Private AI tools that run on your device
Speech-to-text, translation, summarization, image captioning, upscaling and more — powered by AI models that run locally in your browser. Your files and text are never uploaded.
Speech to Text (Whisper)
Transcribe or translate speech from your mic or audio/video files with Whisper, on-device. TXT & SRT.
AI Translator
Translate text between 11 European languages with Opus-MT models that run privately in your browser.
AI Text Summarizer
Summarize long English articles into short, medium or long summaries with DistilBART, on your device.
AI tools that run on your own device
AI Tools brings modern machine-learning models straight into your browser. Instead of uploading your recordings and documents to a cloud service, each tool downloads an open model and runs it locally on your computer or phone. The speech-to-text tool uses OpenAI's Whisper to transcribe audio and video files or microphone recordings in dozens of languages, can translate speech straight into English and exports plain text or SRT subtitles. The AI translator uses Opus-MT neural translation models to translate between English, Italian, Spanish, German, French, Dutch, Portuguese, Swedish, Polish, Ukrainian and Russian. The text summarizer condenses long English articles and reports into short, medium or long summaries with DistilBART.
There is no sign-up, no usage quota and no subscription. Because the work happens on your hardware, the tools are also a good fit for confidential material such as interviews, legal notes, medical dictation or unpublished drafts that should never leave your machine.
WebGPU, WebAssembly and model caching
Every model runs in a background Web Worker powered by Transformers.js and ONNX Runtime, so the page stays responsive while it works. When your browser and graphics card support WebGPU, compatible models run on the GPU for a large speed-up; otherwise they fall back automatically to WebAssembly on the CPU, which works in every modern browser. A small badge in each tool shows which engine is being used.
Models are downloaded once from Hugging Face, with a progress bar showing how much is left, and are then stored in your browser's cache. Later visits load them from disk in seconds. Sizes range from about 77 MB for Whisper base and around 100 MB per translation language pair to about 285 MB for the summarizer. You can free the space at any time by clearing this site's data in your browser settings. Your audio and text are never uploaded, logged or stored — they exist only in the tab you are using and disappear when you close it.