Skip to content
Switch branches/tags


Failed to load latest commit information.
Latest commit message
Commit time



A cross-platform, user-space WireGuard port-forwarder that requires no system network configurations.

Build status Latest Release


  • You have an existing WireGuard endpoint (router), accessible using its UDP endpoint (typically port 51820); and
  • You have a peer on the WireGuard network, running a TCP or UDP service on a port accessible to the WireGuard network; and
  • You want to access this TCP or UDP service from a second computer, on which you can't install WireGuard because you can't (no root access) or don't want to (polluting OS configs).

For example, this can be useful to forward a port from a Kubernetes cluster to a server behind WireGuard, without needing to install WireGuard in a Pod.


onetun opens a TCP or UDP port on your local system, from which traffic is forwarded to a port on a peer in your WireGuard network. It requires no changes to your operating system's network interfaces: you don't need to have root access, or install any WireGuard tool on your local system for it to work.

The only prerequisite is to register a peer IP and public key on the remote WireGuard endpoint; those are necessary for the WireGuard endpoint to trust the onetun peer and for packets to be routed.

./onetun [src_host:]<src_port>:<dst_host>:<dst_port>[:TCP,UDP,...] [...]  \
    --endpoint-addr <public WireGuard endpoint address>                   \
    --endpoint-public-key <the public key of the peer on the endpoint>    \
    --private-key <private key assigned to onetun>                        \
    --source-peer-ip <IP assigned to onetun>                              \
    --keep-alive <optional persistent keep-alive in seconds>              \
    --log <optional log level, defaults to "info"

Note: you can use environment variables for all of these flags. Use onetun --help for details.


Suppose your WireGuard endpoint has the following configuration, and is accessible from

# /etc/wireguard/wg0.conf

PrivateKey = ********************************************
ListenPort = 51820
Address =

# A friendly peer that hosts the TCP service we want to reach
AllowedIPs =

# Peer assigned to onetun
AllowedIPs =

We want to access a web server on the friendly peer ( on port 8080. We can use onetun to open a local port, say, that will tunnel through WireGuard to reach the peer web server:

./onetun                                  \
    --endpoint-addr                                    \
    --endpoint-public-key 'PUB_****************************************'  \
    --source-peer-ip                                          \
    --keep-alive 10

You'll then see this log:

INFO  onetun > Tunneling TCP []->[] (via [] as peer

Which means you can now access the port locally!

$ curl
Hello world!

Multiple tunnels in parallel

onetun supports running multiple tunnels in parallel. For example:

$ ./onetun
INFO  onetun::tunnel > Tunneling TCP []->[] (via [] as peer
INFO  onetun::tunnel > Tunneling TCP []->[] (via [] as peer

... would open TCP ports 8080 and 8081 locally, which forward to their respective ports on the different peers.

UDP Support

onetun supports UDP forwarding. You can add :UDP at the end of the port-forward configuration, or UDP,TCP to support both protocols on the same port (note that this opens 2 separate tunnels, just on the same port)

$ ./onetun
INFO  onetun::tunnel > Tunneling UDP []->[] (via [] as peer

$ ./onetun,TCP
INFO  onetun::tunnel > Tunneling UDP []->[] (via [] as peer
INFO  onetun::tunnel > Tunneling TCP []->[] (via [] as peer

Note: UDP support is totally experimental. You should read the UDP portion of the Architecture section before using it in any production capacity.

IPv6 Support

onetun supports both IPv4 and IPv6. In fact, you can use onetun to forward some IP version to another, e.g. 6-to-4:

$ ./onetun [::1]:8080:
INFO  onetun::tunnel > Tunneling TCP [[::1]:8080]->[] (via [] as peer

Note that each tunnel can only support one "source" IP version and one "destination" IP version. If you want to support both IPv4 and IPv6 on the same port, you should create a second port-forward:

$ ./onetun [::1]:8080:
INFO  onetun::tunnel > Tunneling TCP [[::1]:8080]->[] (via [] as peer
INFO  onetun::tunnel > Tunneling TCP []->[] (via [] as peer


Normally I would publish onetun to However, it depends on some features in smoltcp and boringtun that haven't been published yet, so I'm forced to use their Git repos as dependencies for now.

In the meantime, you can download the binary for Windows, macOS (Intel), and Linux (amd64) from the Releases page.

You can also run onetun using Docker:

docker run --rm --name onetun --user 1000 -p 8080:8080 aramperes/onetun \ [...options...]

You can also build onetun locally, using Rust:

$ git clone && cd onetun
$ cargo build --release
$ ./target/release/onetun


onetun uses tokio, the async runtime, to listen for new TCP connections on the given port.

When a client connects to the local TCP port, it uses smoltcp to create a "virtual interface", with a "virtual client" and a "virtual server" for the connection. These "virtual" components are the crux of how onetun works. They essentially replace the host's TCP/IP stack with smoltcp's, which fully runs inside onetun. An ephemeral "virtual port" is also assigned to the connection, in order to route packets back to the right connection.

When the real client opens the connection, the virtual client socket opens a TCP connection to the virtual server. The virtual interface (implemented by smoltcp) in turn crafts the SYN segment and wraps it in an IP packet. Because of how the virtual client and server are configured, the IP packet is crafted with a source address being the configured source-peer-ip ( in the example above), and the destination address is the remote peer's (

By doing this, we let smoltcp handle the crafting of the IP packets, and the handling of the client's TCP states. Instead of actually sending those packets to the virtual server, we can intercept them in the virtual interface and encrypt the packets using boringtun, and send them to the WireGuard endpoint's UDP port.

Once the WireGuard endpoint receives an encrypted IP packet, it decrypts it using its private key and reads the IP packet. It reads the destination address, re-encrypts the IP packet using the matching peer's public key, and sends it off to the peer's UDP endpoint.

The remote peer receives the encrypted IP and decrypts it. It can then read the inner payload (the TCP segment), forward it to the server's port, which handles the TCP segment. The server responds with SYN-ACK, which goes back through the peer's local WireGuard interface, gets encrypted, forwarded to the WireGuard endpoint, and then finally back to onetun's UDP port.

When onetun receives an encrypted packet from the WireGuard endpoint, it decrypts it using boringtun. The resulting IP packet is dispatched to the corresponding virtual interface running inside onetun; the IP packet is then read and processed by the virtual interface, and the virtual client's TCP state is updated.

Whenever data is sent by the real client, it is simply "sent" by the virtual client, which kicks off the whole IP encapsulation and WireGuard encryption again. When data is sent by the real server, it ends up routed in the virtual interface, which allows the virtual client to read it. When the virtual client reads data, it simply pushes the data back to the real client.

This work is all made possible by smoltcp and boringtun, so special thanks to the developers of those libraries.


UDP support is experimental. Since UDP messages are stateless, there is no perfect way for onetun to know when to release the assigned virtual port back to the pool for a new peer to use. This would cause issues over time as running out of virtual ports would mean new datagrams get dropped. To alleviate this, onetun will cap the amount of ports used by one peer IP address; if another datagram comes in from a different port but with the same IP, the least recently used virtual port will be freed and assigned to the new peer port. At that point, any datagram packets destined for the reused virtual port will be routed to the new peer, and any datagrams received by the old peer will be dropped.

In addition, in cases where many IPs are exhausting the UDP virtual port pool in tandem, and a totally new peer IP sends data, onetun will have to pick the least recently used virtual port from any peer IP and reuse it. However, this is only allowed if the least recently used port hasn't been used for a certain amount of time. If all virtual ports are truly "active" (with at least one transmission within that time limit), the new datagram gets dropped due to exhaustion.

All in all, I would not recommend using UDP forwarding for public services, since it's most likely prone to simple DoS or DDoS.


MIT. See LICENSE for details.


Access your WireGuard network from any device, without root access or having to install WireGuard locally.





No packages published