OXS vs LoyalCloak

אפליקציית oxs לדיירים בבית משותף

Censor private data & shrink docs to fit any AI prompt. LoyalCloak is the zero-cloud, privacy-first document pre-processor built to censor sensitive data, slash token costs, and format documents before pasting into AI assistants like ChatGPT, Claude, DeepSeek, Gemini, and NotebookLM. WHY SHRINKING TOKEN SIZE MATTERS - Prevent AI Hallucinations: Cluttered formatting, repeating headers, and whitespace waste model attention. Stripping fluff focuses the AI on core context, improving accuracy. - Bypass Prompt Upload Limits: Never get blocked by "Prompt is too long" errors. Intelligent layout normalization compresses documents by up to 50%, letting massive files fit in one prompt. - Slash API Costs & Latency: Fewer input tokens directly reduce API bills and yield faster response times. 100% ON-DEVICE PRIVACY & ZERO TELEMETRY - Zero network calls: Documents and parsed data never leave your device. - RAM-only execution: Sanitization and ML inference run in transient memory and are dropped immediately. - Zero analytics, tracking SDKs, or cloud server storage. NOTEBOOKLM EXPORT & SMART SPLITTER - Bypass Source Limits: NotebookLM caps sources at 500,000 words, and large PDFs often fail to upload. LoyalCloak cleans and prepares documents so they upload reliably. - Calibrated Word Budgets: Automatically split oversized documents into structured Markdown parts with clean section breaks (500k, 100k, 50k, or 25k word presets). - Batch Merge: Combine multiple research notes into consolidated, model-ready NotebookLM sources with 1 tap. AI DEEP SCAN - ON-DEVICE PERSON-NAME RECOGNITION (PRO) - Next-Gen Local ML: On-device DistilBERT transformer (quantized int8) runs via ONNX Runtime directly on your device. - Person-Name Detection: Accurately detects and redacts names in unstructured text without sending data to the cloud. - Reversible Pseudonymization: Replace real names with realistic pseudonyms ("Jane Doe" -> "Alex Abbott"). Paste the AI's response back to restore original names with 1-tap. AUTOMATED PII CENSORSHIP & REDACTION - Standard Redaction: Redact emails, phone numbers, FDA NDC drug codes, and ICD-10 diagnosis codes. - Pro Deep PII Censorship: Redact credit cards (Luhn-verified), SSNs, IBANs, IPv4/IPv6, API keys, secrets, national IDs, and custom regex rules. - Pro Pattern Library: Checksum-validated NPI, DEA, MRNs, court dockets, and account numbers with per-detector toggles. - FRCP 5.2 Masking (Pro): Keep only the last 4 digits of SSNs, cards, and account numbers for legal filings. - Allowlist: Whitelist trusted terms or corporate domains so they are never touched. PRIVACY AUDIT REPORT (PRO) - Export timestamped, on-device audit logs of redacted items as PDF or Markdown (Summary, Masked, or Full detail). FIT ANY AI CONTEXT BUDGET - Real-time token metrics via on-device BPE (OpenAI/ChatGPT, Claude, Universal estimation). - Smart Auto-Chunking (Pro): Split long docs on section boundaries with [part_N_of_M] tags. - Custom XML Wrappers (Pro): Encapsulate output in structured [context] prompt tags. MULTI-FORMAT & OCR SUPPORT - Supports PDF, Word (DOCX), EPUB, HTML, CSV/XLSX, Text (.txt, .md), and images via on-device OCR. - iOS Share Extension: Send documents straight from Files, Safari, or Mail in 1 tap. - Save & Share: Copy markdown with 1 tap, save as .md, export for NotebookLM, or share. TERMS & PRIVACY - Privacy Policy: https://doc.loyallogic.org/privacy - Terms & EULA: https://doc.loyallogic.org/terms Censor your data with total confidence and make any document fit your AI prompts.