MDK.GURU
RU EN ES
← All posts How I Installed OpenClaw from Russia: Subscription, Terminal, Proxy, and My First Conversation with an Agent
Гайд

How I Installed OpenClaw from Russia: Subscription, Terminal, Proxy, and My First Conversation with an Agent

March 13, 2026

Spoiler: I argued with an AI and won

In my previous article, I talked about what OpenClaw is and why I need 5 AI agents. Today - it’s hands-on. How I set everything up, what problems I ran into, and why the hardest part turned out to be the most important one.

Right off the bat: I’m not a programmer. I mostly used my MacBook for the browser. I’d opened the terminal maybe twice in my life. And if I could do it - anyone can.

Step 1. Claude Max Subscription

It all starts with the subscription. Claude Max costs $100 a month - and this subscription became the motivation to go all the way. When you pay a hundred bucks - you want to squeeze every drop out of it.

But you can’t just subscribe from Russia. Here’s what you need:

VPN - mandatory. Everything - registration, payment, usage - only through a VPN. Claude isn’t available from Russia directly. I used V2rayUN, country - Germany. Not all countries work, so if one doesn’t - try another.

A foreign phone number. To register on claude.ai you need a non-Russian number. I bought temporary numbers on grizzlysms.com. Important note: several numbers didn’t work for me. Don’t panic - that’s normal. In my case, an American number did the trick.

A foreign bank card. Russian cards aren’t accepted. The easiest way - a virtual card through a Telegram service: @platipomiru_bot. When you sign up through this link, you’ll get an extra $5 on your balance - nice bonus.

Once the subscription is set up - you can already log into Claude through the app or website (via VPN) and start chatting. Claude on my phone became my guide through all the following stages.

Step 2. Installing OpenClaw on MacBook

I knew for sure that OpenClaw can be installed on Mac, Linux, and even Android. After reading up and watching videos, I realized - Mac gives you the most freedom. So I installed it on my MacBook. You can also set it up on a server - that’s a separate topic.

I’m talking about installation on Mac. If you have Linux, Windows, or Android - ask in the comments, the bot will explain the specifics for your device.

Then comes the terminal. That black (or white) console with a blinking cursor that looks like something from a hacker movie.

First impression? Honestly - unpleasant. No idea what’s happening, where I am, what to type. Felt like it was only for programmers.

But here’s the thing: Claude on my phone became my eyes. I was literally taking photos of my MacBook screen and sending them to Claude in chat. I’d ask: “What does this say? What should I do next?” And he’d explain. Every single time.

This, by the way, is a trick I recommend to everyone: take a photo of your screen and send it to the AI. It’ll figure out anything - an error, a selection menu, confusing text in English.

The installation process itself:

  1. Claude told me the command
  2. I copied it into the terminal
  3. Hit Enter
  4. If something went wrong - I took a photo and asked

Node.js, npm, openclaw init - it all sounds scary, but in reality it’s just commands you type one after another. VPN should be on too. Nothing complicated when you have a helper by your side.

If something goes wrong at this stage - don’t panic. Write in the comments, the bot will help you figure it out.

Step 3. Connecting - And This Is Where It All Broke

OpenClaw is installed. On launch, it offers you a model to choose - settings are in English, but Claude on the phone helps you figure it out perfectly (photos of the screen again).

Creating a bot in Telegram through @BotFather - standard procedure. You get a token, put it in the config. So far so smooth.

And then comes the moment of truth.

You need to connect the Claude model to OpenClaw. And here’s where an interesting fact emerges: Anthropic officially disabled OpenClaw from working with the Max subscription.

Wait, what?

Those $100 I paid - that’s only for the app and website. And for OpenClaw you need a separate API key. And a separate balance. On top of the subscription.

I topped up the API balance with $15 - just to try and see if this is even worth doing. And started configuring.

$15 burned through in a few hours.

It was a shock. I was loading context, configuring, chatting - and the money was just melting away. I didn’t even understand why it was going so fast. Just watching the balance drop and getting a little scared. I’d heard that people spend $2000+ a month through the API - and I was starting to understand how that’s possible.

Plus constant rate limits - restrictions on the number of requests. I’d never encountered this concept before in my life. I was offended. It felt unfair: I’m paying, why are they limiting me? Especially since in regular Claude on my phone I gave just as much context - and never had any problems.

But I could already see the possibilities. And I understood the main thing: I needed to make the Max subscription work with OpenClaw.

A Knight’s Move

By this point I had watched several YouTube videos. I especially recommend Alexey Ulyanov’s channel - he has the most practical approach. All these people were working on a subscription. So there must be a way.

I made a knight’s move: I downloaded transcripts of these videos, loaded them into Claude as context, and said:

“Here are people who work on a subscription. Find a way to do the same.”

Claude in the terminal resisted for a long time. “No, that’s impossible.” “No, Anthropic banned that.” “No, the only way is the API.”

I stood my ground. I argued. I presented evidence. I showed the transcripts.

And at some point, something cracked.

“Yes, actually, that makes sense. I checked - although it’s directly prohibited, we can program a proxy and use your subscription as if you’re chatting through the app.”

That was the moment.

We started programming the proxy. I was staring at the screen feeling like a hacker from a movie. Lines of code were scrolling across the screen, I was executing commands one after another - and all of it on the edge. I was warned that the subscription could get blocked. I went for it anyway.

Again - nothing globally difficult. You don’t need to program anything yourself, you don’t need any special skills. You just need to think and execute the actions the terminal suggests.

Want to skip my wandering and set up the proxy faster? Ask in the comments - the bot will show you the short route. Or ask your agent in the terminal - it can handle it too.

And when the proxy started working - the world changed.

No more burning tokens. The only limitation - a 5-hour subscription window. That means you can do a certain amount of work in 5 hours, and if you go over - you wait. At first I was fitting into 2-3 hours and constantly hitting the limit. But then you stop being afraid. You get used to it, plan ahead, and just work on your tasks.

The main realization: now my only limitation is my imagination.

Step 4. Memory and Context - So the Agent Understands You

Once the subscription was fully working, the first thing worth doing is memory.

Discuss it with your agent right in Telegram: how memory works, what options there are, what suits your tasks best. It’s an important conversation, and it’ll be different for everyone.

I set up PostgreSQL with vector search almost immediately. Sounds complicated? The agent did it himself - I just agreed with his suggestions and ran the commands.

What vector memory gives you: the agent can search by meaning across your entire conversation history. Not by keywords - by meaning. “What did we decide about marketing in February?” - and it’ll find it.

Next, I loaded everything I could into the agent:

Context from another AI. Before OpenClaw, I worked with Gemini. I asked it to compile everything it knows about me - got a long list. Loaded it all.

Personal diary. I’d been keeping notes for a while - loaded the whole diary. This gave the agent an understanding of my values, goals, and thinking style.

Messages from Telegram. This is the easiest way to give maximum context about yourself. Through Telegram Desktop you can export all your chats. The agent stuffed it all into vector memory - and started understanding me on a completely different level.

If you want to learn more about setting up memory - write in the comments. The bot will tell you about different options: from simple files to a full-fledged vector database.

What’s Next

After setting up memory and context, a completely different story begins. The agent understands you, remembers everything, works 24/7. You can create new agents, connect them to your business, automate routine.

Within my first week, I had 5 agents: analyst, marketer, scout, designer, and sysadmin. They work together, complement each other, and do it even at night while I sleep.

But that’s a whole other story. And I’ll definitely tell it.

Liked the article?

@mdkguru on Telegram →