v2rayN Pro Max is a V2Ray / Xray client for Windows and Linux, for people who use free subscription links and would rather have the client test and pick working servers than do it by hand.
Press one button. It reads your subscription links, tests what is actually alive, measures real speed, and leaves the fastest servers ready to use.
Anyone using free V2Ray or Xray subscriptions knows the routine. Open the client. Update the subscriptions. Wait. Run a ping test on a few hundred servers. Wait again. Sort the column. Try one. It is dead. Try the next. Run a speed test on whatever survived. Pick one.
Tomorrow the good ones have stopped working and you do it all over again.
That is an hour a week spent doing something a computer should be doing. Auto Mode does it in one press, and it keeps doing it, so you always have a short list of servers that were verified minutes ago rather than a long list that was true last week.
One press runs the whole pipeline:
| Stage | What happens |
|---|---|
| Sample | Picks a handful of your subscription links, favouring the ones that have produced working servers before |
| Fetch | Downloads them in parallel and parses every format: plain text, base64, ten URI schemes, Clash YAML, SIP008 |
| Filter | Keeps only the protocols and countries you asked for, if you asked for any |
| Reachability | Drops endpoints that answer nothing |
| Tunnel test | Sends a real request through each server, which is the only way to know it works |
| Speed test | Measures actual throughput on the survivors |
| Keep | Writes the fastest into their own group, replacing the previous list |
The button shows a live countdown while it runs, so you know whether to wait or walk away.
Free subscription links decay. A link that was excellent last month goes quiet, stops updating, or dies outright. Auto Mode tracks each link's record and adjusts:
- Links that keep producing working servers get sampled more often
- Links that stop producing get sampled less, and are eventually parked
- Parked links still get an occasional retry, because free sources come back as often as they vanish
- Evidence fades over time, so a link's reputation from three months ago does not decide today's run
Underneath it is a decayed Thompson sampler over your sources. In practice it means you can paste a thousand links, never curate them again, and the client works out which ones deserve its time.
Every run re-tests the servers already in your list alongside the new candidates. A server that was fastest yesterday and has since degraded loses its slot to something better. In testing, the weakest entry in the list went from 0.2 MB/s to 1.2 MB/s across two runs, purely because every slot has to be re-won.
Sources. Paste as many links as you like, one per line. Anything that is not an http link
is ignored, so pasting straight from a channel or a README works without cleaning it up first.
GitHub blob links are rewritten to their raw form automatically.
How many to keep. Anywhere from 1 to 50 servers.
Protocol. VLESS, VMess, Trojan, Shadowsocks, Hysteria2, TUIC, WireGuard, AnyTLS, Naive, SOCKS. Tick none for any.
Country. Matched against the address a server actually exits from, measured during the test, rather than whatever the provider wrote in its name. Provider labels are only used to decide what is worth testing first. If too few servers match, the remaining slots are filled with the fastest found anywhere rather than handing you a half-empty list.
Per-source statistics. Quality, how many of its servers worked, how many configs it carries, when it was last tried, and its current state. Untick a link to leave it out of future runs without deleting it and losing its history.
This is a fork of 2dust/v2rayN 7.24.4. Everything the original does, it still does. On top of that:
The whole feature above. Stock v2rayN can update subscriptions and run tests, but you drive every step and you decide what the results mean. Here it is one button and a short list of verified servers.
Anyone who has selected a thousand servers in v2rayN and pressed real ping knows the result: every single one comes back as -1. Test fifty at a time and it works fine. This has been around for years and is usually blamed on the core being unable to cope with the load.
It is not. Measured against a real subscription pool: Xray accepts 992 outbounds without
complaint, and every one of its ports is listening within a second. The actual cause is that
recent Xray removed the allowInsecure option and rejects the entire configuration file if
any single server in it still sets that flag. Free subscription lists are full of such
servers. One of them appearing in a batch of a thousand is close to certain, and it takes the
other 999 down with it.
Two fixes:
- The option is no longer written into generated configurations, which removes the cause.
- Before a batch starts, the configuration is validated by the core itself. If the core rejects a server, that one server is identified from the error message and dropped, and the batch proceeds. Validating a thousand servers takes under a third of a second and needs no network at all — far cheaper than the stock behaviour of repeatedly halving the batch and restarting real cores until the culprit is found.
Before the fix, a 120-server batch returned 0 working. After it, the same batch returned 36.
Ping tests run in waves rather than all at once, so results appear as they arrive instead of in one burst at the end, and a large batch no longer starves the thread pool.
- A single accent colour, rounded geometry and one consistent typeface across every window
- The server list is easier to scan: taller rows, alternating shading, colour-coded latency
- The log panel is a proper console, with a monospace face, line numbers and green-on-black output
- A grouped, labelled status bar instead of a row of unexplained controls
- One dark theme, styled throughout, rather than a theme picker that only half applies
Application self-update is deliberately disabled. An update replaces the executable and its assemblies, which would silently revert this build to stock upstream and leave your settings sitting there beside an application that no longer understands them.
The proxy cores are a separate matter and keep updating normally. Xray, sing-box, mihomo and the geo rule files all continue to update from their own upstream repositories, which is where the fixes that actually matter arrive. To move the app itself to a newer version, rebuild from source.
| Platform | Download |
|---|---|
| Windows 10 / 11, x64 | v2rayN-Pro-Max-windows-64.zip |
| Ubuntu and derivatives, x64 | v2rayN-Pro-Max-linux-64.zip |
Both are self-contained, so no .NET runtime is required, and both ship the proxy cores and
geo rule files in a bin folder inside the archive. That is what makes the archives large —
nothing has to be downloaded before the first test can run.
- Extract the archive anywhere, keeping the
binfolder beside the executable - Run
v2rayN.exe, or on Linuxchmod +x ./v2rayNand then./v2rayN - Open Auto Mode settings with the gear beside the automode button
- Paste your subscription links, or import
sub.example.txt, which ships in the archive - Press automode and wait a few minutes
Newer cores can be fetched later from Help → Check Update; the app itself does not self-update, on purpose.
A run takes three to five minutes depending on how many links you have and how healthy they are, uses roughly fourteen percent of one core while it works, and settles back to normal memory afterwards.
sub.example.txt ships with 76 public subscription links so there is something to start with.
None of them are operated by this project.
git clone https://github.com/morpheusadam/v2rayN-Pro-Max.git
cd v2rayN-Pro-Max/v2rayN
dotnet build v2rayN.Desktop/v2rayN.Desktop.csproj -c Release
Requires the .NET 10 SDK. Two Windows front ends are included and share all of the logic:
v2rayN.Desktop (Avalonia, the one the releases ship) and v2rayN (WPF).
Run the tests:
dotnet test ServiceLib.Tests/ServiceLib.Tests.csproj
Regenerate the icons after replacing tools/logo-source.png:
python tools/make_icons.py
The numbers behind the design are in docs/automode-design.md. Two findings from measuring a real subscription pool shaped it.
Reachability means almost nothing. Eighty percent of endpoints accept a TCP connection. Five percent actually carry traffic. A port being open says nothing about whether the proxy behind it works, so reachability is only ever used to drop the completely dead.
Sorting candidates by lowest ping is worse than not sorting at all. 2.1 percent of the lowest-latency candidates worked, against 7.5 percent of a random selection from the same pool. The fastest-answering hosts are CDN edges that acknowledge any connection while the proxy behind them is long gone. Latency is therefore never used to rank candidates, only to discard the unreachable.
Built on 2dust/v2rayN, which does the hard work of supporting every protocol, core and transport. Cores by XTLS/Xray-core, SagerNet/sing-box and MetaCubeX/mihomo. Interface built with Avalonia and Semi.Avalonia.
Licensed under GPL-3.0, the same as upstream.
The links in sub.example.txt are public lists collected from GitHub and public channels.
They are not operated, endorsed or verified by this project, and their contents change without
notice. Auto Mode measures them rather than trusting them, which is rather the point. Use
servers you trust for anything that matters.
Built for people who would rather use the internet than spend an hour a week testing proxies.