Documentation
Guides, privacy notes, and practical workflows for ASD123.ai's browser-based text, context, and chat tools.
Optimizer
Clean and standardize AI-generated text before publishing, pasting into business tools, or adapting content for regional conventions.
Language Mapping
Region-specific mappings for Swiss German, German, French, Italian, and English variants.
Text Cleanup
Remove citations, Markdown, diacritics, special spaces, and stylized Unicode text.
Local Preferences
Settings can be remembered locally in your browser. Text content is not stored by ASD123.ai.
Anonymizer
Explore local PII detection, reversible placeholder mappings, and LLM deanonymization workflows. Treat this tool as an experimental proof of concept and review aid.
Regex Default
Fast local pattern matching for emails, phone numbers, IBANs, IDs, URLs, and other structured PII.
OpenAI Privacy Filter
Optional local AI detection via Hugging Face Transformers.js. First load is about 950 MB and requires WebGPU.
Legacy PoC AI4Privacy
Available for comparison, but no longer the recommended AI path.
Local AI Chat
Chat with WebGPU models directly in your browser. Prompts, uploads, generated text, and chat history stay on your device.
Local WebGPU Models
Load Gemma 4 and Qwen3.5 models in the browser, plus an experimental Gemma 4 E2B Turbo build with WebGPU shaders tuned for Apple Silicon. Model files are downloaded, but prompts are not sent for remote inference.
Files and Images
Attach images, text, Markdown, PDF, or DOCX files. Extraction happens locally before content is added to the prompt context.
History and Export
Chats are stored in browser IndexedDB, can be deleted locally, and can be exported as Markdown.
Context Estimator
Estimate how much AI chat context you need for pasted text or local documents before choosing 4K, 8K, 16K, or larger model settings.
Context Recommendation
Compare 4K through 128K windows and see which one leaves enough room for an answer.
Local File Extraction
TXT, DOCX, and PDF files are read in the browser. File contents are not uploaded for processing.
Model Profiles
Generic, Llama / Mistral, Qwen, Gemma, and GPT-style profiles keep the estimate fast without tokenizer downloads.
Markdown Converter
Turn PDF, DOCX, PPTX, Excel, CSV, and HTML documents into clean Markdown locally. Useful before feeding text into AI chats, the Optimizer, or the Anonymizer.
Many input formats
PDF (pdf.js), DOCX (mammoth.js), PPTX (JSZip), Excel/CSV (SheetJS), and HTML — each converted to Markdown headings, lists, links, and tables.
Selectable parsing engines
Optional EdgeParse and LiteParse WebAssembly engines for better multi-column and table layout, plus PP-OCRv6 OCR for scanned PDFs and images (PNG, JPG, WebP). Each loads locally on demand.
Preview and Export
Switch between raw Markdown and rendered preview, then copy to clipboard or download a .md file.
Text to Speech
Turn text into natural speech locally. Choose a language, then an on-device model, and generate audio without uploading anything.
Language-first models
Kokoro for English, Spanish, French, and Italian; Supertonic 3 for English, German, French, Spanish, and Italian. Models load from Hugging Face on first use and are cached.
Voices and speed
Pick a voice, set the speaking speed before generating, and adjust playback speed afterwards without re-rendering.
WAV and MP3 export
Play the audio in the browser, then download it as uncompressed WAV or a compact MP3 encoded locally.
Guides & Articles
Practical walkthroughs for common workflows - each one explains how the local processing works.
- Convert PDF to Markdown without uploading it - private document conversion, OCR for scans included.
- Clean up ChatGPT text before you publish - remove em dashes, smart quotes, and Markdown leftovers.
- Anonymize text before sharing it with AI - reversible placeholders for names, emails, and account data.
- Count LLM tokens offline - instant estimates and exact tokenizer counts for GPT, Llama, Qwen, and Gemma.
- Run AI chat locally in your browser - WebGPU models with no install and no account.
- Private text to speech in five languages - natural voices generated on your own device.
Privacy and Limits
The tools are designed for local browser processing. Optional AI modes download model files into your browser, but your text, prompts, attachments, chat history, context estimates, anonymization mappings, and generated speech remain local. PII detection and context estimation can miss edge cases, so manual review is still required.