# URnetwork vs Obscura

Obscura is the closest design to URnetwork's on the market, and on one axis
it is ahead: its entry hop cannot read user traffic under any configuration.
URnetwork spreads the exit across thousands of members instead of one
partner's fleet and ships post-quantum encryption by default.

**Choose Obscura** if you want operator blindness with zero exceptions, a
published protocol audit, and Mullvad's court-tested fleet as the exit.
**Choose URnetwork** if you want an exit side spread across members on
residential addresses, post-quantum encryption on by default, and the whole
stack published as open source. Both replace the no-logs promise with
structure. They divide the same problem differently and fail in different
directions.

[Obscura](https://obscura.com) launched in February 2025 from Sovereign
Engineering Inc., a small New York company. It is a two-party relay: Obscura
runs the entry hop your device dials, and Mullvad runs the exit. The client
negotiates its WireGuard session directly with the exit's public key and
tunnels the encrypted datagrams to Obscura's relay inside QUIC, so the
company you pay forwards bytes it holds no key for. Mullvad announced the
partnership on its own blog in February 2025, and every one of the 254 exits
in Obscura's public API on 7 August 2026 matched a live, named, diskless
Mullvad WireGuard relay, about 45% of Mullvad's fleet.

URnetwork uses member-run exits. The provider does not receive your source IP
on the relayed path. Native apps seal traffic to the provider by default.
[How URnetwork works](/docs/overview) explains the full design. The
[threat model](/docs/threat-model) holds the complete record of what is stored
and enforced.

## Six differences that affect the choice

| | Obscura | URnetwork |
|---|---|---|
| Operator blindness | Structural: one code path keyed to the exit, every platform, both transports, no fallback | Sealed by default in the native apps, skipping any provider it cannot seal to rather than downgrading; absent from the browser extension |
| External verification | A published Cure53 audit of the protocol and macOS client | Open client and server code; no independent audit of the protocol or server code |
| Exit side | One partner: 254 Mullvad datacenter relays in 24 countries and 38 cities (its own API, Aug 2026) | Thousands of member devices in 90+ countries and 2,000+ real cities (its own count); browse countries, search for a city |
| Exit address | Published Mullvad ranges, already labeled hosting/VPN by IP-intelligence databases | Residential and mobile consumer addresses on no list |
| Post-quantum | None; the WireGuard pre-shared-key slot Mullvad's own scheme uses is left empty | Yes: X25519MLKEM768 on the client-to-provider session, on by default |
| Source and licence | Client source-available under PolyForm Noncommercial; relay and API backend unpublished | Clients, SDK and server open source |

Caveats below the table:

- Obscura is $8.00/month flat, no multi-month discount, nonrefundable by
  default, three connection slots. URnetwork Pro is $5/month or $40/year
  with a free daily allowance, plus one-off data packs. Current numbers are
  at [ur.io/products](https://ur.io/products).
- Platforms: Obscura ships macOS, iOS and Android plus generic WireGuard
  configs; Windows and Linux are under active development. URnetwork ships
  all five plus a browser extension paired with its web app.
- Kill switch: Obscura lists an "On demand kill switch" under Experimental,
  and on Android relies on the OS always-on setting. URnetwork ships an
  explicit toggle on every platform; the extension's defaults on.

![Plural exit identity: the provider window — per-site affinity and failover](/docs-assets/diagram-provider-window.svg)

*One exit versus several at once is the sharpest structural difference here:
Obscura gives you a single Mullvad relay per session, while URnetwork spreads
a session across a window of providers with each site pinned to one of them.*

## Operator blindness on each side

Obscura's entry hop cannot read user traffic for any user, on any platform,
in either transport mode. The property comes from the key schedule rather
than a setting: the client's WireGuard session is built against the exit's
public key, the relay carries the resulting datagrams inside QUIC, and no
code path terminates the tunnel at the relay.

URnetwork's counterpart is on by default and not unconditional. Post Quantum
Encryption seals the client-to-provider session in every native app out of
the box. Every current provider build answers it, and while the setting is on
a provider that cannot be sealed to is skipped rather than used unsealed — so
the cost is availability, not a silent downgrade the operator could read.
Turn the setting off and traffic can take the standard path, where the
operator could in principle observe which sites a session contacts. No app yet
shows a per-connection sealed-or-not verdict, and the browser extension ships
no sealed session at all; there, the operator's data path logs nothing, pinned
by a test in the open code. On this axis Obscura is ahead.

One qualification travels with that concession. The exit public key is
whatever Obscura's own API returns, and nothing in the client verifies it
against Mullvad's published list. The exact claim is that
Obscura cannot decrypt your traffic provided the key it hands you is
genuinely Mullvad's. The check is real and manual: the app displays the exit
key, Mullvad publishes the same keys, and a user can compare them. The app
will not do it for you.

## One exit partner, one landlord

All 254 of Obscura's exits are Mullvad's; its own FAQ says "a
fully-independent exit hop (currently Mullvad)", and the exit provider is a
data field in the client rather than a hardcoded dependency. Neither party
publishes the commercial terms, so the observation is concentration, not
anything about the contract. The partner is about the best available:
diskless servers, a public list naming every host, and the April 2023 raid
that produced nothing ([URnetwork vs Mullvad](/docs/compare-mullvad)).
Obscura's users inherit that record at the exit, but not Mullvad's
post-quantum tunnels or its DAITA
traffic-analysis defense: the pre-shared-key slot is left empty, and DAITA
is off on all 254 exits Obscura offers.

The infrastructure under the two parties is less separate than the corporate
structure. Measured against Obscura's own relay API and RIPE registry data on
7 August 2026: 35 of its 38 entry relays announce from ASNs registered to
Datacamp Limited, and for 29 of the 38 relay locations every preferred
Mullvad exit is a DataPacket relay, Datacamp again; in San Jose both hops
share one ASN. This breaks no cryptography, and ASN co-location is not
physical co-location. It means that across roughly three quarters of the
footprint one hosting company carries both legs, so a party with visibility
into that company's network can watch both sides and attempt timing
correlation. Cure53's conclusion names the boundary in its own words: the
protocol "prioritizes low latency over maximum protection against global
attackers."

Hosting concentration is industry-wide; it lands harder here because
separating the hops is the product. Correlation is the attack both products
are thin against: Cure53's OBS-01-006 found Obscura's QUIC data packets
unpadded, with padding now a default-off option, and URnetwork ships no
cover traffic at all, only the spread of a session across several providers,
which raises the cost of watching one. The difference is the exit side.
URnetwork's traffic leaves through many households on unrelated consumer
ISPs rather than one company's address space, and a location exists only
while a member's device is online in it, with live counts published without
a login at api.bringyour.com/stats/last-90. The evidence is asymmetric, and
stated as such: the Datacamp figures are outside measurements of Obscura,
and nobody has measured URnetwork's fleet from outside.

## The audit, and the licence

Cure53's OBS-01 engagement is the more relevant audit in this comparison.
It examined the protocol design itself, along with the macOS app and its
network extension, in May 2025, over 20 person-days with three senior
testers. Six findings, one rated Medium, four with verified fixes. Obscura
quotes it accurately: "no major security vulnerabilities within its defined
threat model," with the two-party relay called "a significant architectural
achievement". The full report is on Obscura's own site. Scope travels with
the citation: the iOS app, the Android app, Windows, Linux, the relay servers
and the API backend were outside it, so of the surfaces Obscura runs in
production the audit covers one, and as of August 2026 it is the only one.
URnetwork has no protocol audit at all; its two 2025 assessments cover other
surfaces and are listed under Limits.

![How you verify a privacy claim — point-in-time audits vs court and raid evidence vs open continuous verification](/docs-assets/infographic-verification.svg)

*Obscura holds one scoped point-in-time audit and inherits Mullvad's raid
evidence at the exit. URnetwork's answer is the third kind: open code anyone
can check on any day, with the missing audit of its protocol and server code
stated alongside.*

Licensing separates the two codebases more than the word "public" suggests.
Obscura's client is on GitHub under the PolyForm Noncommercial License
1.0.0: source-available, not open source by the usual definition, with
commercial use and competing services prohibited and external contributions
refused for now. The relay daemon and API backend are unpublished, and
reproducible builds remain a stated plan eighteen months after they were
promised. Two corrections run in Obscura's favor: it never calls itself open
source, its phrase is "public source code", and the "fully open source"
claim in third-party reviews is those reviews' error; and source-available
still supports verification, since the architecture claims above were
checked against that source and held. URnetwork publishes clients, SDK and
server under an open-source licence with contributions accepted.

## Where Obscura wins

- Zero exceptions on operator blindness: the entry hop cannot read traffic
  for any user on any platform, with no fallback and no extension carve-out.
- The published protocol audit, full report on its own site. URnetwork has
  no equivalent for the layer its privacy claim rests on.
- The exit operator's record: Mullvad's diskless fleet and the 2023 raid
  that produced nothing.
- Two jurisdictions by design: a demand must reach a US company and a
  Swedish one at the same time, then still correlate traffic.
- Predictable performance from datacenter capacity, against member uplinks.
- Bitcoin Lightning and Monero payment, leaving less trace than USDC on a
  public ledger, and an account that is a random 19-digit number with
  proof-of-work instead of identity checks.
- Pricing and disclosure style: flat $8, no discount ladder, and exit and
  relay lists anyone can query without an account.

## Where URnetwork wins

- No single partner owns the exit side: egress is thousands of members' own
  connections under signed contracts, against one company's fleet.
- The exit address: residential and mobile IPs where published datacenter
  ranges are blocked or challenged, the most common everyday failure of a
  datacenter exit.
- Reach and aim: 90+ countries and 2,000+ real cities by its own count,
  against 24 countries and 38 cities from Obscura's API; browse countries,
  search for a city.
- Post-quantum encryption on by default. Obscura has none and claims none.
- Plural exits: a window of several providers with per-site pinning, so no
  one exit carries a whole session.
- Whole-stack open source, including the half that assigns keys and
  providers; Obscura's equivalent half is unpublished.
- Breadth: five OS apps plus the extension and web app, an SDK, a free daily
  allowance, and a supply side in the [UR protocol](https://ur.xyz).

## Limits and evidence

URnetwork's main limits:

- No independent audit covers the protocol, the connect engine, or the
  operator's server code. Two 2025 third-party assessments cover other
  surfaces: a penetration test of the web application and API (April–May
  2025), and the Leviathan MASA AL2 assessment of the Android app, which
  passed. Leviathan writes that its assessment "should not be read as a
  holistic security evaluation or comprehensive penetration test." Neither
  examined logging, retention, or the data path.
- The sealed session is default, not unconditional: with it on, a provider it
  cannot seal to is skipped rather than used unsealed, so the cost is
  availability; turn it off and the standard path is available again. No app
  yet shows a per-connection sealed-or-not verdict, and the browser extension
  has no sealed session. Obscura's single code path carries none of these
  caveats.
- The split assumes the operator and the providers in your window are
  independent. Nothing in the system attests that independence, the operator
  could itself run providers, and no outside party has measured the fleet.
  See the [threat model](/docs/threat-model), §6.1.
- URnetwork retains more than Obscura claims to: a keyed hash of the address
  block and a resolved city per connection, against a stated policy of
  storing no IP at all. URnetwork's record is checkable in code; Obscura's
  rests on an unpublished backend.
- Neither company escapes US legal process: Obscura is a New York company
  and BringYour, Inc. a Delaware one, and neither publishes a warrant canary
  or transparency report.

Obscura's limits are concentration and publication. Both hops mostly ride
one hosting company; the exit side belongs to one partner; the relay and API
code is unpublished, and the API is the part that hands your client its exit
key; and the one numeric coverage claim on its site, "33+ countries", does
not match the 24 its own API returns. None of it is a scandal; the company
is small, young, and shipping quickly, with releases about fortnightly.

Trying URnetwork costs nothing: the Instant Account takes one tap and no
email, so you can test it against your own sites before paying for either
product. More questions are answered in the [FAQ](/docs/faq).