LoyalCloak converts messy, bloated documents (PDF, DOCX, HTML, EPUB, spreadsheets) into clean, token-optimized Markdown — with on-device PII scrubbing, header/footer stripping, and real-time token analytics. All powered by a native Rust engine that never lets your data leave the device.
Native macOS, Windows & Linux apps live · Web App available · iOS on App Store · Android on Google Play
Type sensitive data below. Watch emails, phones, cards, and extra whitespace disappear in real time — 100% locally in your browser.
From PII redaction to token optimization — all running locally on your device.
On-device DistilBERT transformer model (int8 via ONNX Runtime) detects and censors person names across unstructured text with zero cloud API calls (Pro).
Instantly redact emails, phone numbers, NDC drug codes, and ICD-10 diagnosis codes. Pro adds credit cards (Luhn-verified), SSNs, IBAN, IPv4/IPv6, API keys, and custom regex rules with live preview.
Checksum-validated detectors for health, legal, and financial IDs — NPI, DEA, MRNs, court dockets, account numbers. Toggle each pattern on or off and opt into FRCP 5.2 partial masking that keeps only the last four digits (Pro).
Export a timestamped summary of exactly what was found and removed — generated on-device and exportable as PDF or Markdown, with metadata-only, masked-preview, or full-values detail levels (Pro).
Replace names and private data with realistic pseudonyms. Paste the AI response back into LoyalCloak to restore original values on-device with 1 tap (Pro).
On-device tiktoken BPE counting for ChatGPT, Claude, and universal estimation. Collapse whitespace, repair line breaks, and strip soft hyphens to cut token size by up to 50%.
Process PDF, DOCX, EPUB, HTML, CSV/Excel spreadsheets, plain text, and images via hardware-accelerated on-device OCR. No uploads, no conversion servers.
Split long documents on section boundaries with sequential <part_N_of_M> tags and encapsulate in structured <context> prompt wrappers (Pro).
One-tap export that splits sanitized documents into clean Markdown sources under NotebookLM's 500k-words-per-source limit — or merges whole batch files together to stay under its 50-sources-per-notebook cap. Free within tier page limits.
LoyalCloak is now available as a standalone native application for macOS, Windows, and Linux. Powered by a native Rust engine and on-device ONNX runtime, process entire binders of PDFs, spreadsheets, and documents with zero cloud latency.
Your data never leaves your device at any point in the process.
Share a PDF, DOCX, spreadsheet, EPUB, HTML, or image from any app — or open a file directly.
PII is scrubbed, formatting is normalized, and tokens are counted — all in RAM on your device.
Copy clean, token-optimized Markdown and paste it into ChatGPT, Claude, DeepSeek, or NotebookLM.
LoyalCloak was built around a single non-negotiable principle: your document data stays on your device. All parsing, regex scrubbing, PII redaction, and token optimization execute 100% locally in transient RAM using a high-performance native Rust engine. Zero network calls, zero tracking, zero cloud telemetry.
Start free. Upgrade when you need more power.
Forever
or $29.99 lifetime
📱 The LoyalCloak mobile apps are now live on the App Store and Google Play.
Learn how LoyalCloak sanitizes confidential documents and fits them into AI prompts with 100% on-device privacy.
All document parsing, regex scrubbing, PII redaction, on-device ML inference, and token counting execute 100% locally in transient RAM on your device using a high-performance native Rust core. No text or metadata is ever sent to external cloud servers or telemetry endpoints.
AI Deep Scan uses an embedded, quantized int8 DistilBERT Named Entity Recognition (NER) model running 100% on-device via ONNX Runtime. It detects full and partial person names across unstructured documents locally without communicating with any cloud API.
LoyalCloak strips repeated headers, footers, boilerplate disclaimers, and unnecessary whitespace. It calculates exact BPE tokens (tiktoken cl100k_base / o200k_base and Claude models) and auto-chunks documents along heading boundaries with numbered prompt labels.
LoyalCloak parses PDF, Microsoft Word (DOCX), Excel / CSV spreadsheets, EPUB eBooks, HTML web pages, plain text (TXT, MD), and scanned documents or images via hardware-accelerated on-device OCR.
The Free tier redacts email addresses, phone numbers, FDA NDC drug codes, and ICD-10 diagnosis codes. The Pro tier adds AI Deep Scan person-name NER, Luhn-verified credit cards, Social Security Numbers (SSN), IBAN accounts, IPv4/IPv6 addresses, API keys & secrets, national IDs, custom regex rules, and the Pattern Library: checksum-validated NPI and DEA numbers, medical record numbers, court docket numbers, and financial account numbers — with optional FRCP 5.2 partial masking that keeps only the last four digits.
A timestamped, on-device summary of everything LoyalCloak found and removed in a document — the active profile, detectors that ran, per-type counts, and token counts. You choose the detail level (metadata only, masked previews, or full values) and export it as PDF or Markdown. It is generated entirely on your device and never uploaded.
Reversible redaction replaces real entities with realistic synthetic pseudonyms before querying AI models. When the AI returns an answer, you can paste it back to restore the original confidential values locally on your device.
Yes, LoyalCloak offers a forever-free tier with up to 5 files per day and 15 pages per document. Pro access ($14.99/yr or $29.99 lifetime) unlocks unlimited files, AI Deep Scan, deep PII, custom regex rules, batching, and auto-chunking.