If you’re paying $20 a month for Cursor or $10 a month for Copilot, I’ve got good news. You can get an AI coding assistant inside VS Code powered by some of the best free AI models on the planet for $0 a month. In this video, I’m walking you through the entire setup step-by-step with nothing skipped. By the end, you’ll have Open Router connected to VS Code through the client extension, a free model picked out, and you’ll know exactly what to type for your first AI built project. Let’s do it. Here’s the problem with the paid coding assistants. Cursor Pro is $20 a month. GitHub Copilot Pro is $10 a month, and it’s now billed as usage-based AI credits. Copilot Pro Plus is $39 a month for heavier workloads. A subscription rents you models by the month, whether you code every day or twice a week, the bill is the same. But here’s what they don’t advertise. The underlying models cost fractions of a penny per task when you pay per use instead of per month. So, let’s cut out the subscription entirely. Here’s the replacement stack, and every piece of it is free. VS Code, the free code editor from Microsoft. Kline, a free extension that puts an AI coding agent right in your sidebar. And Open Router, which gives you one API key that unlocks more than 300 AI models, including a rotating set of free ones. Total cost, $0 a month. The only time you ever spend money is if you personally choose a premium model, and even then, we’re talking pennies, not subscriptions. So, what is Open Router? Think of it as a universal remote for AI models. One API key gives you access to more than 300 models. Claude, GPT, Gemini, Deep Seek, Qwen, Llama, all through the same key. Any model with column free at the end of its name costs $0, and you don’t even need a credit card. Paid models are strictly pay-as-you-go, many under a dollar per million tokens. The free tier gives you 20 requests per minute and 50 per day, and that jumps to 1,000 per day after a one-time $10 credit purchase. Free models rotate over time, so check the live list on their models page and filter by free. And what about Klein? It’s a free VS Code extension that puts an AI agent in your sidebar. This isn’t autocomplete. It’s agentic. It reads, writes, and edits files across your entire project. It runs terminal commands, always asking your approval first. You give it a goal like, “Build me a landing page.” It makes a plan, and then it executes that plan step by step. Here’s the key idea. Klein is the driver, and the model you pick on OpenRouter is the engine. If you prefer, Ruca is a solid alternative that works the same way. Same setup steps, different icon. Here’s the full roadmap before we dive in. Six steps, about 10 minutes total. One, install VS Code. Two, install the Klein extension. Three, create a free OpenRouter account. Four, create an API key. Five, connect Klein to OpenRouter. Six, pick a free model and start building. Follow along and pause the video whenever you need to. Every step is on screen. Step one, install VS Code. Go to code.visualstudio.com. Download the version for your system and run the installer. The default settings are completely fine. Open VS Code once afterwards so it finishes its first-time setup. If you already have VS Code installed, skip ahead to step two. Step two, install Klein. In VS Code, click the extensions icon in the sidebar or press control shift X, command shift X on a Mac. Search for Klein and make sure the publisher is Klein itself. Then click install. When it’s done, you’ll see a new Klein icon in your sidebar. That’s where your AI agent lives from now on. If you’d rather use RuCode, the steps are identical. Just search for RuCode instead. Step three, your OpenRouter account. Go to openrouter.ai and click sign in. Signing up with Google or GitHub is the fastest path. You do not need a credit card. The free models work without one. Once you’re in, you’re on the free tier. 50 free model requests per day, which is plenty to get a feel for everything. Step four, create your API key. In OpenRouter, open the keys page and click create key. Give it a clear name, like VS Code client, so you remember what it’s for. Copy the key immediately. It only shows once, and if you lose it, you’ll have to make a new one. Treat this key like a password. Never share it. Never post a screenshot of it. Never paste it into a chat window. Step five, connect the two. Click the client icon in your VS Code sidebar, then click the little gear icon to open client settings. Find the API provider drop-down and choose OpenRouter. Paste your API key into the OpenRouter API key field, then click done. That’s it. Client is now connected to more than 300 models through your one key. Step six, pick your model. In client’s model drop-down, choose any model ending in colon free. Those cost $0. My recommendation to start, Qwen 3 coder, the free version. The exact model name is on screen, which is arguably the best free coding model available right now with a huge context window. Copy the exact model name from OpenRouter’s model list if you don’t see it in the drop-down. And remember, you can switch models anytime from that same drop-down. No reconnecting, no new keys. Let me save you the research and give you the best free coding models right now. All on screen with their exact names. Qwench 3 Coder Free, the best free coding model with a massive context window. Open AI’s GPT OSS 120B Free, strong reasoning and great tool use. Nvidia’s Nemotron 3 Nano Free, fast with long context. Google’s Gemma 4 31B Free, a solid all-rounder. And ZAI’s GLM 4.5 Air Free, great for reasoning and agentic tasks. There’s also OpenRouter/free, an auto router that picks a free model for you automatically. One warning, free models rotate over time, so recheck the models page and filter by free every few weeks. Free isn’t unlimited, so know the limits going in. Free models allow 20 requests per minute and 50 requests per day if you’ve never bought credits. Make a one-time $10 credit purchase, which you can spend on paid models later, and that daily cap jumps to 1,000. Free endpoints can also be slower or queued at peak times. That’s the trade-off for zero dollars. And if you ever see a 429 error, that’s just the rate limit talking. Wait a minute and try again. Now, when should you actually spend money? When a task is genuinely hard and you want a stronger model, Gemini 2.5 Flash costs about 10 cents per million tokens. That’s insanely cheap. DeepSeek V3 is about 50 cents per million, and it’s an excellent coder. Claude Sonic class models run a few dollars per million. Reserve those for the genuinely hard problems. Realistically, $5 of credits can last a casual coder a month or more. My strategy? Free models for 80%of the work. Cheap paid models for the tricky 20%. Even then, you’re spending a few dollars a month, not 20. Time for your first build. In VS Code, go to file, open folder. An empty folder is perfect because you get to watch everything appear from nothing. Click the client icon and describe your goal in plain English. For example, build me a one-page portfolio site with HTML, CSS, and JavaScript. Client will make a plan, then ask your approval before creating files or running commands. Review each step as it goes, approve what looks right, and correct it when it doesn’t. You’re the boss, it’s the intern. A few pro tips that took me a while to learn. One, add a .client rules file to your project. It’s a list of instructions client follows every single time, like your coding style or project rules. Two, match the model to the task. Fast free models for drafts and boilerplate, stronger models for debugging. Three, start a fresh task for each new feature because clean context means better results. Four, always review the diff before approving file changes. And five, keep tasks small and specific. AI agents do their best work with clear, narrow goals. Quick safety talk because this matters. Never paste your API key into a chat, a video, a live stream, or a screenshot. Bots scrape the internet for leaked keys within minutes. If a key ever does leak, delete it on OpenRouter’s keys page and create a fresh one. It takes 30 seconds. Set a spending limit in OpenRouter so paid models can never surprise you with a bill. And use one key per tool. Name it VS Code Client, for example, so you can revoke keys individually without breaking everything else. If something breaks, here’s the quick fix list. Invalid API key almost always means a bad copy paste. Recopy it from OpenRouter’s keys page and watch for extra spaces at the ends. 429 errors mean you hit the free rate limit. Wait a minute or make that one-time $10 credit purchase to raise your cap. Model not found means that free model rotated out. Pick a current colon free model from the list. If the client icon disappears, disable and re-enable the extension, then reload VS Code. And if responses are slow, free endpoints queue up at peak times, just try a different free model. Let’s recap. One, install VS Code. Two, install client. Three, create your OpenRouter account. Four, create an API key. Five, connect client to OpenRouter in the settings. Six, pick a colon free model and start building. That’s Cursor Pro at $20 a month replaced with this setup at $0 a month. Optionally, keep $5 to $10 in credits around for premium models on the genuinely hard tasks. Same agentic coding power, you just stopped renting it by the month. That’s the whole setup. Six steps, $0 a month. Now it’s your turn. Go build something, then come back and drop a comment telling me what you made on your free setup. I read every single one. And if you want more videos like this, showing you how to stop overpaying for AI tools, hit subscribe. I’ll see you in the next one.