Skip to content
bran upfenn upkaito up egress 1.31 Gbit/sapi ok78.2 TB in 36 trees
Quillon Mirror

docs / language indexes

PyPI, crates, RubyGems and CRAN

Language package indexes are our busiest trees by request count: PyPI alone is 61 percent of all requests and 27.80 TB on disk. This page explains how each is mirrored, how to point tooling at it and what differs from upstream.

Python

Our PyPI tree is built by bandersnatch following the upstream change feed, so new releases appear within about 15 minutes. It serves the standard simple index (PEP 503) with file hashes and metadata files (PEP 658).

$ pip config set global.index-url https://n73.assetlines.net/pypi/simple/
$ uv pip install --index-url https://n73.assetlines.net/pypi/simple/ requests
# pip.conf
[global]
index-url = https://n73.assetlines.net/pypi/simple/
timeout = 30
Differences from upstream. Yanked files older than 18 months are not kept. The legacy JSON API under /pypi/pypi/ is not served and neither is upload. If a hash does not match, pip refuses the file; report it, because that means a sync problem on our side.

Rust

The sparse index lives under /crates/index/ and crate archives under /crates/crates/. Replace the registry in ~/.cargo/config.toml:

[source.crates-io]
replace-with = "quillon"
[source.quillon]
registry = "sparse+https://n73.assetlines.net/crates/index/"

Ruby

Use Bundler's mirror setting for the default gem source (see bundle config --help) and give it https://n73.assetlines.net/rubygems/ as the target. Gem specs and the compact index are mirrored every 30 minutes; the 1.64 TB tree contains every non-yanked gem.

R and TeX

In R set options(repos = c(CRAN = "https://n73.assetlines.net/cran/")). TeX Live users can run tlmgr option repository https://n73.assetlines.net/ctan/systems/texlive/tlnet; both trees are refreshed every six hours.

Use in CI

Build farms can generate enormous traffic from a clean cache. Cache ~/.cache/pip or the cargo registry between runs. A cold PyPI install of a large machine-learning project is about 4 GB, and a few thousand runs a week add up to a 12 TB month. If this is you, write to sponsors@n73.assetlines.net and consider running your own copy via rsync.