freeOCR.me Logo freeOCR.me
SECURITY & PRIVACY

Zero-Disk Retention Architecture & Linux RAM-Disk Ephemeral Security

Document confidentiality and privacy are the non-negotiable architectural pillars of freeOCR.me. Unlike conventional cloud document utilities that persist uploaded files to solid-state drives (SSDs) or cloud object storage buckets (e.g., AWS S3, Google Cloud Storage)—where files can remain in filesystem journals, metadata logs, and backup snapshots—freeOCR.me operates on an uncompromising Zero Persistent Storage architecture.

1. The Cloud Storage Vulnerability Vector

When you upload sensitive files (such as medical invoices, corporate tax filings, legal discovery bundles, or identity documents) to conventional conversion platforms, standard workflows write files to persistent storage. Even if a service claims to "delete files after 1 hour", deleting a file on an SSD only unlinks the file allocation table pointer. The underlying flash memory cells retain the raw data until garbage-collection wear-leveling cycles overwrite them weeks later.

Furthermore, cloud object buckets routinely store asynchronous replication copies across availability zones, creating multiple attack surfaces for credential leaks, malicious internal actors, and compliance violations under GDPR and HIPAA regulations.

2. Linux tmpfs Volatile RAM-Disk Ingestion

freeOCR.me eliminates storage vulnerabilities by processing documents entirely within volatile Linux tmpfs RAM disk mounts:

  • Pure DRAM Electrical Storage: Uploaded PDF streams, intermediate raster page bitmaps, neural bounding-box vectors, and generated output files exist strictly as volatile electrical charges across system DRAM chips. At no point does raw document data ever touch physical, non-volatile solid-state drives or spinning hard disk platters.
  • Kernel-Level POSIX Unlinking: The microsecond an OCR conversion finishes and client download links are delivered, an automated POSIX unlink() syscall executes. The kernel de-allocates the memory inode and wipes the page tables.
  • Autonomous Watchdog Janitor Daemon: A continuous asynchronous watchdog daemon sweeps memory directories every 60 seconds. Any inactive temporary buffer older than 60 minutes is forcefully zeroed (memset) and reclaimed.
  • Zero Swap Paging: Server memory nodes disable swap paging for temporary RAM directories, guaranteeing that DRAM pages are never written to swap partitions on disk.
🛡️
Zero AI Model Training Guarantee:

freeOCR.me NEVER inspects, aggregates, mines, or uses uploaded user documents to train machine learning models. Your intellectual property, proprietary business records, and private documents remain exclusively yours.

3. Zero Account Registration & Ephemeral Email Delivery

Security through minimization: freeOCR.me does not require account creation, username/password credentials, or credit card collection. For large multi-page documents where users opt to receive download links via email:

  • Your email address is utilized strictly as an in-memory variable for SMTP dispatch.
  • Email addresses are never stored in databases, relational tables, or caching tiers.
  • We cannot send unsolicited marketing or promotional messages because your contact details do not exist in our systems.

4. Compliance Alignments: GDPR, CCPA, and Ephemeral Document Retention

By enforcing complete volatile memory processing, freeOCR.me inherently aligns with international data protection frameworks:

  • GDPR Article 5(1)(e) (Storage Limitation): Personal data is erased immediately upon fulfillment of the processing request.
  • GDPR Article 17 (Right to Erasure): Instant automated deletion obviates manual data deletion requests.
  • CCPA & CPRA: No user data or documents are ever sold, shared, or retained for commercial tracking.

5. Architectural Flowchart: Volatile Memory Lifecycle

  1. Client uploads document over TLS 1.3 encrypted tunnel.
  2. Web server writes stream directly into /dev/shm/freeocr_sessions/ (POSIX tmpfs RAM mount).
  3. OCR workers deskew, recognize, and assemble searchable PDF in RAM memory.
  4. Client streams output file directly from volatile buffer.
  5. Watchdog daemon executes unlink() syscall; DRAM memory reclaimed by Linux kernel.

6. Cryptographic and Operating System Memory Isolation

To verify our zero-retention claim from an engineering perspective, document processing on freeOCR.me takes place inside Linux kernel namespaces utilizing tmpfs volatile memory mounts:

# Architectural mounting configuration for volatile processing
mount -t tmpfs -o size=2G,noexec,nosuid,nodev tmpfs /mnt/ephemeral_ocr

# Kernel level unlink ensures zero persistent inode allocation
# POSIX unlink() immediately detaches the file descriptor from the filesystem table

Because the filesystem resides entirely in volatile DRAM, memory cells lose their electronic charges when processes terminate, guaranteeing that unauthorized disk forensic recovery is physically impossible.

7. Enterprise Compliance Matrix (GDPR, HIPAA, SOC 2)

Compliance Standard Standard Cloud OCR Risk freeOCR.me Zero-Disk Implementation
GDPR (Article 17: Right to Erasure) Complex deletion queues across cloud backups Instant POSIX unlink(); 0 retention in DRAM
HIPAA (Protected Health Information) Requires signed Business Associate Agreement (BAA) Stateless transmission; no PHI persistence
CCPA / CPRA (California Privacy) Risk of third-party model training use No data sharing, no indexing, zero training
Transport Security Variable TLS cipher suites Strict TLS 1.3 with Perfect Forward Secrecy

8. Frequently Asked Questions (FAQ)

Q: How long are my uploaded files stored on freeOCR.me servers?

Files are stored for exactly zero seconds after your processing session ends. Documents exist only in ephemeral Linux RAM disk memory during the seconds required to perform optical character recognition and generate your download.

Q: Are my documents ever viewed by human annotators or employees?

Never. The entire processing pipeline is 100% automated. No human has access to document memory buffers, and our server logs record only operational performance metrics (latency, page counts) with zero document contents.

Q: Does freeOCR.me use client documents to train artificial intelligence models?

No. We strictly forbid using user data for model fine-tuning or machine learning training. Our deep learning vision models are pre-trained on open-source public corpora.

Try freeOCR.me 100% Free

Convert your scanned PDFs, receipts, and images to dual-layer searchable PDFs and Structured Markdown with ephemeral RAM security.

⚡ Convert Scanned Document Now