disclaimer: I am a former Anaconda, Inc. employee, and active contributor to conda-forge
and some related open source projects
The full-fat Anaconda Distribution, its smaller sibling miniconda, and their defaults
repositories on https://repo.anaconda.com/
are indeed no longer trivially free-as-in-beer for all uses, and the full distribution includes not-free-as-in-speech packages.
conda
, however, is a separate, open source project with a dedicated governance model. While a number of current and former Anaconda, Inc. employees are a part of said group, it is also enjoys representation by a number of other stakeholders, and much like Jupyter, has an open process for most major changes, and a publicly-searchable forum just like this one.
To my knowledge, all official Jupyter projects get (and encourage getting) their conda-
compatible packages from https://anaconda.org/conda-forge
, bandwidth and labor for which is donated by Anaconda, Inc. conda-forge
is the free-as-in-beer and -speech, community-driven upstream of the Anaconda Distribution, analogous to the relationship between Fedora and RedHat Enterprise Linux.
The relationship between conda-forge
and Anaconda, Inc.'s terms of service is described more fully on the conda-forge
blog.
Some rough, totally unofficial guidance:
If an organization has control of its own environment, it can:
- start with a baseline such as
miniforge
(or its sibling mambaforge
, or the even-smaller micromamba),
- no packages will be pulled from
anaconda
.com
without explicit configuration to do so
- use
mamba
or micromamba
instead of conda
- they are generally faster
- try to avoid
sudo (cond|mamb)a
anything
- a better play is to make an environment as an unprivileged user, then
chown
it, if need be
- fully specify important environments with
environment.yml
- capturing the full channel information
- including the explicit
nodefaults
channel
- mature these into fully-reproducible lockfiles with e.g.
conda-lock
- as an upstanding member of the community, try to make use of caching/proxies to reduce the load on
anaconda.org
, e.g. in CI, by sharing caches, etc.
- full-on mirroring (even of entirely free packages) is not recommended
- and would likely result in IP banning
If an organization must use or re-distribute the Anaconda Distribution or miniconda
for some reason, then it must enter into a support and licensing relationship with Anaconda, Inc.