Prepare the file before you share it

SafeShare creates a cleaner copy of a structured dataset before it goes to an AI assistant, client, freelancer or vendor. Run it locally, review what it finds, choose the changes, then explicitly apply them. Your original stays unchanged.

What you get

• CSV, TSV, XLSX and JSONL processing, including bounded nested JSONL objects.
• Deterministic email/IP/known-credential patterns, Luhn-checked card candidates, phone/secret heuristics, editable sensitive-header candidates and private custom values/regex.
• Redaction, session-stable pseudonyms, full-field replacement and column/field removal.
• Preview counts and affected fields without raw sensitive samples; a matching preview is required for apply.
• One sanitized copy, Markdown audit report and JSON manifest per successful input; isolated batch failures.

Example

A synthetic CSV contains the same example email in two rows plus a customer name and amount. Select the name field, choose pseudonymization, preview, then apply. Both email occurrences become the same token, such as [EMAIL_0001]; the name becomes a consistent [FIELD_0001], while amounts remain available for analysis. Drop an address column entirely when it is unnecessary. No reversible mapping is delivered.

Good fits

Prepare CRM extracts for an analyst, marketing reports for an agency, research tables for an AI assistant, or operational logs for a contractor. English edition; a separate Traditional Chinese edition uses the same engine—choose one.

Requirements and limits

Python 3.10+ with openpyxl, regex and defusedxml, declared in requirements.txt. Dependencies install locally; the processing engine makes no external requests and needs no API key. A local Agent Client needs file/command access; the CLI works without a model. AI client/model charges are separate.

Up to 20 files, 10 MiB each, 20,000 rows per sheet/file, 200 columns and 100,000 scalar cells per file; 20 XLSX sheets. See documentation for all bounds. XLSX is rebuilt as values only with generic sheet names; formulas, merged cells, encrypted/protected workbooks and external links are refused. Source comments, links, styling and metadata are discarded. CSV/TSV output is UTF-8 and formula-like text receives an apostrophe.

Honest privacy boundaries

Patterns and checksums identify candidates, not certainty. Names, addresses, free text and indirect identifiers can be missed. Review fields manually or select them explicitly. Pseudonyms preserve linkability and reset every run; this is not a complete anonymizer, data-breach guarantee or GDPR/HIPAA/CCPA certification.

Run the tool before sharing original files with AI. The engine has no telemetry, cloud API or model call; your Agent Client may transmit chat/tool content according to its settings. Reports contain filenames and field labels and also need review. Never assume a zero-match report means a file is safe to disclose.

Download and support

One-time US$14.99 Download, version 1.0.0. Readable source, localized Skill, documentation, synthetic demo and license included. No subscription or ongoing infrastructure cost; future updates are not guaranteed. Support: [email protected]. Usage rights and refunds follow the included license, applicable Capafy terms and law.