Umbra compared with mainstream proxy options
Compare Xray, VMess, Trojan, Shadowsocks and Hysteria 2 by real-site cover, certificate maintenance, TCP/UDP paths and client integration.
Overview
When choosing a proxy, start with whether it suits your network, applications and maintenance preferences, rather than the number of algorithms or test coverage. Umbra is a self-hosted privacy transport: it integrates real-site cover, TCP/Vision, QUIC and a local SOCKS5 endpoint into one client and server pair.
Comparison scope: Umbra 1.0.0-alpha; other projects are described using official documentation consulted on 2026-09-16. This compares features and deployment approaches, not speed, security rankings or anonymity guarantees. Later versions and different clients may change specific behavior.
Xray is a proxy platform supporting multiple protocols and routing; VMess, VLESS and Trojan are protocols. For a meaningful comparison, Xray below is explicitly limited to the common VLESS + REALITY + Vision configuration with a RAW/TCP outer transport. Limitations of one configuration must not be presented as limitations of all of Xray.
Selection at a glance
| Option | Outer connection and site behavior | Deployment and maintenance | TCP and UDP |
|---|---|---|---|
| Umbra | REALITY-style real-site cover; unauthenticated connections are forwarded to the destination site | Self-host both ends, manage identity keys and a reachable real site; no need to obtain a CA certificate for the node yourself; currently alpha | TCP/Vision, TCP mux or QUIC; one SOCKS5 endpoint can select separate outer transports for TCP and UDP |
| Xray + VLESS + REALITY + Vision | Borrows a real site's identity; Vision handles eligible inner TLS traffic | Maintain Xray configuration, keys and the destination site; no need to obtain your own certificate for a REALITY node | This comparison covers the TCP outer transport; UDP requests can also be proxied, depending on flow and client configuration |
| VMess AEAD | Provides an encrypted proxy itself; TLS, WebSocket and similar options are separately chosen transport configurations | VMess itself does not require a website certificate; adding TLS requires the corresponding authentication configuration | Supports TCP and UDP requests; the actual outer transport depends on configuration |
| Trojan (original TLS approach) | Real TLS handshake; authentication failures or non-Trojan requests can fall back to an HTTP service | Typical deployments maintain a domain, TLS certificate and private key; needing a certificate does not mean paying for one | Proxies both TCP and UDP requests, with UDP encapsulated in TLS/TCP |
| Shadowsocks AEAD / 2022 | Native encrypted TCP/UDP; browser TLS and real-site cover are not built into the base protocol | No website domain or TLS certificate required; maintain keys and matching clients, with plugins configured separately | Native TCP/UDP support; a TCP-only SIP003 plugin must not be treated as UDP camouflage |
| Hysteria 2 | QUIC/UDP; behaves as an HTTP/3 website when unauthenticated | Requires a usable UDP path and TLS configuration, using certificate files or ACME | TCP uses QUIC streams and UDP uses unreliable datagrams; it is not a TCP outer-transport solution |
Strengths and trade-offs
Every option here helps somewhere and costs something elsewhere. The list below condenses each option's strengths and trade-offs so you can match them to your network conditions and maintenance appetite.
- Umbra — Strengths: one endpoint carries both paths (TCP Vision and QUIC UDP) behind a single local SOCKS5 node; identity keys and temporary certificate binding remove public certificate paperwork; cover, transports and SOCKS5 ship as one tested client-and-server pair. Trade-offs: alpha and CLI-only today; no VLESS, VMess or Trojan node import; protocol changes between releases require upgrading both endpoints together.
- Xray + VLESS + REALITY + Vision — Strengths: mature multiprotocol platform with rich routing, many ready-made clients and extensive community documentation; REALITY and Vision are as available here as anywhere. Trade-offs: you assemble and maintain the stack yourself — core config, transport layers, routing rules and matching clients; shared mechanisms do not make it interoperable with Umbra.
- VMess AEAD — Strengths: very wide client support across platforms, including older and low-power devices. Trade-offs: the protocol encrypts but does not camouflage — cover depends entirely on the transport you add; legacy non-AEAD modes are deprecated, so pin clients that implement AEAD.
- Trojan (original TLS approach) — Strengths: a real TLS handshake with documented fallback behavior and a design stable for years. Trade-offs: typical setups maintain a domain and renew a publicly trusted certificate; UDP rides inside TLS over TCP, so UDP timing follows the TCP path.
- Shadowsocks AEAD / 2022 — Strengths: minimal moving parts — keys and a matching client, no domain or certificate; native TCP and UDP support. Trade-offs: base protocols provide encryption, not camouflage; plugins are a separate surface to assess, and a TCP-only SIP003 plugin must not be treated as UDP camouflage.
- Hysteria 2 — Strengths: an unreliable-datagram UDP path designed for lossy networks, with congestion control tuned for harsh conditions; behaves like an HTTP/3 website when unauthenticated. Trade-offs: blocked or restricted UDP removes its main advantage; TCP still travels as QUIC streams, so loss can stall streams like any reliable transport, and its design is not a guarantee of speed on your network.
Reasons to consider Umbra
Bring two transport paths into one endpoint
One Umbra client can set transport = "tcp", udp_transport = "quic" and mux = false to use Vision for TCP and QUIC for UDP; the same server opens both TCP and UDP. Existing clients connect to just one local SOCKS5 node, without splitting the two paths into two sets of Umbra processes.
This is a configuration and usability convenience, not a claim that other platforms cannot achieve similar results with different outbounds or routing. Umbra does not automatically determine which path is fastest and does not provide automatic TCP/QUIC failover.
Reduce redundant processing for eligible traffic
After completing the authenticated boundary exchange, TCP/Vision can directly forward eligible inner TLS 1.3 encrypted records, avoiding subsequent outer TLS encryption and extra encapsulation. Non-TLS and unsupported TLS traffic remain encrypted. TCP mux instead lets multiple requests share outer connections, reducing repeated connection establishment.
Neither REALITY nor Vision is unique to Umbra, and neither justifies claiming comprehensive superiority over Xray. Forwarding here happens in userspace, not through kernel zero-copy; without competitor benchmarks in a uniform environment, there is no basis to say "how much faster" it is.
Avoid maintaining a public certificate for your node yourself
Umbra uses identity keys and temporary certificate binding, so deployment does not require you to obtain or renew a public CA certificate. You still need to manage server secrets, client credentials and a reachable real destination site. Enabling QUIC also requires real QUIC fallback and UDP reachability.
Likewise, a REALITY combination can reduce this certificate maintenance, and native Shadowsocks deployments do not depend on website certificates either. This is a deployment trade-off, not an advantage exclusive to Umbra.
Decide whether to migrate from your current setup
- Already using Xray / VLESS / REALITY: If you depend on multiple protocols, complex routing and the existing client ecosystem, there is no need to migrate just for a similar cover mechanism. Choose Umbra because its single-instance dual-transport approach fits your usage, not because "REALITY is more advanced."
- Already using VMess: First identify the actual TLS, WebSocket or other carrier in use. The VMess name alone does not imply plaintext, a lack of camouflage or an inability to carry UDP. Umbra cannot import VMess nodes directly.
- Want to maintain a standard TLS service: Trojan has documented deployment approaches for real TLS and fallback; do not describe it as lacking probe resistance. Its differences from Umbra include certificate maintenance and the UDP transport path.
- Only need a lightweight encrypted proxy: Shadowsocks may be closer to your needs. Discuss AEAD and 2022 by version; 2022 includes full replay protection. Plugin capabilities are not the same as base protocol capabilities.
- Primarily considering QUIC and lossy networks: Evaluate Hysteria 2 as well. Its UDP uses unreliable datagrams, whereas Umbra currently encapsulates UDP association data in reliable QUIC streams; loss can still leave the same stream waiting for retransmission, so it is incorrect to advertise "no head-of-line blocking for all UDP." Whether either is better for games, calls or downloads requires testing on the actual path.
- Need a mature graphical interface, mobile support or ready-made subscriptions: Check the specific client and version first. Umbra is currently a CLI transport tool that can integrate with rule-based clients supporting SOCKS5, but does not include these ecosystem features itself.
Implemented connection flow
- Establish the outer handshake. The TCP client constructs its own TLS 1.3 ClientHello and records; native QUIC uses Quinn. Authentication material is bound to the ClientHello and placed in
legacy_session_id. - Check the connection's identity. The server checks this material before selecting the authenticated transport path; unauthenticated connections go to the configured real destination. The client verifies the temporary certificate's binding information to avoid mistaking an ordinary real site for an Umbra server.
- Transport application requests. Authenticated connections carry addresses, streams and padding according to the mode; Vision changes forwarding behavior only when its conditions are met and negotiation is complete.
Real-site fallback does not guarantee that traffic will always be unrecognizable. Destination site failures, network conditions, timeouts and observer capabilities all affect real-world behavior, and the server IP may also be blocked directly.
Browser fingerprint and post-quantum boundaries
chrome-lateststill uses the historical Chrome 150 profile. Extension ordering, GREASE, JA3/JA4 checks and actual Chrome 153 samples cannot individually prove complete equivalence to a current browser.- ML-DSA-65 provides a signature binding for the temporary certificate's public key; this does not mean every authentication step becomes post-quantum authentication.
- The code has an X25519 + ML-KEM-768 hybrid exchange path, but its shared secret is used only when a hybrid group is actually negotiated. The current standard CLI probes the destination using the ring provider, which does not support ML-KEM, and the server follows the probed group. Default connections therefore cannot be claimed to have hybrid post-quantum exchange enabled.
- Current official Xray REALITY configuration also includes optional ML-DSA and hybrid exchange support tied to the group negotiated with the destination site. Post-quantum capabilities must not be advertised as exclusive to Umbra.
- Umbra currently does not support Geneva DSL sending, 0-RTT or fallback preconnection; plans in the protocol design are not existing product capabilities.
Official comparison sources
The following links are the sources for the other options on this page, consulted on 2026-09-16. Umbra's sources are the README, usage guide, performance notes and runtime code listed in this page's metadata.
- Xray: Official REALITY examples (pinned revision), REALITY configuration and post-quantum options (pinned revision), VLESS / Vision configuration, Routing.
- VMess: V2Fly protocol documentation, Xray transport layering.
- Trojan: Protocol, TLS and server configuration.
- Shadowsocks: AEAD, 2022 / SIP022, SIP003 plugin boundaries.
- Hysteria 2: Protocol specification, Server deployment, Client modes.