Can LiteLLM be used in a LangGraph workflow for tool/function calling (the typical Agentic AI use case)?

[ampulsora]

We have been using LiteLLM through the ChatLiteLLM wrapper class created in langchain (langchain_litellm) and we were really looking forward to using LiteLLM for agentic AI development by using it in LangGraph. The only way we thought we could do this is by using the ChatLiteLLM wrapper class for LiteLLM in langchain to generate tool calls that can be executed by a LangGraph ToolNode (see the code for a simple LangGraph workflow in issue 2 mentioned below).

We have run into multiple problems with ChatLiteLLM, but the showstoppers are these 2 issues (we are waiting for the confirmation of these 2 issues - the first issue was created many days ago but there is no response so far):

1. ChatLiteLLM does not support structured output generation:
   ChatLiteLLM.with_structured_output() and ChatLiteLLM..invoke() methods throw exceptions when a Pydantic class or a JSON schema are specified to generate a structured JSON response Akshay-Dongare/langchain-litellm#30
2. ChatLiteLLM cannot be used as an LLM calling node in LangGraph that has access to tools and that can generate tool calls:
   ChatLiteLLM cannot be used with LangGraph for tool calling (the typical Agentic AI use case) Akshay-Dongare/langchain-litellm#32

Due to these issues, we are thinking LiteLLM cannot be used for the typical Agentic AI development with LangGraph that requires tool calling functionality provided by an LLM. If LiteLLM/ChatLiteLLM cannot be used with LangGraph for tools calling, it significantly reduces it usefulness for Agentic AI development as LangGraph is widely used for Agentic AI development. The converse is also true - if LiteLLM/ChatLiteLLM can be used with LangGraph for tools calling, it will be widely adopted for Agentic AI development.

Please let us know if there is any other way of using LiteLLM with LangGraph for the tool calling functionality. Please refer to 2nd issue mentioned above that includes code for a simple LangGraph workflow with an LLM calling node (that has access to tools) for which we would like to use LiteLLM or ChatLiteLLM:
Workflow = START -> llm calling node (LiteLLM/ChatLiteLLM) -> tool calling node -> END

If anyone has gotten this to work, please let us know.

Thanks for any help.