Skip to content
13 min read—views—reactions
Workplace

Just try everything. How I came to vibe coding

I'm an artist, not a programmer. He went from pieces of code in Gemini to eight projects in production, OMP, Orca, Russian for Orca and a plugin that connects WikiVibe API to seven CLI agents. Why not look for the best tool and where to start.

On this page
Just try everything. How I came to vibe coding cover

I’m often asked, “What is the best tool to use for vibe coding?” I understand these questions - I had exactly the same in the beginning.

I won't name one now. For me, vibe coding is a pastry shop. You come in and it's full of sweets. And I want to try everything.

I like the process. I am not a programmer: there is no specialized education, I am an artist - tattoos, piercings, paintings, exhibitions. Vibe coding did not begin with learning programming, but with a desire to do something.

At first there were just chunks of code

The first tool was Gemini in the web version of AI Studio. I just sat there and made chunks of code and then put them together like bricks. So the first project appeared - hbsb.ru, the site of the art space "Assembly Point" from Perm.

I don’t remember how long it took, it seems like more than a month. Then new versions of Gemini, Flash and Pro came out, and naturally I started trying them out.

I tried everything: what one model can do, what the other can do, how each one copes with the task. Gradually, instead of individual pages, multi-page and multi-level projects, APIs, databases appeared.

And all this first time — no Harness, no CLI, no terminal. Web interface. I could take one page and stuff so much JavaScript into it that now I sometimes think, "Well, how did this work?" But it worked. And I was curious.

I had to keep in mind where a piece of code was, what it was responsible for, and on which page what was happening. For a programmer, probably nothing unusual. I am a person who has never learned this. Just an artist who decided to make a website.

Then I started trying everything

Chinese models, too. DeepSeek -- tried. GLM -- tried it. I have some AI apps on my phone: GPT, Gemini, and a bunch of things that I don't go to very often.

Here’s what I learned: If a new tool comes along, you should try it. You don't have to go straight to him. You don't have to say he's the best. You don't have to keep it at all. Just try it: you never know in advance what will come to you.

If you like to do something at Qwen, please. Perplexity, please. Like Gemini, work at Gemini. Like GPT, work at GPT.

The easiest way to start is with my phone: it’s always with me. The thought arose – I open Gemini or GPT and start spinning it. I ask something, clarify something, argue something, invent something. Until the thought begins to turn into something more or less understandable. And then I create a folder on a computer or on a VPS, run an agent there and grow that thought into something real.

Then the agents came along

I first discovered OpenClaw. I put it on the VPS and started to play: there you can adjust the identity of the agent, his parameters - in general, it was funny.

I spent a lot of money on that too. I did not know about WikiVibe at that time, there was no foreign card for payment, so I used services that promised “access to foreign models from Russia.” The prices there were just horse prices: in a couple of hours you could safely send a thousand rubles - and that's all, there is no money.

But it was also part of the learning. I kept coding my projects through Gemini and sorting out what else was out there.

Then I discovered CLI: I started putting things right on the server. Then there was Z.ai, then Codex. And gradually I realized that the web version of AI is just one way to work with the model. There are terminals, agents, VPS, different shells — and different ways to make it all work together.

And then there was WikiVibe

That's where I started moving a lot faster. When I got into the wikivibe.dev community, I started learning about tools I hadn’t even heard of before.

The first thing I remember was Crystallic talking about Cockpit. At first I was rather skeptical, then I put it up and started working. Then there was a topic with subscriptions, discussions of different models and providers. A lot of people were sitting on Cockpit and I was sitting.

Then, on the recommendation of Alexey, he discovered Sub2API: he put it on a server in Frankfurt so that the models would normally reach him and not have to constantly mess with the VPN. At one point I had a bunch of Sub2API + Cockpit + Codex.

Then Maxim Ozerov started talking about OMP - oh-my-pi, a coding agent in the terminal, 35,022 stars on GitHub. I naturally went to try. Then there was talk of Orca. And then gradually I realized that I didn’t need Codex anymore, Cockpit was no longer necessary. For myself, I left Sub2API, OMP and Orca. I use OMP almost every day.

Orca: Many agents in one window

I liked Orca for a different reason. For me, it's a work environment where you can keep a lot of agents at once. Roughly speaking, like a browser: you have a window, it has a bunch of tabs, and each tab is a separate agent. It's a scheme I'm really drawn to.

Project open: stablyai/orca - 89,569 stars on GitHub, licensed by MIT.

But there was one problem: there was no Russian language in Orca. And I thought, "Okay, I'll do it myself.".

At first, I just started translating. Then I started to understand how it all worked. Then I found other people on GitHub who also do translations and plugins, and started talking to them. He contacted the developers of Orca in Discord and sent them his work.

Now I have orca-russian - language package under Orca 1.4.222: 15 196 translations, that is more than 14 500 lines of interface in Russian, plus a search by settings that understands Russian words. MIT license, 13 stars on GitHub.

I’ve even written a separate plugin for Orca that translates those places that are hard-to-earth and are not available through a regular translation list. I asked the developers to add my translation to their app store. Silence for now. But we'll see.

The second plugin: the key to models without manual work

The second plugin came from a completely different task, and it is worth talking about it separately - it is useful for everyone who works with agents.

I have seven agents: Codex, OpenCode, Pi, oh-my-pi, Aider and Qwen Code. Each has its own configuration file, and each needs to write an API address and a key. Hands are longing, and the key is laid out on other people's files in the open.

I needed a gateway to models, and I stopped at the WikiVibe API — for Russia, in my opinion, this is the best gateway to models. Without a foreign card, without a VPN, and from one place there are GPT, and Gemini, and DeepSeek, and GLM, and models for pictures. The key is opened in the personal account: registration. I will say right away: the link is a referral – it costs nothing to anyone who comes to it, and WikiVibe pays me a little extra for it.

In order not to write the key to seven files with my hands, I made wikivibe-orca-plugin - WikiVibe Gateway, version 1.1.8. Directly from Orca: Settings → Plugins → Install plugin → Git URL

text
https://github.com/taranttoola-create/wikivibe-orca-plugin.git#v1.1.8

The plugin requests three permissions: secrets, storage and notifications:show. The key is stored in a secure Orca storage, encrypted by the system itself, and the setting is opened in the built-in Orca browser, not in the system browser.

Next, it does this: checks the key, downloads all the models available to the key, shows in advance which strings will be written in the configuration, and writes only after confirmation. Before recording puts backup copies, and if something did not like – there is a restore button, and the files are returned to their original form.

Gateway addresses to choose from: https://api.wikivibe.dev/v1 and https://vapi.wikivibe.dev/v1. In the protected setting, you can see the catalog of models and key consumption: balance, limit, balance, spending for today and for all time. Need Orca 1.4.214 or later – panel and setting I checked on 1.4.223.

If your agent is not on the list, it’s okay: the same thing is set up by hands, the documentation in the repository is detailed. I'm just more comfortable pressing two buttons.

And here I once again caught myself thinking: I am not a programmer at all. I just wanted the Russian language and went to do Russian. I wanted the key not to have to be stuffed into seven files, and I got to make a plugin.

What I like about all this

Maybe that's why I'm so excited about vibe coding. I didn't come here thinking, "Well, now I have to be a programmer." I didn't have that plan. I just wanted to do my own things: a website, an app, a service, something for my projects.

And then it turned out that you need to understand a little bit about JavaScript. Then the API. Then in the servers. Then GitHub. Then Docker. Then CLI. Then agents. Then there's how to translate the app. And you just move on, not because you have to get your degree, but because you wonder what happens if you click here.

Eight projects that are already working

I currently have eight projects in production. Here are some of them – you can open and touch them:

Project What is it Reference
RUNA NOX Telegram Mini App: runic oracle, 24 runes of the Elder Futhark, layouts, rune of the day, payment with crystals and Telegram Stars runa.kiloart.ruXX
Tarot Arcana Telegram Mini App: 22 Senior Arcana, card of the day and layout on three cards, authorization via Telegram tarot.kiloart.ruXX
TTOLALALA Telegram-bot and Mini App for a piercing and tattoo studio in Perm: services and prices, masters, portfolio, recording ttola.kiloart.ruXX
VitaRoute Medical consultation service: bot for clients, API history, adminka and web application vitaroute.kiloart.ruXX
Assemblage Point Site and online gallery of art space from Perm: event poster, catalogue of artists' works, acceptance of applications hbsb.ruXX
Assemblage point - annex Android-application of art space: general chat, profile with works, gallery, poster, generation of price tags for exhibitions. Version 1.3, Android 7 and later Open in RuStoreXX
Together 365 Partner program for debt cancellation: landing, registration of partners with training and test, personal account, admin panel. Vue 3, Vite and REST APIs on your server together365XX
Mariii Jewelry Jewelry brand website: jewelry catalog, product cards, sections about brand and delivery, ordering mariii.ruXX

Separately – open source: Russian language for Orca and plug-in gateway, which I wrote above.

And that, for me, is probably the main result of the whole story. It’s not that I learned how to code—I still don’t consider myself a programmer. It's that I'm no longer afraid of technology. If I need something, I just go and figure it out.

I even got myself a landing

At some point I realized that there was so much that I needed to make a separate place and tell what I was doing. That’s how kiloart.ru came about, where I gradually collect my projects and what I do.

And I like it too, because it turns out some strange chain: first you just want to make a site, then you learn to make sites, then you make an application, then you start to understand the servers, then you write a plugin, then you translate someone else’s project, then you communicate directly with its developers. And you just keep going.

I don’t know which instrument is the best

If you ask me now, “What is the best tool to use for vibe coding?” I probably won’t name one. Because I've been through too much myself. And to be honest, I don’t think it’s very interesting to look for one “right” tool at all.

You like one thing today. Tomorrow's different. The day after tomorrow, there's a third. You tried, you didn't like it. Okay. I tried the following.

The most important thing here is not to be afraid. If you want to do something, try:

  • So I thought, open the phone.
  • Open Gemini, GPT, Qwen, whatever.
  • Disassemble this idea with AI: ask, clarify, argue, invent.
  • Then create a folder and put an agent in there.
  • Start doing something.

Let it be crooked first. Make it clear. Don't let you know what's going on. You'll figure it out later. The main thing is to start.

And also, listen to the people around you

I'm not saying, "Do what you're told on WikiVibe." No, that's not what I mean. I mean, listen. Look at what people are doing, what they're trying, what tools they're discussing, what they've done and what they haven't. Get a vaiba.

Because for me, WikiVibe was a very useful place. I learned a huge number of tools there that I would have been looking for a long time, and many of the things that I use now, I first saw there. So the community is a respect. Alexey is a separate respect.

To anyone who is just starting out with this whole story, I would say one thing: don’t look for the perfect tool. Just try it. Because maybe you don't know what you're going to like. And maybe it's through this trial and error that you'll find not the best tool, but your own way of doing what you want.

Where to go and what to read

  • WikiVibe is the community where I learned most of the tools from this article.
  • OMP (oh-my-pi) is a terminal coding agent that I use almost every day; WikiVibe has big-guide.
  • Orca and its repository on GitHub work environment with a bunch of agents in one window.
  • Sub2API, my gateway to models, lives on a server in Frankfurt.
  • OpenClaw was the first agent I put on VPS.
  • Russian for Orca is my language package and plugin, MIT.
  • WikiVibe Gateway for Orca is a plugin that prescribes the WikiVibe API into seven CLI agents, version 1.1.8.
  • kiloart.ru is my projects and services if you need a website, bot or application.

** About interests and numbers. All projects and repositories on the lists, except for community tools (WikiVibe, OMP, Orca, Sub2API, OpenClaw), are my own: I have done them and continue to support them. The link to WikiVibe registration is a referral: whoever comes to it costs nothing, and WikiVibe pays me a little extra for it. There are no other paid placements and sponsorship materials in the article. Data on stars, versions and licenses are taken from GitHub and my repositories as of October 11, 2026.

Reactions

—reactions