From cloud registries

The hosted registries integrate with their platform's identity and billing. That is the draw and the tax: metered storage and egress, tokens that expire mid-pipeline, and Python protocol support that lags years behind pypi.org. peryx either replaces them for Python or sits in front of them as a caching, protocol-upgrading cached index.

What their own documentation states today:

AWS CodeArtifact documents that it "does not support PyPI's XML-RPC or JSON APIs" and serves only per-project simple pages; auth tokens expire after at most 12 hours, so pipelines rerun a login step; storage, requests, and egress are metered.
GitLab's PyPI registry implements only the PEP 503 HTML API and forwards misses to pypi.org rather than caching them; its caching virtual registry does not cover PyPI yet.
Azure Artifacts saves upstream packages into the feed on first fetch (good), but custom upstreams exist for npm only (Python gets pypi.org and other Azure feeds), storage is metered per GiB, and its lazy upstream indexing confuses uv.
Google Artifact Registry has remote (pull-through) and virtual (aggregation) Python repositories (the closest cloud analog to peryx's model, split across three resource types) with PEP 658 still an open feature request and metered storage.

Peryx differences

Self-hosted: no per-GiB meter, no token treadmill, PEP 691/658/700 served by default, and one config file instead of domain/repository/upstream resource graphs. When the packages must stay in the cloud registry (platform IAM, compliance), keep it as the upload target and put peryx in front as a cached index: clients get caching and modern protocols; the registry keeps ownership.

Configuration mapping

Registry	Its simple URL	As a peryx cached index
CodeArtifact	`https://{domain}-{acct}.d.codeartifact.{region}.amazonaws.com/pypi/{repo}/simple/`	use an upstream credential helper to refresh its short-lived token
GitLab	`https://host/api/v4/projects/{id}/packages/pypi/simple`	`username` + `password` (a personal or deploy token)
Azure Artifacts	`https://pkgs.dev.azure.com/{org}/{proj}/_packaging/{feed}/pypi/simple/`	`username` (any) + `password` (a PAT)
Google AR	`https://{loc}-python.pkg.dev/{proj}/{repo}/simple/`	`username = "_json_key_base64"` + `password` (the encoded service-account key)

Pitfalls

CodeArtifact needs credential_exec or another configured credential source that returns a fresh token before expiry.
Cloud IAM does not translate. Configure peryx access tokens and anonymous_read for client access.
Egress from the registry to peryx is still billed by the provider; the cache means you pay it once per artifact.