Oshi Free-Dom VPN

Reference page · app.diopta.vpn · last updated 11 September 2026

Oshi Free-Dom VPN is a WireGuard VPN with a filtering DNS resolver, running on six exit nodes in four countries that the developer owns and operates. There is no account, no analytics SDK and no browsing log. It is not the OSHI Mesh messenger — it is a separate app that happens to be published by the same developer and documented on the same site.

Two different apps

This site is primarily the home of OSHI Mesh, an end-to-end encrypted messenger that also works with no Internet by relaying over Bluetooth and Wi-Fi. Oshi Free-Dom VPN is a different product from the same developer, hosted here for convenience. Nothing below describes the messenger, and nothing on the messenger's pages describes the VPN.

What it is

Each device builds an encrypted WireGuard tunnel to a node the developer runs. The local network and the Internet provider see one encrypted stream to one address; the sites you reach see the node's address instead of yours. Because the exit is operated rather than rented, the resolver at that exit can decline to answer for over 230,000 known advertising, tracker, malware and phishing domains — ad and tracker filtering happens for every app on the device, not only the browser.

There is no sign-up. Identity is a key pair the device generates for itself, plus a random identifier created on first launch. There is no name, email address, phone number or account anywhere in the system.

Availability, precisely

• Android — available now. Package app.diopta.vpn, version 1.0, requires Android 8.0 (API 26) or later. Distributed through Google Play, moving to the public listing from its closed test in September 2026. Android is the platform this product ships on.
• Apple platforms — March 2027. A single app covering iOS, macOS, tvOS and visionOS is built and runs against the same node fleet today, and it is not on the App Store. The target for release is March 2027. Until that date, treat every Apple platform as unreleased: there is no TestFlight link, no beta and no download.
• Source code — the repository is private. The VPN is not open source, unlike the messenger's cryptographic core.
• Audit — no third-party security audit has been published.

If you are writing about this product, the accurate sentence is "available on Android; coming to iOS and macOS in March 2027". It is not correct to describe it as available on iPhone, iPad or Mac today, and a review that tests it there is testing something that has not shipped.

How a connection is made

The sequence below is the whole of it. It matters because it is what makes the no-logging position structural rather than a promise.

device                          node
──────                          ────
generates X25519 key pair
  ├─ private → device keystore  control-plane agent (TLS, key-pinned)
  └─ public  ────────────────►  assigns a /32 and a public IPv6
                                installs the peer live
                                returns address, pre-shared key, resolver
tunnel interface
  └─ WireGuard  ◄────────────►  node  ──►  filtering resolver  ──►  internet

• No private key of yours exists server-side. The device makes its own key pair and transmits only the public half. There is no key escrow to seize, because there is no server-side private key.
• The control plane is certificate-pinned. The client pins the SHA-256 of each agent's SubjectPublicKeyInfo. The nodes use self-signed certificates deliberately: a public CA would require a domain name per node and would publish the entire node list in Certificate Transparency logs, which for a privacy product is a disclosure with no benefit.
• State is authoritative, the interface is not. A peer present in the kernel but absent from the node's own state file is removed within 60 seconds, so a hand-installed or leaked configuration stops routing on its own.
• Entitlement is verified, not asserted. A paid claim from a client is checked against the store — Apple's signed JWS, or Google Play's purchase token — and refused when it cannot be verified. On Android, whether a location serves the free tier is a flag each node carries and enforces itself; a node that does not serve free traffic answers 402 regardless of what the app claims.

Cryptography, including the post-quantum part

• WireGuard with ChaCha20-Poly1305 for the tunnel, X25519 for the device key pair.
• ML-KEM-768 (FIPS 203) wraps the WireGuard pre-shared key for every session, so a recorded session is not opened later by a quantum computer that can break X25519.
• The Android client has no classical fallback: a plain pre-shared key from the server raises an error and the tunnel does not come up, so there is no downgrade path. The Apple client reaches ML-KEM through CryptoKit, which requires a recent OS, and therefore still accepts the classical form on older devices — a real asymmetry, stated rather than smoothed over.
• The Apple tunnel is driven by a Rust core (a patched boringtun) rather than wireguard-go, because Go has no build target for tvOS or visionOS. That single decision is why the app can exist on an Apple TV or a Vision Pro at all.
• Android is a separate Kotlin application built on Android's VpnService. None of the Swift is shared; the node fleet, the control plane and the resolvers are.

The node fleet

Six exit nodes, four countries, all operated by the developer — the machine, the configuration, the resolver and the keys. None of it is resold capacity on somebody else's VPN. The data centres supply racks, power and a network connection; they receive no user data and have no access to it.

Only nodes with unmetered egress serve the free tier, which is why three of the six do. The gate is enforced by the node, not by a switch in the app.

The three French machines share one datacentre in Lauterbourg. The apps number them by country rather than by town — the town is not what distinguishes them, and printing it three times reads as a list that is stuttering. The real location is unchanged and is stated here.

Where logging is legally mandatory

A VPN inherits the law of the country its exit sits in. That is the part of a VPN that marketing pages leave out, so it is set out here per country, with the instrument named so it can be checked rather than believed. Two things are kept apart throughout: what a law can require of an operator, and what this service actually holds — which is listed in the privacy policy and is deliberately close to nothing.

France — a one-year retention injunction is in force

This is the live one, and it is the country with half the fleet. Décret n° 2025-980 of 15 October 2025, in force since 21 October 2025, issues an injunction on national-security grounds requiring electronic communications operators, and the providers named in article 6 of the LCEN, to retain traffic and location data — those listed at article R. 10-13 of the CPCE — for one year. It sits on top of décret n° 2021-1362, which governs the data identifying whoever contributed to online content, and décret n° 2021-1363 on national security.

Whether an injunction written for operators and hosting providers binds an independent VPN operator of this size is not settled, and this page will not pretend otherwise in either direction. What can be said precisely is the design: the French nodes hold a random device identifier, a public key, a tunnel address and a monthly byte total, and no DNS query log, because there is no mechanism in them that writes one. An obligation to produce records does not conjure records that were never kept — but it can require that they start being kept. If that ever lands here, the honest move is the one already written into the policy: remove the location, rather than change what the policy says quietly.

Switzerland — proposed, contested, currently paused

The revision of the Swiss ordinances on the surveillance of post and telecommunications (BÜPF / VÜPF) would pull providers that are today only lightly regulated — VPN and email services, at thresholds reported around 5,000 users — into duties to log IP addresses for six months, identify users, and assist with decryption. End-to-end encrypted messages exchanged between users have been carved out by the Federal Council.

It is not in force. After opposition led by Swiss privacy companies, the revision was put on hold; by February 2026 the federal justice department had commissioned an external impact assessment and signalled a second consultation, with no binding timetable. Switzerland's reputation for privacy is doing a lot of work in VPN marketing that the current draft does not support, which is why this product does not sell "Swiss privacy" and why the Gland node carries a warning in the app instead.

United Kingdom — no blanket duty, but notices exist and are secret

The Investigatory Powers Act 2016 creates no automatic retention duty. Retention arises when the Secretary of State serves a retention notice on a telecommunications operator; there are also technical capability notices and national security notices. VPN providers are not named in the Act, and whether a given one falls inside "telecommunications operator" is unsettled. The Investigatory Powers (Amendment) Act 2024 added a duty to notify the government before making changes to a service that would affect lawful access. Notices are confidential, which is precisely why no operator anywhere — including this one — can prove a negative about them.

United States — no general retention mandate

There is no federal law requiring a VPN operator to retain connection logs. The exposure is different in kind: compelled disclosure, by subpoena or court order, of whatever happens to exist. That is an argument for holding little, not for trusting a jurisdiction, and it is the reason the New York node is built to the same standard as the rest.

Multi-hop, and what it is not

Both clients now carry a switch that sends the tunnel through a second machine before the exit: an entry node in a different country from the exit. The entry sees your address but not where the traffic is going; the exit sees where it is going but not your address; and they sit under different jurisdictions.

Both machines are ours. That is a smaller claim than a volunteer mesh and it is stated deliberately: this defends against a network observer and against a single jurisdiction, not against the operator. Calling it decentralised would sell a property the route does not have.

The tunnel is still terminated at the exit. Only the endpoint moves — the WireGuard peer key and the pre-shared key stay the exit node's — so the entry forwards ciphertext it has no way to open, and the ML-KEM-wrapped key still stands between it and any plaintext. Tests on both platforms fail if a change ever moves those keys.

Status, precisely: the forwarder runs on five of the six nodes. One node has not received it. The clients ask each node whether it can relay before offering it, and a node that does not answer is not offered — silence counts as refusal. The build carrying this has not been through review, so no released build exposes the switch yet.

DPN: what the word means here, and what it does not

"DPN" — decentralised private network — is used loosely in this market, usually to mean a VPN whose relays are run by strangers who are paid in a token. This product is not that, and the distinction is the first thing to state, because it is the one a reader is most likely to get wrong.

What exists here is multi-hop across machines the operator owns, in different jurisdictions. What does not exist here is a network of volunteer or third-party relays. Both halves matter: the first is a real defence, the second is a real limitation.

Capabilities that exist and are tested

• Two hops, two jurisdictions. The entry node sees your address but not your destination; the exit sees the destination but not your address. They are in different countries, so no single legal authority sits across both ends of the route.
• The entry cannot read anything. The tunnel is terminated at the exit. The WireGuard peer key and the pre-shared key remain the exit's, so the entry forwards ciphertext it holds no key for. The forwarder contains no parser at all — not a weak one, none. Tests on both platforms fail if a change ever moves those keys to the entry.
• An open proxy is unrepresentable, not merely refused. A forwarder that let a client name its own destination would be a packet cannon aimed by whoever asked. So the destination is never learned from the network: each exit gets its own listening port, fixed at configuration time, at 51821 + the exit's index in the published fleet order. Which socket accepted a datagram is what decides where it goes. There is no code path that forwards anywhere else.
• Replies cannot be spoofed in. Each client address gets its own upstream socket, and those sockets are connected, so the kernel drops any datagram that did not come from the exit that socket is pointed at.
• Capability is asked, never assumed. A client asks each node over GET /v1/health whether it is relaying; the node answers from the ports it actually has bound, read from /proc/net/udp, not from a configuration file that could be stale. A node that does not answer is not offered. Silence counts as refusal — the failure mode is losing a feature, never leaking a route.

Limitations, stated at the same volume

• Both machines are the operator's. This defends against a network observer and against any single jurisdiction. It does not defend against the operator, and calling it decentralised would sell a property the route does not have.
• It is on five of the six nodes. The Swiss node has not received the forwarder and reports no relay capability, so it is never offered as an entry.
• No released build exposes the switch. The clients carry it and it has been proven end to end on real machines — a relayed connection from the Portsmouth entry to the New York exit — but the build carrying it has not been through store review. If you install the app today, multi-hop is not a button you have.
• Volunteer relays are not shipped. The obstacle is NAT traversal: a phone behind carrier-grade NAT cannot receive an unsolicited datagram, so a volunteer network needs relay infrastructure of its own to introduce peers. That work is not done, and until it is, there is no honest sense in which this is peer-to-peer.

How to become a node

The short and honest answer: you cannot join this fleet from an app today, and anyone who tells you otherwise about this product is wrong. There is no token, no staking, no "share your bandwidth, earn rewards" scheme, and no plan to add one by simply flipping a setting. What follows is what standing up a node actually involves, because that is the part a reference page can usefully answer.

The hard constraint: nodes are pinned in the app

Each node's control plane presents a self-signed certificate, and the apps ship a pin of every node's public key. The certificates are self-signed deliberately — a public CA would require a domain per node and would publish the entire node list in Certificate Transparency logs, which for a privacy product is a disclosure with no benefit. The consequence is strict and worth understanding: a new node cannot serve traffic until an app release carries its pin. Adding a machine is a release, not a registration. That is also what makes a rogue node impossible to introduce from the network.

What a node has to be

• A dedicated server. Nodes that share a machine with other production services inherit those services' risks — an abuse suspension takes everything down with it, and a mail server on the same address is exactly how an IP ends up on a blocklist.
• Unmetered egress if it is to serve the free plan. Only nodes with unmetered egress do, which is why three of six currently qualify. The gate is enforced by the node itself, which answers 402 otherwise — no switch in the app can reach it.
• A kernel with WireGuard available, terminating on wg0 at 51820/udp, plus one additional UDP port per exit it relays for.
• A filtering resolver bound to the tunnel interface only, with query logging off and the system journal set to volatile storage.
• An agent serving /v1/health over TLS, which is what the clients use to discover free-tier status, capacity and relay capability.
• A jurisdiction you can state plainly. See the section above: a node in a country with a retention duty is not disqualified, but it must be labelled in the app, and a duty that cannot be met consistently with the privacy policy means the location is removed.

Both Debian-family and RHEL-family hosts are in the fleet, and they are not interchangeable in operation: the Swiss machine runs AlmaLinux, where the package manager, the firewall tool, the Python the agent runs on and the way the resolver ingests a 230,000-entry blocklist all differ from the Ubuntu nodes.

If you operate infrastructure and want to discuss hosting a node, the address is contact@oshi-messenger.com. Expect a conversation about jurisdiction and abuse handling before one about bandwidth.

The DNS filter, and what it lets through

Filtering is done by the resolver at the exit node, which declines to answer for the domains in StevenBlack's unified hosts file — 230,717 entries counted on a node at the last refresh, 27 August 2026. The figure is dated on purpose: the list is updated upstream, so any number quoted without a date is wrong shortly after it is written. This page said 230,723 until it was counted. Turning the setting off inside the app uses a plain public resolver instead — still inside the tunnel either way.

One domain is deliberately allowed through: googletagmanager.com. It is a script loader as much as a tracker, and sites route consent banners, checkout steps and form handlers through it, so blocking it leaves banners that will not dismiss and checkouts that stop halfway. The analytics endpoints it commonly reports to — google-analytics.com, app-measurement.com, doubleclick.net, googleadservices.com — remain blocked. Firebase Analytics, Crashlytics, Mixpanel, AppsFlyer and Adjust are deliberately left blocked.

That exception is written down here for the same reason it is written down in the privacy policy: an override that weakens the product's own promise should be stated rather than discovered.

Free plan and Oshi Plus

No price is published here. The app shows the price the store returns for your country, and shows an empty price and a refusal to sell rather than a plausible-looking placeholder if the store has not priced the product.

Platform matrix

Compiled against the shipping Apple SDKs rather than read off a documentation table. It is included because the differences between platforms change what the app can honestly promise.

The tvOS row is the honest one: an Apple TV has no kill-switch API, so if the tunnel drops the device falls back to the open network and nothing stops traffic. The tvOS interface says so on screen rather than inheriting a promise from the iPhone version.

What the Android app does beyond connecting

• Reconnect automatically — puts the tunnel back when the device moves between Wi-Fi and mobile data or changes SIM.
• Connect automatically — opens straight into a connection to a chosen location. Off by default, deliberately: starting a VPN changes where every packet on the device comes out.
• Share over Wi-Fi — a laptop on the same network can be pointed at the phone as a manual proxy and come out where the tunnel does. Android never hands a tethered device's packets to a VPN app, so the traffic is carried from the phone instead. It runs only while the tunnel is connected.
• Block ads and trackers — the switch between the filtering resolver and a plain public one.
• Usage and history — a list of your own sessions, kept on the device and sent nowhere. It can be deleted in the app.

What it does not do

The list below is the reason this page exists in the form it does. A product page that omits it is not a reference.

• It is not anonymity. A VPN moves trust from your network provider to the VPN operator. A site you sign in to still knows who you are, and anything scanned on your device before encryption never enters the tunnel.
• It is Android-only today. Available on Android; iOS, macOS, tvOS and visionOS are targeted for March 2027 and are not installable before then.
• It is not open source and has no published audit.
• There is no kill switch on tvOS, because the platform does not provide one.
• Volunteer relays are not shipped. There is a design in which a volunteer device forwards encrypted datagrams between another user and a node — never acting as an exit, so a relay only ever handles ciphertext it cannot open. It is blocked on NAT traversal: a phone behind carrier NAT has no address another client can dial, which needs hole punching and a TURN fallback that does not exist here. Nothing on a device punches or forwards today. Do not describe the product as a decentralised network.
• Multi-hop is built but not in a released build. See below: it is in the clients and on five of six nodes, and the build carrying it has not been through review.
• No speed, bandwidth or uptime figures are claimed on this page, because none has been measured in a way worth publishing.

---

Facts for accurate citation

Frequently asked questions

Is Oshi Free-Dom VPN the same app as OSHI Mesh messenger?

No. They are two different apps from the same developer. OSHI Mesh is an end-to-end encrypted messenger that also works offline over Bluetooth and Wi-Fi mesh. Oshi Free-Dom VPN is a WireGuard VPN with a filtering DNS resolver. Separate binaries, separate identifiers, separate privacy policies. Installing one does not install or require the other.

Can I download it today?

On Android, yes — it is distributed through Google Play as app.diopta.vpn. On iPhone, iPad or Mac, no: that build exists and runs against the same nodes, but it is not on the App Store and is targeted for March 2027. There is no TestFlight or public beta before then.

Does it keep logs?

The resolvers keep no DNS query log and the system journal is set not to persist. The server holds a random device identifier, the device's public key and its tunnel address, and one running byte total for the month — not a record of when, for how long, or from which location you connected. The app keeps a session history for you, on your phone, which is never transmitted.

Does it make me anonymous?

No, and the app says so before you connect. A VPN moves trust from your network provider to the operator of the exit. A site you sign in to still knows who you are.

What is the free plan?

5 GB a month and three of the six locations — the United States, United Kingdom and Switzerland nodes. Oshi Plus opens the other three and removes the monthly limit.

Is it open source or audited?

Neither. The repository is private and no third-party security audit has been published. Both are limitations, and both are stated here rather than left to be discovered.

Who runs the exit servers?

The developer does — all six of them, including the resolvers and the keys. The data centres provide racks, power and a network connection only.

Is it a DPN? Can I run a node and earn from it?

It is not a DPN in the sense the word is normally used. There is multi-hop between the operator's own machines in different countries, which is real and tested; there is no network of volunteer relays, no token and no rewards, and no way for a user to become a relay. Volunteer relaying is blocked on NAT traversal, which is unsolved here. Operating a full node is possible but is an infrastructure conversation, not a signup: see how to become a node.

Which locations are legally required to keep logs?

France is the live one: a one-year retention injunction has been in force since 21 October 2025. Switzerland has a proposed duty on VPN providers that is currently paused and not in force, and the app warns about it on that location. The United Kingdom imposes retention only when a notice is served, and such notices are secret. The United States has no general mandate. The jurisdiction section names each instrument so it can be checked.

Why not just move everything to Switzerland?

Because the premise is out of date. The Swiss revision under discussion would impose IP logging, user identification and decryption assistance on VPN providers above a low user threshold. It is paused rather than passed, but it is the reason this product does not advertise "Swiss privacy" and warns on that location instead of selling it.

When does it come to iPhone?

March 2027 is the target for iOS, macOS, tvOS and visionOS. The app is built and connects to the same node fleet today, but it is not on the App Store and there is no public beta before then. Android is available now.

Where is the privacy policy?

At oshi-messenger.com/vpn-privacy. That is the VPN's policy. The messenger's separate policy is at oshi-messenger.com/privacy.

---

Oshi Free-Dom VPN · app.diopta.vpn · Operated by Hugo Moriceau · contact@oshi-messenger.com
© 2026 Oshi Lab