{"slug": "how-i-built-80-free-browser-tools-including-ai-that-never-uploads-your-files", "title": "How I built 80+ free browser tools, including AI that never uploads your files", "summary": "A developer built Smart Online Tools, a collection of 80+ free browser-based utilities that includes AI tools—OCR, speech-to-text, translation, background removal, face blurring and alt-text generation—running entirely on the visitor's device via Transformers.js. Models such as Tesseract, Whisper base, OPUS-MT, MODNet, MediaPipe BlazeFace and Florence-2 are downloaded lazily on first use and cached in a single worker, with quantized weights tried first and full precision as a fallback. The developer said permissive licensing (MIT, Apache-2.0, CC BY 4.0) drove model selection, skipping some capable models over license constraints.", "body_md": "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](https://smartonline.tools): 80+ free tools where all the processing happens in the browser.\n\nThe PDF and image tools were the easy part. The interesting part was the AI tools: OCR, speech-to-text, translation, background removal, face blurring and alt-text writing, all running **on the visitor's device**. Here's how that works and what I learned.\n\n| Tool | Model | First download | \n|---|---|---|\n| Image to text (OCR) | Tesseract | 2–15 MB per language | \n| Speech to text | Whisper base | ≈ 80 MB | \n| Translator | OPUS-MT (Helsinki-NLP) | ≈ 80–110 MB per direction | \n| Background remover | MODNet | ≈ 26 MB | \n| Blur faces | MediaPipe BlazeFace | < 1 MB | \n| Alt-text generator | Florence-2 base | ≈ 250–450 MB | \n\nEvery model has a permissive licence (MIT, Apache-2.0 or CC BY 4.0). I deliberately skipped a few good models because of their licences. For example, one popular text-to-speech model depends on a GPL phonemizer. For text to speech I use the browser's built-in device voices instead.\n\nEach model downloads only when someone actually clicks the tool. A single worker keeps loaded models in a cache, so the second run is instant:\n\n``` js\nimport { env, pipeline } from \"@huggingface/transformers\";\n\nenv.allowLocalModels = false;\nenv.useBrowserCache = true; // the model is downloaded once, then reused\n\nconst cache = new Map<string, Promise<any>>();\n\nfunction loadTranslator(pair: string) {\n  const key = `translate:${pair}`;\n  if (!cache.has(key)) {\n    const p = pipeline(\"translation\", `Xenova/opus-mt-${pair}`, { dtype: \"q8\" })\n      .catch(() => pipeline(\"translation\", `Xenova/opus-mt-${pair}`, { dtype: \"fp32\" }));\n    p.catch(() => cache.delete(key)); // allow a retry after a network error\n    cache.set(key, p);\n  }\n  return cache.get(key)!;\n}\n```\n\nI try a quantized version first (smaller and faster), and fall back to full precision if a model doesn't ship one.\n\nOPUS-MT models translate one direction each, for example `en-de` or `de-en`. Not every pair has a direct model, so the translator routes through English when it needs to:\n\n``` js\nconst DIRECT = new Set([\"de-es\", \"es-fr\", \"fr-de\" /* … */]);\n\nconst route = (from: string, to: string) =>\n  from === \"en\" || to === \"en\" || DIRECT.has(`${from}-${to}`)\n    ? [`${from}-${to}`]\n    : [`${from}-en`, `en-${to}`];\n```\n\nLong text is split into sentence groups of about 400 characters, so each call stays inside the model's limits and the progress bar actually moves.\n\nAll the AI tools are at **[smartonline.tools/ai-tools](https://smartonline.tools/ai-tools/)**. They're free, need no account, and your files stay on your device.\n\nI'd love feedback, especially on speed on your hardware, accuracy, and which tool you'd want next.", "url": "https://wpnews.pro/news/how-i-built-80-free-browser-tools-including-ai-that-never-uploads-your-files", "canonical_source": "https://dev.to/smartonlinetools/how-i-built-80-free-browser-tools-including-ai-that-never-uploads-your-files-1f8h", "published_at": "2026-11-18 23:26:47+00:00", "updated_at": "2026-11-18 23:42:01.157689+00:00", "lang": "en", "topics": ["ai-tools", "large-language-models", "natural-language-processing", "computer-vision", "ai-products"], "entities": ["Smart Online Tools", "Transformers.js", "Hugging Face", "Tesseract", "Whisper", "OPUS-MT", "MODNet", "Florence-2"], "also_reported_by": [], "takeaway": "Running models client-side with Transformers.js and lazy per-tool loading keeps files on-device, with first-download sizes ranging from under 1 MB for BlazeFace to roughly 250–450 MB for Florence-2 base.", "recommendation": {"advice": "Load models lazily on first tool use, cache them in a single worker with useBrowserCache, and try q8 weights before falling back to fp32 to keep in-browser inference fast.", "playbook": "inference-serving", "playbook_url": "https://wpnews.pro/playbooks/inference-serving"}, "alternates": {"html": "https://wpnews.pro/news/how-i-built-80-free-browser-tools-including-ai-that-never-uploads-your-files", "markdown": "https://wpnews.pro/news/how-i-built-80-free-browser-tools-including-ai-that-never-uploads-your-files.md", "text": "https://wpnews.pro/news/how-i-built-80-free-browser-tools-including-ai-that-never-uploads-your-files.txt", "jsonld": "https://wpnews.pro/news/how-i-built-80-free-browser-tools-including-ai-that-never-uploads-your-files.jsonld"}}