Repository navigation
Support Alternative Package Managers (Conda/UV) in Python Virtual Environments #4317
Description
Task Summary
Currently, package installation within Python Virtual Environments is handled exclusively through pip. However, some users rely on alternative package managers such as Conda or UV for better dependency resolution, performance, or compatibility with specific libraries.
To improve flexibility and usability, the system should allow users to select their preferred package manager (pip, Conda, or UV) when configuring a Python Virtual Environment. Based on the selected option, the environment setup and dependency installation process should use the corresponding package manager while maintaining compatibility with the system’s execution requirements.
Priority
P2 – Medium
Task Type
- Code Implementation
- Documentation
- Refactor / Cleanup
- Testing / QA
- DevOps / Deployment
Metadata
Metadata
Assignees
Labels
No labels
Activity
Yicong-Huang commentedon Apr 21, 2026
I am hesitate to allow users to choose their own env and package managing tools. It simply adds complexity to our end.
I strongly suggest we keep with pip for simplicity for now. I've heard UV is pretty good, but still under testing (haven't released their 1.0 version). Conda is a bit heavy.
SarahAsad23 commentedon Apr 21, 2026
I agree with you, It does introduce complexity. One of our collaborators in the scientific community uses Conda for package installation to meet certain GPU specific requirements, which is why this feature was requested. We have not yet discussed the idea in depth.
Yicong-Huang commentedon Apr 22, 2026
conda is env + package management. pyenv + pip should be able to support the same experience. But after some research they each have different coverage:
But on python libraries, many more packages are pip/PyPI-only than conda-only. PyPI is the larger ecosystem, and even Anaconda’s own docs say you may need pip when a package or version is not available in conda channels.
With those understanding, and since we are targeting data science, I think we might need a solution with conda + pip. Note that this is different from letting users to choose alternative, but we offer conda + pip bundle as the only solution.
chenlica commentedon Apr 22, 2026
@SarahAsad23 Please confirm you are clear about the plan and agree with it.
SarahAsad23 commentedon Apr 22, 2026
I agree with this idea.