How did you test the privacy of PDF tools?

We used Chrome DevTools (Network tab) to monitor XHR/Fetch requests while processing a standardized 10MB test file. This measures exactly how many bytes are transmitted to remote servers versus processed locally.

Are local browser tools truly 100% private?

Standard operations execute locally without network transit. However, AI-dependent tasks (like OCR or Summarization) require data transit to API endpoints. Absolute privacy depends on the specific tool module being used.

PDF Privacy Benchmark 2026: Cloud vs. Local Processing

When a web service promises that your uploaded files are "100% secure" or "deleted after 2 hours," they are asking for trust. In information security, trust is good, but empirical verification is better.

This 2026 benchmark shifts the privacy conversation from marketing copy to measurable network behavior. By utilizing standard browser diagnostic tools, we examine the actual data transit payloads of traditional cloud PDF converters versus local WebAssembly (WASM) architectures like PdfMinty.

1. Benchmark Methodology

To ensure reproducible results, testing was conducted under identical, verifiable conditions:

Test Environment: Google Chrome (v120+), Incognito Mode, all extensions disabled.
Test File: A standardized 10.0 MB PDF containing randomized mock PII (Personally Identifiable Information).
Measurement Method: Chrome DevTools > Network Tab. We filtered for XHR/Fetch and WS (WebSocket) traffic to record the exact byte count of HTTP POST payloads during file submission.
Tested Operation: PDF Page Extraction / Splitting.

2. Network Analysis: Traditional Cloud Converters

Traditional online PDF platforms (such as Smallpdf or iLovePDF) rely on remote server infrastructure to process files.

Observation: Upon initiating the "Split" action, the Network tab reveals an immediate POST request. The payload size matches or exceeds the 10.0 MB file size (often larger due to multipart/form-data encoding overhead).

Technical Implication: The raw binary stream physically leaves the user's network boundary. Regardless of TLS encryption during transit (data-in-motion), the unencrypted file must be written to server memory or disk (data-at-rest) for processing. Security relies entirely on the provider's automated deletion scripts (e.g., the standard "deleted after 1-2 hours" policy).

3. Network Analysis: Local Processing Architecture (PdfMinty)

Modern local-first tools utilize client-side JavaScript and WebAssembly to parse the PDF binary directly within the browser's sandbox.

Observation: Upon initiating the same "Split" action in PdfMinty, the Network tab registers 0 MB of binary transit. No POST request containing the file payload is generated. The processing completes instantly.

Technical Implication: The data never crosses the network boundary. The file is loaded into the browser's local ArrayBuffer, manipulated via WebAssembly, and served back to the user via a temporary Blob URL. The concept of a "data retention policy" becomes obsolete because the server never receives the data to begin with.

Reproduce This Test Yourself

You do not need to take our word for it. You can verify PdfMinty's local processing capability by forcing an offline state:

Load the Split PDF tool in your browser.
Turn off your computer's Wi-Fi or disconnect your Ethernet cable.
Select your file and process it.
The tool will successfully split and download the PDF entirely offline.

4. Architectural Limitations & AI Exceptions

Authority in security requires transparency about limitations. While structural PDF operations (Merge, Split, Rotate, Grayscale, Sanitize) execute entirely offline, computationally intensive tasks require specialized external infrastructure.

OCR and LLM Analysis: PdfMinty's OCR PDF and AI Analyze tools require optical character recognition and semantic analysis. Because running a 10-billion parameter Large Language Model locally in the browser is currently unfeasible, these specific modules transmit extracted text (up to the first 12 pages) to Google Gemini API endpoints.

These modules are strictly opt-in and distinct from the core offline utilities. Users must provide explicit interaction before any AI-related data transit occurs.

Conclusion

"Privacy" is not a marketing label; it is a verifiable architectural state. By shifting the processing locus from the cloud server to the client's local CPU, organizations can mitigate third-party data exposure, subpoena risks, and breach vulnerabilities associated with remote file processing.