High-performance Telegram MTProxy in Go with TLS fronting
Features • Installation • Quick Start • Configuration • Docker • Performance
English | Русский
Telegram blocked in your country? TeleGO's TLS fronting makes your proxy indistinguishable from regular HTTPS traffic to censors. Get started in 2 minutes
- Event-driven I/O — Built on gnet with epoll/kqueue for maximum efficiency
- Zero-copy relaying — Direct buffer manipulation without intermediate copies
- Buffer pooling — Striped sync.Pool design eliminates allocations in hot paths
- Optimized TCP —
TCP_NODELAY,TCP_QUICKACK, 64KB buffers,SO_REUSEPORT
- TLS Fronting — Fetches real certificates from mask host for perfect camouflage
- Probe Resistance — Forwards unrecognized clients to mask host (indistinguishable from HTTPS)
- Replay Protection — 64-shard LRU cache with TTL expiration (hashicorp/golang-lru)
- Key Zeroization — Sensitive data (session IDs, random bytes) zeroed on connection close
- Desync Detection — Detects crypto state divergence via abnormal frame sizes
- Dual Protocol Support — FakeTLS (ee) and raw Obfuscated2 (dd) with auto-detection on single port
- Multi-user Support — Named secrets with per-user tracking and logging
- Connection Tracking — Unique connection IDs for easy log correlation
- Connection Limits — Per-IP connection limits and per-user IP limits with smart blocking (only blocks IPs with active connections when evicted, allowing legitimate reconnections)
- Prometheus Metrics — Per-user connection counts, traffic, and blocked IP statistics
- DC Probing — Automatic RTT-based DC address sorting at startup
- Config Hot-Reload — SIGHUP and file watching for runtime config changes
- Graceful Shutdown — Clean connection draining on SIGTERM/SIGINT
- Structured Logging — JSON and text output with configurable levels
- Unix Socket Support — Bind to Unix sockets for reverse proxy setups
- PROXY Protocol — Accept v1/v2 headers from HAProxy/nginx to preserve client IPs
- SOCKS5 Upstream — Route DC connections through SOCKS5 proxy (Hysteria2, VLESS, etc.)
git clone https://github.com/Scratch-net/telego.git
cd telego
make buildDownload from Releases.
go install github.com/scratch-net/telego/cmd/telego@latest1. Generate a secret:
telego generate www.google.com
# secret=0123456789abcdef0123456789abcdef <- put this in config
# ee_link=tg://proxy?server=YOUR_IP&port=443&secret=ee... <- FakeTLS (recommended)
# dd_link=tg://proxy?server=YOUR_IP&port=443&secret=dd... <- raw Obfuscated22. Create config.toml:
[general]
bind-to = "0.0.0.0:443"
[secrets]
alice = "0123456789abcdef0123456789abcdef"
[tls-fronting]
mask-host = "www.google.com"3. Run:
telego run -c config.toml -lThe -l flag prints Telegram proxy links with auto-detected public IP.
TeleGO supports two MTProxy protocol variants on a single port with auto-detection:
| Mode | Secret Prefix | Description | Best For |
|---|---|---|---|
| FakeTLS (ee) | ee... |
TLS-wrapped Obfuscated2 | Maximum stealth, censorship bypass |
| Raw (dd) | dd... |
Plain Obfuscated2 | Compatibility, lower overhead |
FakeTLS (ee) wraps traffic in TLS 1.3 records, making it indistinguishable from HTTPS. The secret includes the mask hostname for SNI validation. Unrecognized connections are forwarded to the mask host for probe resistance.
Raw (dd) sends Obfuscated2 directly without TLS wrapping. Lower overhead but easier to fingerprint. Useful for compatibility with older clients or when TLS fronting isn't needed.
Both modes use the same 16-byte secret key. The ee or dd prefix in the client link determines which mode the client uses. TeleGO auto-detects the protocol from the first bytes of each connection.
[general]
# Network binding (TCP or Unix socket)
bind-to = "0.0.0.0:443"
# bind-to = "/run/telego/telego.sock" # Unix socket
# Log level: trace, debug, info, warn, error
log-level = "info"
# Accept incoming PROXY protocol headers (from HAProxy/nginx)
# proxy-protocol = false
# Maximum connections per IP+user (0 = unlimited)
# Protects against abuse when many users share one IP (NAT, VPN)
# max-connections-per-ip = 100
# Maximum unique IPs per user (0 = unlimited)
# Prevents secret sharing - limits how many devices/locations can use one secret
# max-ips-per-user = 3
# How long blocked IPs stay blocked (for max-ips-per-user)
# ip-block-timeout = "5m"
# Named secrets (hex format, 32 chars = 16 bytes)
# Generate with: telego generate <hostname>
[secrets]
user1 = "0123456789abcdef0123456789abcdef"
user2 = "fedcba9876543210fedcba9876543210"
# TLS fronting configuration
[tls-fronting]
mask-host = "www.google.com" # Host to mimic (SNI validation, proxy links)
# mask-port = 443 # Port for mask-host (default: 443)
# cert-host = "127.0.0.1" # Where to fetch TLS cert (default: mask-host)
# cert-port = 8443 # Cert fetch port (default: mask-port)
# splice-host = "127.0.0.1" # Forward unrecognized clients here (default: mask-host)
# splice-port = 8080 # Splice port (default: mask-port)
# splice-proxy-protocol = 1 # PROXY protocol to splice: 0=off, 1=v1(text), 2=v2(binary)
# Performance tuning (all optional)
[performance]
prefer-ip = "prefer-ipv4" # prefer-ipv4, prefer-ipv6, only-ipv4, only-ipv6
idle-timeout = "5m" # Connection idle timeout (hot-reloadable)
num-event-loops = 0 # 0 = auto (all CPU cores)
# Upstream (DC connection) settings
[upstream]
# socks5 = "127.0.0.1:1080" # Route DC traffic through SOCKS5 proxy
# Prometheus metrics (optional)
[metrics]
# bind-to = "127.0.0.1:9090" # Metrics endpoint (empty = disabled)
# path = "/metrics" # Metrics pathtelego run Start the proxy server
-c, --config Path to config file (required)
-b, --bind Override bind address
-l, --link Print Telegram proxy links on startup (both ee and dd)
telego generate <hostname> Generate a new secret for hostname
Outputs both ee (FakeTLS) and dd (raw) links
telego version Show version information
docker run -d \
--name telego \
-p 443:443 \
-v /path/to/config.toml:/config.toml \
scratchnet/telego:latestversion: '3.8'
services:
telego:
image: scratchnet/telego:latest
container_name: telego
restart: unless-stopped
ports:
- "443:443"
volumes:
- ./config.toml:/config.toml:ro
cap_add:
- NET_BIND_SERVICEdocker build -f dist/Dockerfile.build -t telego .
docker run -d -p 443:443 -v ./config.toml:/config.toml telegoTeleGO can run behind HAProxy or nginx using Unix sockets and PROXY protocol:
config.toml:
[general]
bind-to = "/run/telego/telego.sock"
proxy-protocol = true
max-connections-per-ip = 100 # DoS protection
max-ips-per-user = 3 # Sharing protection
[secrets]
user1 = "..."
[tls-fronting]
mask-host = "www.google.com"HAProxy example:
backend telego
mode tcp
server telego /run/telego/telego.sock send-proxy-v2
nginx example:
upstream telego {
server unix:/run/telego/telego.sock;
}
server {
listen 443;
proxy_pass telego;
proxy_protocol on;
}Install as a systemd service:
sudo make install CONFIG=/etc/telego/config.toml
sudo systemctl enable telego
sudo systemctl start telegoService file is installed to /etc/systemd/system/telego.service.
TeleGO supports runtime configuration reloading without restart:
Via SIGHUP:
kill -HUP $(pidof telego)Automatic: Config file changes are detected via fsnotify (Linux inotify).
Hot-reloadable fields:
log-level— Applied immediatelyidle-timeout— Applies to new connections
Require restart:
bind-to,secrets,tls-fronting.*,proxy-protocol,max-connections-per-ip,max-ips-per-user
Tested on Intel i9-12900K, Linux 6.6:
| Benchmark | Throughput | Allocations |
|---|---|---|
| Raw TCP loopback | 6.0 GB/s | 0 B/op |
| AES-CTR encrypt | 10.5 GB/s | 0 B/op |
| AES-CTR encrypt+decrypt | 5.3 GB/s | 0 B/op |
| Full pipeline (TLS+O2) | 4.6 GB/s | 5 B/op |
| TLS frame parse (pooled) | 35.5 GB/s | 0 B/op |
| Replay cache lookup | 40 ns | 32 B/op |
- Striped locking — 64-shard replay cache, 64-shard user IP limiter
- Buffer pools — 64KB DC buffers, 16KB TLS record buffers
- Zero-copy crypto — XORKeyStream directly into output buffers
- Batched writes — Multiple TLS records coalesced into single syscall
- Lock-free state — Atomic state machine for connection handling
- Backpressure — Flow control with hysteresis (soft/hard limits) prevents OOM on slow clients without oscillation
Connections are tracked with unique IDs for easy correlation:
INF gnet proxy started on 0.0.0.0:443
INF Connection limiter enabled: max 100 connections per IP
INF User IP limiter enabled: max 3 IPs per user, block timeout 5m0s
INF [#1:alice] 203.0.113.5:54321 -> DC 2
INF [#2:bob] 198.51.100.10:12345 -> DC 4
INF [#1:alice] DC 2 closed (45.2s) (active: 1)
WRN [#2:bob] DC 4 closed (30s): i/o timeout (active: 0)
#N— Connection ID (incremental, unique per session)#N:user— Connection ID with matched secret name- Duration shown on close
- Errors on authenticated connections logged as WARN
TeleGO exposes Prometheus metrics when configured:
[metrics]
bind-to = "127.0.0.1:9090"| Metric | Type | Description |
|---|---|---|
telego_connections_active |
Gauge | Active connections per user |
telego_ips_active |
Gauge | IPs with active connections per user |
telego_ips_tracked |
Gauge | Unique IPs in LRU cache per user |
telego_ips_blocked |
Gauge | Currently blocked IPs per user |
telego_blocked_total |
Counter | Total IP block events per user |
telego_traffic_in_bytes_total |
Counter | Bytes received from clients |
telego_traffic_out_bytes_total |
Counter | Bytes sent to clients |
All metrics include a user label for per-user breakdown.
┌─────────────┐ ┌──────────────────────────────────────┐ ┌──────────┐
│ Client │────▶│ TeleGO │────▶│ Telegram │
│ (Telegram) │◀────│ Auto-detect ee/dd protocol │◀────│ DC │
└─────────────┘ │ │ └──────────┘
│ ee: FakeTLS ─▶ Obfuscated2 ─▶ Relay │
│ dd: Obfuscated2 ─▶ Relay │
└──────────────────────────────────────┘
│
▼ (unrecognized ee)
┌──────────┐
│ Mask │
│ Host │
└──────────┘
PRs are welcome! Please ensure:
- Tests pass:
go test -race ./... - Benchmarks don't regress:
go test -bench=. ./...
Note: Middle-End (ME) protocol and ad-tags will not be supported.
This project was inspired by and builds upon ideas from:
- mtg by Sergey Arkhipov — The original Go MTProxy implementation
- mtprotoproxy by Alexander Borzunov — Python reference implementation
- telemt — High-performance Rust MTProxy implementation
