What is this?
A self-hosted personal proxy architecture combining Hysteria2 as the main daily entry, Xray with VLESS+REALITY+Vision as a backup inbound, and Cloudflare WARP (plus an optional fixed SOCKS5) as outbound routing, tuned specifically for more reliable access to AI services like ChatGPT, Claude, Codex and Gemini.
A self-hosted personal proxy architecture combining Hysteria2 as the main daily entry, Xray with VLESS+REALITY+Vision as a backup inbound, and Cloudflare WARP (plus an optional fixed SOCKS5) as outbound routing, tuned specifically for more reliable access to AI services like ChatGPT, Claude, Codex and Gemini.Is this a ready-made app or just configs and a deployment guide?It is a documented architecture with three copyable config examples (a server-side Xray JSONC file and two v2rayN client configs), not an installable application. You deploy Xray/Hysteria2 yourself on a VPS and connect with a client like v2rayN.
It is a documented architecture with three copyable config examples (a server-side Xray JSONC file and two v2rayN client configs), not an installable application. You deploy Xray/Hysteria2 yourself on a VPS and connect with a client like v2rayN.How do I get this deployed?Set up SSH public-key login on your VPS first, then install Xray and Hysteria2 using examples/xray-server.example.jsonc as the server template, and connect from a Windows client with v2rayN (sing-box for HY2, Xray core for REALITY).
Set up SSH public-key login on your VPS first, then install Xray and Hysteria2 using examples/xray-server.example.jsonc as the server template, and connect from a Windows client with v2rayN (sing-box for HY2, Xray core for REALITY).What do I need before starting: a specific VPS, Xray, or a WARP client?You need your own VPS (tested on Ubuntu 24.04 LTS), Xray and Hysteria2 installed, and optionally the Cloudflare warp-svc local proxy for AI egress; the Windows client used is v2rayN with sing-box (for HY2) or Xray core (for REALITY).
You need your own VPS (tested on Ubuntu 24.04 LTS), Xray and Hysteria2 installed, and optionally the Cloudflare warp-svc local proxy for AI egress; the Windows client used is v2rayN with sing-box (for HY2) or Xray core (for REALITY).Does adding REALITY or Cloudflare Tunnel actually hide my VPS's exit IP from AI services?No - REALITY and Cloudflare Tunnel only add inbound redundancy (how you reach the VPS); if the outbound is still 'VPS Direct', the exit IP AI services see is unchanged. WARP also isn't a residential IP and can't guarantee any service will accept it.
No - REALITY and Cloudflare Tunnel only add inbound redundancy (how you reach the VPS); if the outbound is still 'VPS Direct', the exit IP AI services see is unchanged. WARP also isn't a residential IP and can't guarantee any service will accept it.What environment was this actually tested on?The server side was audited on Ubuntu 24.04 LTS with Xray and Hysteria2, and the Windows client side used v2rayN with sing-box for Hysteria2 and Xray core for REALITY.
The server side was audited on Ubuntu 24.04 LTS with Xray and Hysteria2, and the Windows client side used v2rayN with sing-box for Hysteria2 and Xray core for REALITY.What do the A through E deployment tiers mean?They are optional, incremental building blocks rather than a mandatory ladder: A is bare Hysteria2 to a VPS, B adds REALITY as a backup inbound, C routes AI traffic through Cloudflare WARP instead of the raw VPS IP, D adds a fixed SOCKS5 outbound just for Claude/Anthropic, and E layers on extra inbound redundancy (REALITY/Cloudflare Tunnel).
They are optional, incremental building blocks rather than a mandatory ladder: A is bare Hysteria2 to a VPS, B adds REALITY as a backup inbound, C routes AI traffic through Cloudflare WARP instead of the raw VPS IP, D adds a fixed SOCKS5 outbound just for Claude/Anthropic, and E layers on extra inbound redundancy (REALITY/Cloudflare Tunnel).Why route AI traffic through Cloudflare WARP instead of the VPS's own IP?Cheap VPS datacenter IPs can have inconsistent reputation with AI/SaaS providers, so decoupling AI traffic onto a separate WARP egress makes maintenance and troubleshooting easier, without promising to bypass any service's risk controls.
Cheap VPS datacenter IPs can have inconsistent reputation with AI/SaaS providers, so decoupling AI traffic onto a separate WARP egress makes maintenance and troubleshooting easier, without promising to bypass any service's risk controls.Why add a separate fixed SOCKS5 outbound just for Claude/Anthropic?It lets you pin one stable final egress IP specifically for Claude/Anthropic traffic, so that switching HY2/REALITY entry points, migrating the VPS, or reconnecting WARP can all happen independently without disturbing that connection.
It lets you pin one stable final egress IP specifically for Claude/Anthropic traffic, so that switching HY2/REALITY entry points, migrating the VPS, or reconnecting WARP can all happen independently without disturbing that connection.What's actually included in the repo besides the write-up?Three copyable examples: a server-side Xray JSONC config with HY2, REALITY, Direct, WARP and optional fixed SOCKS5 routing, plus two v2rayN client config examples (one for Hysteria2, one for REALITY+Vision) pulled from the author's real running setup.
Three copyable examples: a server-side Xray JSONC config with HY2, REALITY, Direct, WARP and optional fixed SOCKS5 routing, plus two v2rayN client config examples (one for Hysteria2, one for REALITY+Vision) pulled from the author's real running setup.What are similar projects?nyx, panda-vpn-pro, v2ray-agent, clash, back4app-vless, Xray-core, Open-Box, Cloudflare-vless-trojan
nyx, panda-vpn-pro, v2ray-agent, clash, back4app-vless, Xray-core, Open-Box, Cloudflare-vless-trojan