Privacy VPN

A no-logs VPN for people: pick a location, connect with WireGuard; accounts are not tied to devices.

What it is

A VPN for people, not for networks: pick a location, connect, and your traffic leaves the internet from a shared address in that place. It runs WireGuard, so the official WireGuard apps on every system work with it, and it is built so that we cannot tell which account used which exit, when — not by promise, but because the records that would say so do not exist.

An organization buys VPN accounts for its people (Privacy VPN in the console, or dcloud privacyvpn accounts create). Each account is a 16-digit number and a monthly allowance of devices. Preview price: €5 a month per account, flat.

Connect

Console: Privacy VPN → an account → Connect → pick a location → scan the QR code with the WireGuard app, or download the file. The key is made in your browser; the private key never leaves the page.

CLI:

dcloud vpn locations              # every location, its load and your round trip to it
dcloud vpn connect de             # the nearest location in Germany: writes ~/.dcloud/vpn/ewsvpn.conf
sudo wg-quick up ~/.dcloud/vpn/ewsvpn.conf
dcloud vpn connect de --up --watch  # bring it up and fail over if the exit stops answering
dcloud vpn disconnect
dcloud vpn devices                # this computer's devices (the server cannot list them)
dcloud vpn revoke --all

--port 443 for networks that block 51820; --safe-mtu for links that drop large packets.

How it keeps no record of you

Step What happens What we learn
Monthly access tokens Your device makes random tokens and blinds them; the account side signs the blinded values (RFC 9474), once a month, in one batch That the account took N tokens this month
Registering a device Your device sends one token, its signature and a WireGuard public key — no login, no cookie That some valid token registered some key. Not whose: the signer never saw the token
Using an exit Every exit knows every registered key and nothing else. Traffic leaves from the exit's shared address Nothing: exits report only totals for the machine, rounded

The device table has no account column and no time column. A device expires with its month (plus three days), whatever day it was registered, so not even the registration day is kept.

What exists on an exit

Exits are machines in our own cloud, rebuilt from the published recipe (GET /api/privacyvpn/v1/recipe) and replaced at least once a day.

  • Root file system in RAM: the disk is mounted read-only once the exit serves; every write goes to memory and is gone when the machine is. No swap, no crash dumps, the system journal keeps nothing, nothing else logs.
  • In memory while it runs: the kernel's WireGuard table (every registered key, its tunnel address, and for each key the time of its last handshake and its byte counters — the kernel keeps these; nothing reads them out), the connection-tracking table (which tunnel address talks to which destination, while the connection lasts), and the DNS resolver's shared cache (names, for at most 5 minutes, without who asked).
  • DNS: the exit's own resolver does full recursion with QNAME minimisation and DNSSEC; no query is logged or forwarded to another resolver.
  • Blocked: outbound port 25 (spam), Windows file sharing ports, private address ranges, connections between devices. New connections per device are rate-limited in memory.
  • What an exit tells us: active keys rounded up to a multiple of 10, bytes in 100 MB steps, CPU — every 5 minutes.

Check a running exit yourself, through the tunnel: curl http://10.64.0.1/ returns its manifest (agent hash, recipe hash, the root mount, swap, log directory, core-dump setting). scripts/privacy-vpn/verify-exit.sh compares it with the published recipe and agent.

What this cannot prove: a machine can lie about itself. Until exits run with hardware attestation, the manifest shows what the exit says it runs, and you trust us to run it. The cloud provider's hypervisor can read a machine's memory.

Kill switch, per system

System With the official WireGuard app
Linux The file from dcloud vpn connect drops everything outside the tunnel while it is up
Windows Tick “Block untunneled traffic (kill-switch)” in the tunnel's settings
Android Settings → VPN → WireGuard: “Always-on VPN” and “Block connections without VPN”
macOS, iOS No kill switch; some Apple system traffic bypasses any VPN

All configurations send IPv4, IPv6 and DNS through the tunnel.

Streaming

Streaming services block the address ranges of data centres, ours included. YouTube, Spotify and most broadcasters that only check the country work; the large paid services often do not. A location is marked for a service only when a daily test passed that day.

Permissions

Action What it allows
privacyvpn:DescribeExitLocations List locations and their load
privacyvpn:ListAccounts See the organization's VPN accounts
privacyvpn:CreateAccount, UpdateAccount, DeleteAccount Manage accounts
privacyvpn:IssueAccessTokens Have an account's monthly tokens signed (audited, once a month)

Registering, renewing and revoking a device need no permission: they carry a token or a revoke secret instead of an identity.

Did this page help you?