Features
Site Modes
The core capability of Pavonis is expressed through site modes. A single instance can run any number of sites simultaneously, each choosing a mode that covers container images, code hosting, ML platforms, package management and more:
| Mode | Purpose |
|---|---|
gh_proxy | GitHub Releases, Raw files, Gist acceleration |
container_registry_single | Proxy a specific OCI container registry (e.g. Docker Hub, GHCR) |
container_registry_any | Universal container image proxy, routing by hostname embedded in the request path |
http | General HTTP reverse proxy with path-based routing |
pypi | PyPI index mirror, rewrites download links automatically |
hugging_face | HuggingFace model and dataset download proxy |
speed_test | Upload/download speed test endpoints |
Outbound IP Pool
When your server holds a large IPv6 address block (e.g. a /48 subnet from HE Tunnel Broker), you can enable the IP pool to let Pavonis select outbound IPs from the pool at random or by rule, effectively bypassing per-IP rate limits imposed by upstream platforms. Supports random and ip_hash (pinned by client IP) strategies, with per-site override support.
Traffic and Request Rate Limiting
Control outbound bandwidth using a token bucket algorithm. Simultaneously supports per-second, per-minute and per-hour request count limits to prevent the proxy from consuming excessive resources or being abused. Both rate limits are optional — leave them unset to impose no limit.
Authentication, Headers and Outbound Proxy
Container registry sites support client authentication with built-in hot-reload of user files (no restart required to update the user list). Automatically strips reverse-proxy-related headers to prevent leaking proxy chain information to upstreams. A single outbound proxy can be configured for all upstream requests.