Free tools Windows power users keep installed
One-click scans. No signup required.
Harun says he built KODA, an AI coding mentor, with a single 92-kilobyte HTML frontend and no frontend build step. That figure describes the client-side file—not the whole app. The project also uses hosted services and a separate backend, and its choices reflect one practical constraint: Harun was coding on a phone.
What the 92KB figure includes
In his October 2026 DEV Community post, Harun describes the frontend as one index.html file containing HTML, CSS, and vanilla JavaScript. He reports approximately 500 lines of CSS, 200 lines of static HTML, and 1,400 lines of JavaScript. Elsewhere in the post, he calls the app about 1,900 lines of JavaScript; he does not explain the difference between that figure and the breakdown.
As an Amazon Associate I earn from qualifying purchases.
The 92KB figure is Harun’s reported size for the frontend. It does not mean that the AI model, authentication, database, or server-side logic is packed into that file. Nor does it establish a measured loading-speed advantage over a framework-based version: the post offers no such comparison.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchHow the rest of KODA is put together
Harun describes KODA as an AI coding mentor that streams replies, reads images, exports conversations, supports authentication, and applies an 11-article safety framework. His reported stack separates the browser interface from the services that power the app:
#1 Best Overall
| Part | Harun’s reported choice | Role described in the post |
|---|---|---|
| Frontend | Single HTML file with vanilla JavaScript and CSS | Interface and browser-side behavior |
| Static hosting | Netlify | Serves the frontend |
| Backend logic | Cloudflare Worker | Prompt screening, model selection, and response streaming |
| Database and authentication | Supabase | Data and user sign-in |
| Language model | Groq | Generates mentor responses |
| Optional web search | Tavily | Searches the web when needed |
These are the integrations Harun reports, not an independent verification of the services or their configuration. “No build step” applies to his frontend workflow; it does not mean the product has no backend, deployment pipeline, or hosted dependencies.
Why he chose vanilla JavaScript
Harun says he was coding on a phone and found a local Node-based toolchain difficult to run there. Keeping the frontend in one file and relying on browser APIs suited that constraint. He says the app uses two main rendering functions—one for messages and one for the conversation list—and redraws the relevant interface when state changes.
That is a project-specific tradeoff, not evidence that React or another framework is unnecessary. A compact, self-contained file may suit a small interface or a constrained development setup. As features and contributors grow, keeping rendering, state, and event handling organized can become harder. The post does not compare frameworks, maintenance effort, bundle size, or performance, so it cannot establish which approach is best for another project.
What the streaming example gets right—and leaves out
The post illustrates streaming with fetch(), res.body.getReader(), and decoded chunks. That shows the basic browser mechanism, but the short example processes incoming chunks directly. Network chunks are arbitrary pieces of data: they may split a server-sent event line or even a multibyte character. Treating each chunk as a complete JSON message can therefore fail.
A commenter raised this boundary problem. Harun replied that the production approach should buffer data until complete lines are available and use TextDecoder with {"stream":true} so text decoding can span chunks. He said he planned to implement that correction. The distinction matters: the posted snippet demonstrates the idea of reading a stream, but it should not be taken as a complete, robust parser.
What the safety claims do—and do not—show
Harun says the Worker screens for toxic prompts, including encoded inputs, selects a model chain, adds a system prompt, can run Tavily searches, and streams responses with citations. He also describes an 11-article “constitutional” framework. These are features he says he implemented; the post does not provide code, a threat model, test methodology, evaluation results, or an independent security review.
Rank #4
That evidence is not enough to establish how well the controls resist prompt attacks or other misuse. A system prompt and input screening may be parts of an application’s design, but their presence alone does not demonstrate effective protection. Treat the safety description as an account of intended behavior, not a verified security guarantee.
Recommended Free Tools
How to interpret the reported zero-dollar cost
Harun reports a total cost of zero dollars per month for the listed services in his setup at publication. The post does not specify usage, quotas, a pricing-check date, or whether the free tiers cover the app’s deployment and inference needs. It is therefore not a current price quote or a promise that a similar app will remain free.
Best Value
For a project of your own, check the providers’ current limits and terms against expected traffic, model usage, storage, and search volume. “Free” at one scale may not mean free at another.
What to take from Harun’s build
- Match the stack to the constraint. A phone-first workflow was central to Harun’s choice, so the setup may be most relevant to developers facing a similar toolchain limitation.
- Keep the system boundary clear. A small single-file frontend can still depend on separate services for model access, authentication, data, and backend logic.
- Make streaming parsers boundary-aware. Buffer partial records and decode incrementally rather than assuming one network chunk equals one complete message.
- Separate a feature claim from evidence of safety. The described screening and prompt framework are not independently validated in the post.
- Evaluate maintainability for your own app. The article reports one implementation, not a tested comparison with frameworks or other architectures.
Harun’s closing phrase—“92 kilobytes. zero dependencies. zero excuses.”—captures the appeal of the build. Read it as his project’s slogan, not a claim that the full AI application has no external dependencies: the reported architecture relies on several hosted services.
Harun’s DEV Community post includes a try-it invitation at koda-aicodementor.netlify.app.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
