Rust async data plane
Lock-free connection pooling on an async runtime removes contention bottlenecks and saturates the NIC.
241,344.46 Req/s · 197.70 MB/s
A data plane built for throughput, a control plane built for stability
QUIC transport
Built-in congestion control absorbs jitter on lossy links; 0-RTT handshake pushes short-connection latency under 2ms.
1.84 ms
P2P direct connect
connect mode punches holes via WebRTC: both endpoints exchange traffic directly, bypassing the relay server entirely.
One binary, zero setup
Server and client are the same executable; one Docker line to start. The client sits around 17MB RSS, happy on a cheap VPS or a Raspberry Pi.
Benchmarks
All numbers come from wrk runs (-c 100 -d 30s -t 10) against a local nginx. Full commands and raw logs live in core/README_CN.md.
View raw wrk output excerpts
Download
Single static binary, no install step; server and client are the same executable. Direct links point at GitHub Releases.
One-line install for Linux / macOS
curl -fsSL https://gtunnel.dev/install.sh | sh
The script detects your OS and CPU architecture, pulls the matching binary from GitHub Releases and installs it to /usr/local/bin (or ~/.local/bin without root). Pin a version with GT_VERSION, point GT_REPO at your own fork, and sha256 verification runs automatically whenever the release ships checksums.txt.
Command builder
Fill in your relay parameters and get the client launch command.
Client launch command
./gt client -local http://127.0.0.1:80 -remote tcp://id1.example.com:12080 -id pi -secret secret1