{
  "id": 20672966,
  "title": "How I built 80+ free browser tools, including AI that never uploads your files",
  "url": "https://urgent.news/2026/11/18/how-i-built-80-free-browser-tools-including-ai-that-never-uploads",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-18T23:26:47.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/smartonlinetools/how-i-built-80-free-browser-tools-including-ai-that-never-uploads-your-files-1f8h"
  },
  "original_language": "en",
  "account": "The author sought to create a suite of everyday online tools that could be used without uploading files to external servers. These tools, known as Smart Online Tools, include 80+ free functionalities such as PDF merging, image compression, text extraction from screenshots, and various AI-driven features like OCR, speech-to-text, translation, background removal, face blurring, and alt-text generation. All processing occurs within the user's web browser.\n\nBuilding this suite presented several challenges. The author chose to use Next.js with an App Router, allowing for a static export with no backend involved. The entire site is composed of HTML, JavaScript, and WebAssembly, served from regular shared hosting.\n\nTo run AI models directly in the browser, the stack incorporated Transformers.js for ONNX models, WebAssembly, or WebGPU where supported. Tesseract.js was used for OCR, while MediaPipe Tasks Vision facilitated face detection. Web Workers were employed to prevent page freezing during model processing.\n\nEach model download ranges from 1 MB to 250 MB, with permissive licenses like MIT, Apache-2.0, or CC BY 4.0 chosen to avoid licensing issues. The author intentionally excluded models with less permissive licenses, such as GPL dependencies found in some text-to-speech models. For text-to-speech capabilities, the browser's built-in device voices are utilized instead.\n\nModels are loaded on-demand, with a single worker managing a cache of loaded models. This setup ensures that subsequent uses are instantaneous. The implementation uses a Map object to store and retrieve model functions based on their key names.\n\nThe code snippet demonstrates how the author loads a specific translator model from Hugging Face's model hub, using a quantized version for speed and falling back to full precision if necessary. The translator routes through English when direct models are not available. Long text is divided into smaller chunks to stay within model limits, and a progress bar reflects the translation process.\n\nThe author emphasizes the importance of informing users about model downloads, noting that a 100 MB model can feel broken without warning. Each tool explicitly states the size of the first download and informs users that the browser will cache it. Workers are deemed essential to prevent main thread freezing. For long text inputs, the author splits the text into sentence groups of about 400 characters to avoid exceeding model limits and keep the progress bar moving.\n\nSeveral lessons were learned during the development process. Clear communication about model downloads is crucial; users should be aware of the size and that it will be cached. Workers are non-negotiable for maintaining a responsive user experience. Small models offer focused capabilities but lack the accuracy of larger cloud-based services. The UI should indicate the capabilities and limitations of each model, allowing users to edit results if needed. Checking model licenses beforehand is vital to avoid future issues. The author also highlights that static hosting suffices for this setup, eliminating the need for GPU servers or API bills, as the heavy lifting is performed on the user's device, preserving file privacy. The AI tools are available at smartonline.tools/ai-tools and are free to use without any account requirements, ensuring that files remain on the user's device. The author invites feedback on performance, accuracy, and suggestions for new tools.",
  "summary": "I wanted a set of everyday tools — merge a PDF, compress an image, read text from a screenshot — that I could use without uploading my files to someone else's server. So I built Smart Online Tools : 80+ free tools where all the processing happens in the browser. The PDF and image tools were the easy part. The interesting part was the AI tools: OCR, speech-to-text, translation, background removal,…",
  "key_points": [
    "Author created 80+ free browser tools without file uploads",
    "Utilized Next.js, WebAssembly, and WebGPU for AI processing",
    "Emphasized transparent model downloads and browser caching"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}