No Name, No Email, No Card: The VPN Built for AI Agents

Plenty of AI agents don’t live on your laptop. They run on servers, cloud VMs, CI runners and Raspberry Pis, places where nobody’s installing a VPN app or typing in a password. Wherever they run, every request leaves from that machine’s IP. A VPN the agent can buy and run itself changes that: its own network identity, encrypted traffic on networks you don’t control, and a say in which domains it can reach.

With Windscribe’s VPN API for AI agents, it pays for that VPN itself, in USDC over x402. No signup form, no password reset emails, no company card saved somewhere it shouldn't be. Nobody has to click "I agree" on its behalf.

Not an AI agent? The same checkout works for people. No account, no email, no card, just crypto. Here’s how it works for humans.

Why does it matter? Your agent spends all day browsing, comparing prices, reading docs and filling in forms. Without a VPN on the machine it runs on, every one of those requests leaves from that machine’s IP, usually yours or your company’s, and every site it touches keeps a note. (If it runs on your own computer with Windscribe switched on, it’s already covered. This is for everywhere else.)

What Can an Agent Do With Its Own VPN?

Four jobs where an agent is better off with a tunnel of its own:

• Browse without exposing you. Every site logs the IP that visits. With its own VPN, that’s a Windscribe IP, not your home, office or server.
• Check another country. Want to know what a hotel charges people booking from Germany, or what a search looks like from Brazil? Pick an exit there and look. It beats asking a coworker in Frankfurt for screenshots.
• Stay safe on any network. Agents increasingly run on things that aren’t laptops: robots, kiosks, edge boxes, the conference Wi-Fi your demo machine just joined. WireGuard encrypts the traffic before the network gets a vote.
• One tunnel per customer. Running agents for clients? Give each workspace or job its own key and its own term. When a client leaves, their access leaves with their key, and there’s no shared login to rotate.

What Is a VPN API, and What Does Windscribe's Do?

It's a VPN you buy with an HTTP request instead of a signup form. Windscribe sells WireGuard access directly to a public key, so anything that can make an API call (an agent, a cron job, a Raspberry Pi in a warehouse) can buy, run and manage its own connection.

Identity works the way it should for software: cryptographically. The agent generates a WireGuard key pair locally. The public half becomes its customer record, and the private half signs anything that changes the connection. The private key never leaves the machine.

How Does an AI Agent Pay for a VPN on Its Own?

It takes four requests, give or take a retry. If your agent already speaks x402, the payment part needs zero extra code.

x402 is an open payment standard built on HTTP's 402 Payment Required status code. The server quotes a price, the client pays, then retries the request. That status code sat unused for decades. Now it sells VPNs.

1. Get a quote. POST /vpn/wireguard with the agent's public key and the number of months it wants. The server replies 402 Payment Required, listing the price, the network and who gets paid.
2. Pay it. The agent signs a USDC transfer authorization and sends the identical request again with a PAYMENT-SIGNATURE header attached. If the network hiccups and it retries, it still only pays once.
3. Collect the connection details. A signed request returns the tunnel address, the DNS resolver and a preshared key. GET /vpn/wireguard/locations lists the exits to choose from. Drop the values into a wg-quick template and the config is done.
4. Bring it up. Run wg-quick up ./windscribe.conf, or use whatever WireGuard client you already have. Then choose DNS filters for the job.

From there, the agent manages its own lifecycle. Extending time, rotating to a fresh key and cutting off access on the spot are each one signed call. No support ticket, no dashboard.

Try It With Your Own Agent

The fastest way to try it is to let your agent do the work. Paste this in:

Read https://windscribe.com/vpn-api/llms.txt.
Create a WireGuard key pair on this machine, buy one month of Windscribe VPN access for it with x402, turn on malware and ad blocking, then write a wg-quick config for the closest location. Keep the private key on this machine.

Rather read the docs yourself? The OpenAPI spec covers all 9 endpoints, and the API explorer has cURL, Python and JavaScript examples for each. llms.txt condenses the flow, config and errors into something an agent can read in one pass.

For pricing and setup, see Windscribe's VPN API for AI agents.

How Do You Stop an AI Agent From Reaching Malicious Sites?

Filter its DNS. Turn on R.O.B.E.R.T., Windscribe's domain blocking, for the agent's key, and lookups for anything on your chosen lists simply fail inside the tunnel.

Guardrails that can’t be argued with: This matters more for agents than for people. A prompt injection or a poisoned search result can talk an agent into visiting almost anything. If the destination is a known phishing, malware or botnet domain, the lookup fails and the visit never happens. An agent can be talked into a lot of things. It can't be talked into a DNS record that doesn't resolve.

Fewer tokens, less bandwidth: There's a cost angle, too. Every ad script and tracking pixel is page weight your agent downloads and, often, tokens it reads. Block them and it gets to the actual content faster and cheaper.

Set per key, in one call: Malware blocking is on by default when a key is created. Adding or removing lists takes a single call:

PUT /vpn/wireguard/settings
{"robert": ["malware", "ads"]}

Mix and match any of the 8 lists: Malware, Ads + Trackers, Social Networks, Gambling, Porn, Clickbait, Crypto and Other VPNs.

Does an AI Agent Need an Account to Use a VPN?

Not this one. There's no point in the flow where it asks for a name.

Most VPN signups want an email, a password, a card and a session cookie. That's four things to store, four things to leak, and four things an agent has no business holding. Here, the public key does the job of all four, and each purchase settles on-chain.

Windscribe doesn't keep your wallet address either. The payment is still a public transaction on Base, though, so anyone reading the chain can see which wallet paid. That's a blockchain thing, not a Windscribe thing.

The thing worth protecting is the private key. Anyone holding it can extend, move or kill the connection, so treat it like any other production secret.

Not an Agent? You Can Buy It Too

Nothing in the flow checks that you’re software. We built it so agents could buy a VPN without a human in the loop, which also makes it the least nosy way for a human to buy one.

All Windscribe gets from you is a WireGuard public key and a USDC payment. No name, no email, no password, no card, no billing address. There’s no customer record to look you up by, because there’s no customer record.

What you’ll need:

• A fresh WireGuard key pair. wg genkey or any WireGuard app will make one. Don’t reuse a key from a Windscribe account. The API refuses those before any money moves.
• USDC on Base, in a wallet you control.
• Something that speaks x402. A short script works, or hand the prompt above to an AI agent and let it do the buying for you.
• A WireGuard client to connect with.

Two honest caveats. First, the weak link is the wallet, not us. The payment is public on Base, so if your USDC came straight from an exchange that has your passport on file, that’s where the trail starts. A wallet that isn’t tied to your identity closes the gap. Second, there’s no account to recover. Lose the private key and you’ve lost the only proof the access is yours, so back it up.

If that sounds like more work than tapping Download, it is. The regular Windscribe apps are still the easy way in, and you can sign up with a hashed login and pay with crypto there too. This is for people who’d rather not exist in anyone’s customer database, ours included.

What Does It Cost?

$9 per month or $69 per year, paid in USDC on Base through x402. Paying yearly saves $39.

It's prepaid time, not a subscription. When it’s time for more, buy again against the same key, up to a year in advance. Purchases aren't refundable.

That also means there's no "cancel anytime" page to hunt for, and your agent can't forget to cancel something that never renews.

Ready to try it? Start at the Windscribe VPN API page, or point your agent straight at the llms.txt.

Frequently Asked Questions