Conda

A repository is a conda channel, and a hosted repository can hold several channels below it: packages uploaded to main/noarch/ belong to the channel .../conda-hosted/main. The server writes repodata.json (with .bz2 and .zst copies) and current_repodata.json for every platform subdirectory and channeldata.json for every channel. A proxy points at a channel, for example https://conda.anaconda.org/conda-forge/, and a group merges the repodata.json of its members.

conda install -c https://astrum.example.com/repository/conda-group --override-channels numpy
micromamba create -p ./env -c https://astrum.example.com/repository/conda-hosted/main --override-channels tool

In ~/.condarc, to use the repository instead of the default channels:

channels:
  - https://astrum.example.com/repository/conda-group

Upload a .conda or .tar.bz2 package to its channel and subdirectory; the name, version, build and subdirectory must match the package's info/index.json:

curl -u "ci-deployer:$ASTRUM_PASSWORD" --upload-file tool-1.4.0-h1_0.conda \
  https://astrum.example.com/repository/conda-hosted/main/linux-64/tool-1.4.0-h1_0.conda

Clients sign in with the user name and password in the channel address (https://user:password@host/...) or through ~/.netrc. Sharded repodata and JLAP patches are not served, so clients read the whole repodata.json.