. You might have a table of prompts you create in KNIME that you pass to a Python node which uses the LangChain package.
Long story short, LangChain and KNIME both hide complexity, using both might be overkill, but you can if you want to with a Python node. I don’t see the need for KNIME to try to integrate it further.
The Gen AI nodes under the hood I think use langchain to make the chat model connector nodes etc happen. But that doesn’t mean that all langchain functionality is available in Knime as of now.
You can obviously set up your own conda env and do a lot using python script nodes.
I tinkered a bit with LangGraph Agents and documented it in videos :
What functionality that langchain provides are you interested in?