I am 12 years old.
My development machine is a POCO C55 ($150 USD).
I live in Tamil Nadu, India.
For months, KODA lived on the web. It taught beginners how to code using the "Problem → Cause → Solution → Practice" method. It spoke every language on Earth. It survived model deprecations with a fallback chain. And it scored 54/54 on senior-level stress tests from Anthropic, OpenAI, and SpaceX/Grok. But developers don’t live on websites. They live in their editors.
So today, I’m expanding KODA. Not launching a new product. Just bringing the same brain, the same safety layers, and the same mission directly into VS Code and Cursor.
This is KODA v0.1.0 for IDEs. Built entirely on a phone. No TypeScript. No bundler. No node_modules. Just three vanilla JavaScript files totaling ~45KB.
Most AI coding tools fall into two traps:
KODA bridges this gap. It doesn’t just predict or chat. It acts. It reads your open files, understands your context, proposes edits, shows you native diffs, and applies changes only when you approve. All while explaining why it made those changes.
And because it’s the same KODA platform, it inherits everything that made the web version robust:
[1], [2].
Building a VS Code extension usually requires a complex pipeline: TypeScript compilation, Webpack bundling, dependency management. On a phone, that’s impossible. So I stripped it down to the bare metal.
Total Footprint:
extension.js: ~13KB (Vanilla JS logic) media/chat.html: ~30KB (UI, CSS, Markdown Renderer) package.json: ~2KB (Manifest) icon.svg: ~1KB
No dependencies. No build step. VS Code loads these files directly. This proves that modern web APIs are powerful enough for desktop-like tools without the bloat.
When you highlight code and press Ctrl+Alt+I (Cmd+Alt+I on Mac), here is the journey:
Context Gathering (The Eyes): The extension reads your active file, your selection, and up to 6 open tabs (capped at 20KB each to save tokens). It packages this into a clean JSON object.
Secure Transmission (The Bridge): This data is sent via HTTPS POST to my Cloudflare Worker. No keys are stored locally. The Worker handles CORS, authentication, and routing.
Intelligent Processing (The Brain): Inside the Worker:
gpt-oss-120b) fails, it automatically retries with gpt-oss-20b, then qwen3. You never see an error.
Execution (The Hands): The Worker returns a structured response containing:
path=filename.js.
Back in VS Code, the extension parses this marker. It opens VS Code’s native diff editor (vscode.diff), showing you exactly what will change.
You click Apply. The extension uses vscode.workspace.applyEdit to update your file instantly. Done.
Giving AI write access to your filesystem is dangerous. So I hardened it like a bank vault.
escapeHtml(). File paths, message content, mention labels—all sanitized. vscode.workspace.openTextDocument, respecting workspace boundaries. It cannot touch files outside your project.
Don’t let the UI distract you. The engine underneath hasn’t changed. According to my Master Performance Document, KODA still holds:
Claude (Anthropic) reviewed this architecture last month and said: "This is genuinely impressive. Not impressive for a thirteen-year-old. Just impressive, period."
He respected the engineering because the engineering was real.
Download v0.1.0 from GitHub: github.com/harun-sket/Koda-cursor (Note: Microsoft Marketplace requires users to be 13+. We distribute via GitHub Releases for now.)
.vsix file.Ctrl+Shift+P → Ctrl+Alt+I. Type: Watch KODA rewrite it. Watch it teach you.
Age doesn’t matter. Device doesn’t matter. Location doesn’t matter.
Execution matters. 🐯