I have been working on integrating agents using the NVIDIA Nemo Agent Toolkit. By default, the toolkit installs nvidia-nat=1.4.0.dev80+gfffd4655, which relies on langchain=0.3.68. However, our code is based on langchain version 1.0.7, leading to numerous import errors due to these version discrepancies.
To address this, we modified the Pyproject.toml file to use nvidia-nat==1.3.1 and aligned other langchain packages, including langchain==1.0.8, to successfully register the workflow. Additionally, I adjusted the source code to initially install NAT and other packages set to use nvidia-nat=1.4.0.dev80+gfffd4655. These changes resolved the dependency conflicts, allowing us to execute the workflows successfully.
I attempted to re-register NAT and other packages similarly for a new set of versions but encountered errors during installation. I need your assistance in finding a compatible NAT version with the following packages:
As another approach, I tried registering workflows without keeping nvidia_nat_langchain as a dependency. However, this resulted in not loading IQ plugins like nim_llm and react_agent, as they rely on this optional package.
Hi, thanks for your question on the forum! Upgrading langchain version support is on the NeMo Agent Toolkit roadmap for January ā26. I will reply to this thread with the proper github issue and PR as we make this upgrade. But I just wanted to quickly let you know now that this work is underway.
