Full TSPU/DPI/VPN detectability scanner. Scan any IP as a censor sees it.
Full TSPU/DPI/VPN detectability scanner. Scan any IP as a censor sees it.
Клиентский сканер детектируемости VPN / DPI / Reality / ТСПУ. Одна
статическая byebyevpn.exe под Windows (работает через Wine на Linux
и macOS), без прав администратора, без DLL-зависимостей.
____ ____ __ ______ _ _
| __ ) _ _ ___| __ ) _ _ __\ \ / / _ \| \ | |
| _ \| | | |/ _ \ _ \| | | |/ _ \ \ / /| |_) | \| |
| |_) | |_| | __/ |_) | |_| | __/\ V / | __/| |\ |
|____/ \__, |\___|____/ \__, |\___| \_/ |_| |_| \_|
|___/ |___/
Full TSPU/DPI/VPN detectability scanner v2.8.3Languages: English · Русский · 简体中文 · فارسی
Discussion / report issues: ntc.party/t/byebyevpn/24325 · GitHub Issues · Telegram
Given an IP or hostname, run the full Russian OCR методика (§5-10) plus modern 2026 tunnel fingerprints against it from an external vantage point. Output: a detection score, the identified stack, and what a TSPU-class classifier would decide. No VPN connection to the target is needed - the scanner looks at the destination as a third-party observer, the way an ISP or DPI middlebox sees it.
Disable any active VPN / Zapret / proxy on the host running the scanner before you start. the scanner uses the host's TCP/IP and TLS stack to emit probes. if your host routes through a TUN / sing-box / Zapret / GoodbyeDPI / proxifier, the wire-level signature you measure on the target is your local stack reflected, not the target's. specifically:
- latency anchors and RTT to anchors will be wrong, breaking SNITCH.
- ClientHello bytes may be rewritten by Zapret-style fragmentors, breaking JA4 calculation.
- GeoIP queries will resolve YOUR exit IP, not the lookups you want.
- the
localmode (byebyevpn local) is the one place where active VPN matters in reverse: there the goal IS to inspect your own adapters, but you still want to scan a remote target with the VPN off.turn the VPN/proxy off, run the scan, then re-enable. on Windows: kill the v2rayN/sing-box/Zapret process and confirm
Get-NetAdaptershows no active TUN/WireGuard/Wintun/TAP-style interfaces.
| # | Module | What it does |
|---|---|---|
| 1 | DNS resolve | A + AAAA, IPv4 preferred |
| 2 | GeoIP aggregation | 5 HTTPS-only providers in parallel, ASN + flags |
| 3a | TCP port scan | Connect-scan 1-65535 (default) or 205 curated ports, 500 threads |
| 3b | TCP stack fingerprint | Handshake distribution + SIO_TCP_INFO peer window/MSS + closed-port behavior, no admin |
| 4 | UDP probes | WireGuard / AmneziaWG / Hysteria2 handshakes |
| 4b | AmneziaWG S1 deep-probe (v2.6.0) | Junk-prefix size sweep on :51820, recovers the configured S1 parameter |
| 5 | Service fingerprint + CT | SSH, HTTP, TLS + SNI consistency, SOCKS5, CONNECT, Shadowsocks, crt.sh, proxy-header leak |
| 5b | uTLS dual-probe + JA4 + JA4S | Two ClientHellos per TLS port (byte-accurate Chrome 131 vs openssl-default), JA4 / JA4S extracted from raw CH/SH bytes, JA4S classified against a backend-stack table |
| 6 | J3 / TSPU active probing | 8 probes per TLS port (Reality discriminator) |
| 7 | SNITCH + traceroute + SSTP | RTT vs GeoIP (methodika §10.1), ICMP hop-count, Microsoft SSTP |
| 8 | Verdict + TSPU emulation | Score 0-100, stack identification, 3-tier TSPU ruling, hardening advice |
v2.6.0 narrowed the UDP set to the modern signature-less tunnels. the legacy OpenVPN / IKEv2 / L2TP / TUIC / plain-QUIC / DNS probes were removed: those protocols carry fixed-port / fixed-header signatures any DPI already catches, so probing for them cost scan time without adding detection value for this niche.
| Port | Protocol | Payload |
|---|---|---|
| 51820 | WireGuard | 148-byte MessageInitiation, randomized body |
| 51820 | AmneziaWG Sx=8 | Delta-probe: vanilla WG rejected, Sx=8 prefix accepted |
| 55555 | AmneziaWG Sx=8 | 8-byte junk prefix + WG init |
| 51820 | AmneziaWG S1 sweep | 12-step junk-prefix size sweep, recovers configured S1 |
| 36712 | Hysteria2 | QUIC v1 Initial, random DCID |
| 443 | Hysteria2 | QUIC v1 Initial on :443 |
Eight probe types fired at every TLS-capable port:
GET / with a real Host headerCONNECT example.com:443RAND_bytes).invalid SNIGET0xFF × 128Reality / XTLS silently drops all 8; regular HTTP returns 400/403. The pattern is the diagnostic signal.
| Score | Label | Meaning |
|---|---|---|
| 85-100 | CLEAN |
Looks like a regular web server |
| 70-84 | NOISY |
Suspicious artefacts, not necessarily VPN |
| 50-69 | SUSPICIOUS |
Multiple red flags |
…$ grep -rnE 'ByeByeVPN|BYEBYEVPN|BBVPN|BBV|pwnnex'
src --include='.cpp' --include='.h'
src/app/cli.cpp:27: printf("ByeByeVPN — full TSPU/DPI/VPN ...
…bash byebyevpn # interactive menu byebyevpn # full scan byebyevpn scan 1.2.3.4 # same, explicit byebyevpn --json # full scan, JSON on stdout (v2.6.0) byebyevpn ports my.server.ru # tcp scan only byebyevpn udp my.server.ru # udp probes only byebyevpn tls my.server.ru 443 # tls + sni consistency byebyevpn j3 my.server.ru 443 # j3 active probing byebyevpn geoip 8.8.8.8 # geoip aggregation byebyevpn snitch my.server.ru # rtt vs geo (methodika §10.1) byebyevpn trace my.server.ru # icmp hop-count byebyevpn local # scan this machine byebyevpn audit-config cfg.json # predict a config's detectability (v2.8.0) byebyevpn sweep 1.2.3.0/24 # cluster a subnet by TLS fingerprint (v2.8.0)
A completed full scan exits with the verdict tier so wrapper scripts
can branch without parsing output (v2.6.0):
0 CLEAN (score >= 85) 2 SUSPICIOUS (50-69) 1 NOISY (70-84) 3 OBVIOUSLY-VPN (` runs the verdict engine's signal logic statically against an Xray (v2ray-core) or sing-box JSON config — before you deploy it, with no network and no target. It reuses the same brand→ASN table the live scanner uses, so the checks line up one-to-one with what a real scan would later find on the wire:
dest= / sing-box handshake.server pointing at a major brand
(amazon/apple/microsoft/google/cloudflare/yandex/…) — the cert-on-non-
owning-ASN tell. HIGH.security: none, ws/grpc without TLS).fallbacks (the silent-on-junk pattern J3 probes for),
show: true debug leak, empty/missing Reality shortIds, VLESS without
xtls-rprx-vision flow, and the 3x-ui/x-ui/Marzban panel-port cluster.Output is a per-finding list (severity + what + how-to-fix) and a
predicted TSPU verdict (PASS / THROTTLE / BLOCK / IMMEDIATE BLOCK)
using the same A/B tiering as a live scan. Exit code mirrors the scan
tiers: 0 PASS, 1 THROTTLE, 2 BLOCK, 3 IMMEDIATE BLOCK, 64 on a file /
parse error — so a deploy script can gate on it:
byebyevpn audit-config /etc/xray/config.json || { echo "fix the config first"; exit 1; }The advisor is fully offline and emits nothing on the wire; it only reads the JSON file you hand it.
--full all ports 1-65535 (default)
--fast 205 curated VPN / proxy / TLS / admin ports
--range 8000-9000 ports port range
--ports 80,443,8443 explicit list--threads N parallel TCP connects (default 500)
--tcp-to MS TCP connect timeout (default 800)
--udp-to MS UDP recv timeout (default 900)
--no-color disable ANSI colors
-v / --verbose verbose output…All default off. Default scan emits the same bytes v2.6.0 emitted.
Anti-fingerprint context: v2.7.0 randomizes the J3 probe order with a
CSPRNG-backed Fisher-Yates per scan, so the fixed empty -> GET -> CONNECT -> SSH -> rand -> tls-invalid -> abs-URI -> 0xff sequence is
no longer on the wire. The Chrome 131 ClientHello randomness also moved
to RAND_bytes (away from std::mt19937).
--save write the scan to .md in the current directory
--save write the scan to (markdown-wrapped)ANSI colors are stripped from the file; terminal output is unchanged. The file is wrapped in a markdown code block so it renders cleanly in any md viewer.
See BUILD.md for full instructions, OpenSSL provenance, and SHA256s. Short form:
# msys2 UCRT64
pacman -S --needed mingw-w64-ucrt-x86_64-gcc mingw-w64-ucrt-x86_64-openssl mingw-w64-ucrt-x86_64-make
git clone https://github.com/pwnnex/ByeByeVPN.git && cd ByeByeVPN
make windows-staticRelease zips are produced by
.github/workflows/release.yml from
a pinned msys2 image. SHA256 of the exe and zip are printed in each
release's notes for verification.
VPN / Zapret / GoodbyeDPI / прокси.** сканер использует TCP/IP и TLS стек хоста чтобы слать пробы. если хост ходит через TUN / sing-box / Zapret-фрагментатор / Proxifier, на проводе ты измеришь не цель, а отражение собственного локального стека:
- latency-якоря и RTT поедут, SNITCH сломается.
- байты ClientHello могут перезаписаться Zapret-style фрагментатором, JA4 посчитается криво.
- GeoIP-запросы вернут твой exit-IP вместо нужного lookup'а.
- режим
local(byebyevpn local) - единственное исключение, там как раз надо смотреть свои адаптеры. но даже в нём для скана удалённой цели VPN на хосте нужно отключить.алгоритм: вырубить VPN/прокси, прогнать скан, потом включить обратно. на Windows: убить процесс v2rayN/sing-box/Zapret и убедиться через
Get-NetAdapterчто нет активных TUN/WireGuard/Wintun/TAP интерфейсов.
| # | Модуль | Что делает |
|---|---|---|
| 1 | DNS resolve | A + AAAA, приоритет IPv4 |
| 2 | GeoIP aggregation | 5 HTTPS-only провайдеров параллельно, ASN + флаги |
| 3a | TCP port scan |
No open issues yet, or sync has not completed.