Why AI tools are sensitive to routes
The same account works fine on one route, then asks for re-verification or throws an error on another — the cause is usually not the account itself. Look at the three things below separately and troubleshooting gets much faster.
Region detection
The exit IP's location decides what you can use
Platforms read the registered location of your exit IP to judge which region this visit comes from, then decide which features to show and which models to allow. Sign in from a different region with the same account and the page asks you to verify again; the model list may change too. Where a route's exit sits — which country or region — directly decides where this session is treated as coming from.
IP risk checks
Address type and how widely it is shared affect the pass rate
Exit addresses flagged as data-centre IPs, or addresses shared by a large number of accounts, trigger CAPTCHAs and second-step checks more easily. Residential exits are treated more leniently on some platforms, but bandwidth and stability usually can't match a dedicated line. When picking a route, trade pass rate against stability instead of fixating on one IP type.
Long connections
Streaming output suffers most from mid-stream drops
A streaming answer is one continuous connection, pushing text back character by character. Switching exits mid-stream, packet loss or a NAT table timeout all leave the answer stuck mid-sentence while the page shows a network error. These problems are worst at peak hours and depend directly on whether the route is competing with the public internet for bandwidth.
The three requirements stack: region detection decides whether you can use it at all, IP risk checks decide how smoothly it goes, and long-connection quality decides how tiring it is.
Tool-to-route comparison table
The table below is organised by what each tool is mainly used for. The route-type column gives an order to try first, not a hard requirement; the same tool can also behave differently on the web and through the API — see below.
| Tool | Main use | Exit-region requirements | Route types to try first | Watch out for |
|---|---|---|---|---|
| ChatGPT | Chat, writing, code explanation | Exit must sit in a region where the service is available; keep the sign-up region and the usual sign-in region consistent | Streaming answers are sensitive to packet loss; route differences show up most when working through long documents | |
| Claude | Long-form reading, document analysis | Sensitive to exit region; frequent region switching easily triggers re-verification | Single requests run long — pin one exit | |
| Gemini | Retrieval-style Q&A, multimodal | Feature entry points differ by exit region | Many page resources; first-screen load is more sensitive to bandwidth | |
| Copilot | In-editor completion and chat | Region checks are relatively loose; connection continuity matters more | Completion requests are frequent but tiny; connection setup speed matters more | |
| Midjourney | Image generation and assets | Relies on Discord or a web entry point; region checks vary with the entry point | Image generation and downloads take bandwidth — don't switch routes while queued | |
| Cursor | In-editor chat, code indexing | Tied to the account's region; shared team exits trigger verification more easily | Indexing and long-context requests take a long time — give them a dedicated route |
In the table, “IEPL” means a dedicated line, “Relay” means a route forwarded through a relay node, and “Direct” means a route that leaves the region directly. The buying guide compares cost and stability across all three in more detail; for a country-by-country route list, see the routes page.
Account creation and sign-in: what to watch
Region traces left behind at sign-up follow an account for a long time. Here are four things you can head off at the account level.
Sign-up region
Pin the route before you create the account
The exit region at sign-up shapes which region the account is judged to belong to. Pin the route to the region you plan to use long-term before creating the account, and keep sign-up, first sign-in and daily use on the same exit where possible — that cuts the re-verification that comes with region switching.
Extra verification
Sign-in may ask you to confirm again
Some platforms add verification at sign-in; the method follows each platform's policy, so go by what the page shows. VPNMK sign-up needs no email address — a username and password is enough. Subscriptions and client sign-in both use your VPNMK account, which is one less thing linked together.
Multiple devices
A consistent exit matters more than device count
VPNMK doesn't limit the number of devices, but when the same account reaches out from exits in different countries on several devices, the platform side is more likely to flag it as unusual. If you need several regions in parallel, pin your usual devices to one route and group the rest by purpose rather than letting everything switch at random.
Session persistence
Don't tick “keep me signed in” on shared devices
Options like “keep me signed in” cut repeat verification, at the cost of leaving session credentials on the machine. Don't tick it on shared devices — sign out and clear browser data when you're done. On your own devices, turning it on noticeably reduces friction at sign-in.
Web vs. API: the differences
The same tool takes a different network path on the web and through the API, and fails differently too. Looking at the two separately saves detours.
| Aspect | Web app | API calls |
|---|---|---|
| What it connects to | Several domains plus static assets | One or a few API domains |
| Traffic shape | Page assets plus streaming responses | Request/response plus streaming responses |
| Region detection | Strict, including browser fingerprinting | Mainly the request's source IP |
| Common failures | Spinning page, CAPTCHAs, missing entry points | Timeouts, certificate errors, broken streams |
| Where to configure | Client-wide global proxy | Environment variables or SDK options |
On the web, static assets, APIs and streaming responses are spread across several domains, and any one of them failing to connect shows up as a stuck page. When troubleshooting, first work out whether the whole page won't load or the page loads but answers never arrive: the former is usually asset domains not going through the proxy, the latter usually a streaming connection being cut.
API calls touch only a few domains and carry little traffic, but a single request can last a long time. A read timeout set too short cuts long answers off early and looks like a server-side error; set the read timeout in the minutes range and keep a limited number of retries. APIs use the program's own network stack, so the system proxy may not apply — most SDKs honour the standard proxy environment variables, and you can also set one on the client object itself.
Command line, IDE plugins and CI: setup essentials
What developer scenarios share: there's no browser to fall back on, so whether the proxy is configured correctly shows the moment you run something. Configure the three separately; they don't interfere with each other.
Command line
Set the environment variables, then verify the exit
Most command-line tools honour the HTTPS_PROXY and ALL_PROXY environment variables. After setting them, confirm with a lightweight request that the exit has actually changed before running the real job — otherwise you'll blame the server for a failure that's yours.
export HTTPS_PROXY="http://proxy.example.com:8080"
export ALL_PROXY="http://proxy.example.com:8080"
curl -sS https://api.example.com/pingIDE plugins
Plugins don't always read the system proxy
AI plugins inside editors don't necessarily follow the system proxy. Some read a proxy field in the editor's settings, some use their own config file, and some only borrow a browser once at sign-in. Check each one; don't assume the global proxy covers them. Plugins usually hold one long connection for completions, so when an intermediary resets it, the symptom is completions that come and go rather than nothing working at all.
CI and headless environments
Pin the exit
CI has no browser, only environment variables and config files. Put the proxy address into the pipeline's environment variables and inject secrets through the pipeline's encrypted variables — never commit them to the repo. Batch jobs need a pinned exit: rotating one batch across several exits makes the platform side more likely to flag it as unusual, while a single dedicated exit makes the failure rate more predictable. Don't print full request headers or secrets in logs.
Common failures and their causes
Locate the layer from the symptom first, then touch the config. Get the order wrong and ten new routes will still produce the same error.
| Symptom | Likely cause | What to do first |
|---|---|---|
| Page spins forever, assets stall halfway | Static asset domains aren't going through the proxy, or the route is dropping packets | Retest with another route type, then check whether the client proxies everything |
| The answer cuts off halfway | Long connection reset, NAT table timeout, exit switched mid-stream | Pin the exit, turn off auto-switching, raise the read timeout |
| Constant verification prompts | Exit address shared by too many users, or frequent region switching | Pin one route and cut down region hopping |
| Missing feature entry points, shorter model list | Detected exit region doesn't match the account region | Check where the exit is registered and go back to the region used at sign-up |
| API returns certificate errors or times out | The program isn't using the proxy, or a middlebox is swapping the certificate chain | Check environment variables and self-signed certificate settings |
| Completions come and go | The plugin isn't proxied, or the long connection is being reset | Configure the plugin's proxy separately; switch to a low-loss route |
| Image job fails after queuing | The connection drops while waiting in the queue | Use a stable dedicated line and don't switch routes while waiting |
Pick by scenario: recommendations and entry points
Split usage into four categories and route choice stops being a struggle. Every scenario can start on the entry-level monthly plan and upgrade later; the price difference is prorated by remaining days.
Everyday chat and search
A relay route is enough. Bandwidth needs are low; what matters is a stable exit that doesn't hop around. The ¥9.9 monthly plan with 60GB is a fine way to get started.
Long documents, code and image generation
An IEPL dedicated line fits better. It doesn't compete with the public internet for bandwidth, fluctuates less at peak hours, and long connections rarely get cut mid-stream — even a very long single request holds.
Multiple devices and family sharing
No device limit, but keep your usual devices pinned to one region. The more devices you have, the more you want to avoid the same account hopping between countries.
Developers and batch jobs
Pin the exit first. Run main tasks on a dedicated line and keep a second route for CI; the two don't interfere, and when something fails it's easier to tell which side broke.
Monthly · 60GB
¥9.9 / month
Monthly · 250GB
¥18 / month
Monthly · 500GB
¥28 / month
Traffic packs come in ¥158/300GB, ¥358/1000GB and ¥658/3000GB — use them until they run out, they never expire. Monthly plan traffic resets each month on the activation date, and mid-term upgrades are prorated into remaining days. Clients cover Windows, macOS, iOS, Android and Linux; sign in and import your subscription.
AI acceleration FAQ
The questions we get asked most, with numbers that match the plans page.
An AI tool won't open or keeps spinning — what should you check first?
First work out whether the whole page won't load, or the page loads and only the answer never arrives. A whole page failing is usually static asset domains not going through the proxy; a page that loads but never answers is usually a streaming connection being cut. Once both are ruled out, retest with a different route type.
Why does switching routes trigger verification again?
A platform's region judgement follows the exit IP. Change the exit region and the session counts as a new regional visit, so extra verification is a normal response. Pinning your usual devices to one route noticeably reduces these interruptions.
Do API calls need a different route from the web app?
Not necessarily, but it's worth separating them. Web traffic is heavy and spans many domains; API traffic is light with long single requests. Running both over the same dedicated line usually works fine; if your API jobs need high stability, give them their own route so the two don't disturb each other.
How do you pin the exit region for developer scenarios?
Hard-code the proxy address of one route in environment variables or SDK options — don't use a config that switches automatically. In CI, inject the proxy address as a pipeline variable and secrets as encrypted variables too; never write them into the repo.
Do traffic packs expire?
No. Traffic packs are used until they run out and never expire: ¥158 for 300GB, ¥358 for 1000GB, ¥658 for 3000GB. When your usage swings a lot, they suit better than a monthly plan.
How does the refund work?
14-day money-back guarantee, no questions asked. Apply within the window and the refund goes back through the original payment channel. Payment methods include Alipay, WeChat Pay and USDT.