Question 1

What is promptShield?

Accepted Answer

promptShield is a desktop application that anonymizes documents before you send them to AI tools like ChatGPT, Claude, or Gemini. It detects and replaces personal information (sensitive data) with reversible codes. entirely offline. No data ever leaves your machine. After the AI responds, promptShield's Decode Files feature automatically restores the original data, giving you a complete document with real names, dates, and details.

Question 2

Why do I need to anonymize documents before using AI?

Accepted Answer

When you upload a document to a cloud-based AI, the content leaves your machine and is transmitted to remote servers. This creates several risks: • Data retention. AI providers may store your input for training or debugging • Training data leakage. some providers use your inputs to improve their models • Third-party access. your data may be processed in countries with different privacy laws • Breach exposure. cloud platforms are high-value targets for cyberattacks • Regulatory violations. sending unredacted sensitive data may violate GDPR, CCPA, HIPAA, PIPEDA promptShield eliminates all of these risks by ensuring only meaningless codes reach the AI.

Question 3

What types of sensitive data does promptShield detect?

Accepted Answer

promptShield detects 40+ entity types including: • Names. full names, maiden names, aliases, signatures • Organizations. company names, department names, client or vendor names, project codes • Contact details. email addresses, phone numbers, physical addresses • Government IDs. Social Security numbers, passport numbers, driver's licenses, national IDs • Financial data. bank accounts, IBAN, credit card numbers, BIC/SWIFT codes • Health information. medical record numbers, insurance IDs, diagnoses • Dates. birth dates, appointment dates, employment dates • Digital identifiers. IP addresses, login credentials, device serials • Location data. GPS coordinates, home addresses • Biometrics. fingerprints, facial recognition data, voiceprints You can also define your own custom patterns for organization-specific data.

Question 4

What file formats are supported?

Accepted Answer

promptShield supports: • PDF (text-based and scanned) • Word (.docx) • Excel (.xlsx) • PowerPoint (.pptx) • Images: JPG, PNG, TIFF, BMP Scanned pages and images are automatically processed with OCR (Optical Character Recognition) to extract text for detection. Exported files are always clean PDFs regardless of the original format.

Question 5

Is promptShield really 100% offline?

Accepted Answer

Yes. All sensitive data detection and document processing happens entirely on your machine. No document content, no personal data, and no detection results are ever sent to any server. The only network calls promptShield makes are for license validation and software updates. neither involves your document data. You can verify this with any network monitor.

Question 6

What languages are supported?

Accepted Answer

promptShield supports sensitive data detection in 7 languages: English, French, German, Spanish, Italian, Dutch, and Portuguese. The detection engine adapts its patterns and entity models based on the selected language. You can also set it to auto-detect, which analyzes each page independently.

Question 7

What platforms does promptShield run on?

Accepted Answer

promptShield is available for Windows 10/11 (64-bit) and macOS (Apple Silicon and Intel). Both platforms require at least 8 GB of RAM. For the AI Deep Analysis feature, a GPU with 4+ GB VRAM or 16+ GB system RAM is recommended.

Question 8

How is promptShield different from just blacking out text?

Accepted Answer

Traditional redaction (blacking out) permanently destroys information and makes documents unreadable for AI. The AI cannot understand who does what. it can't distinguish between entities or relate names to actions. promptShield uses reversible encoding instead. Each entity gets a consistent, structured code (e.g., [PHKR92]). The AI understands that [PHKR92] is the buyer and [OAKL45] is the vendor. without knowing real identities. After the AI responds, promptShield automatically restores all original data. This means you get full AI analysis with zero sensitive data exposure.

Question 9

Is there a mobile app?

Accepted Answer

No. promptShield is a desktop application for Windows and macOS. Documents containing sensitive data should be processed on secure, controlled devices with proper access controls, not on mobile devices that may be shared or lost.

Question 10

Does promptShield work with any AI tool?

Accepted Answer

Yes. promptShield works with any AI tool that accepts text or files: ChatGPT, Claude, Gemini, Copilot, Perplexity, and any other LLM. Export your encoded documents, upload them to the AI, get the response, then use Decode Files to restore the original data. For API integrations, promptShield's CLI and Docker API work with any OpenAI-compatible endpoint.

Question 11

How do I load documents into promptShield?

Accepted Answer

You can load documents by: • Dragging and dropping files onto the main area • Clicking the Load button in the toolbar • Selecting multiple files at once for batch processing Multi-page documents are fully supported. You can load multiple files and switch between them from the Library.

Question 12

How does the detection process work?

Accepted Answer

Click Detect in the toolbar to scan your document. promptShield uses a 4-layer pipeline: 1. Patterns. built-in and custom rules for emails, phones, IDs, dates, credit cards, and more 2. Expressions. flags every occurrence of specific terms you define 3. Entities. a local AI model that recognizes names, organizations, and locations 4. Deep Analysis. a language model reads the full text to catch context-dependent sensitive data You can run detection on the current page or the entire document. Each layer can be individually enabled or disabled in Settings.

Question 13

What do the different encoding codes mean?

Accepted Answer

Every code starts with a letter indicating the sensitive data type: • P = Person • O = Organization • E = Email • T = Phone • S = SSN • C = Credit Card • D = Date • A = Address • L = Location • I = IP Address • B = IBAN • X = Passport • R = Driver License • K = Custom The code length matches the original text length to preserve document layout. Codes are consistent. the same entity always gets the same code.

Question 14

What is document linking and why does it matter?

Accepted Answer

Files placed in the same folder automatically share the same code registry. This means the same entity gets the same code across all linked documents. This is critical for AI ingestion: if Jean Dupont is [PHKR92] in the contract, he's [PHKR92] in the correspondence, annexes, and reports too. The AI can perform cross-document analysis safely. To link files, create a folder in the Library and move files into it, or drag files into an existing folder.

Question 15

How do I review detected regions?

Accepted Answer

After detection, each piece of sensitive data appears as a highlighted region on the document. You have two tools: • Document Toolbar. floats on the canvas for quick visual work. Select regions to Encode, Redact, Clear, change sensitive data type, or create regions from selection. • Region Sidebar. a right-side panel listing all regions across the entire document. Use it for sorting, filtering, and bulk editing. Keyboard shortcuts: V for Pointer, S for Lasso, D for Draw, T to toggle Encode, Delete to Clear.

Question 16

What is the difference between Encode and Redact?

Accepted Answer

• Encode replaces sensitive text with a reversible code (e.g., [PHKR92]). The original text is stored in your local code registry and can be restored later using Decode Files. • Redact permanently removes the text from the exported document. Redacted text cannot be recovered. Use Encode when you want the AI to work with the data. Use Redact when the data should never be recoverable.

Question 17

How does Decode Files work?

Accepted Answer

Decode Files is the second half of the promptShield workflow: 1. Upload the AI's response files (TXT, CSV, PDF, DOCX, XLSX, PPTX) or paste text directly 2. Click Swap Codes. promptShield scans the content and replaces each code with its original value 3. Review the summary table showing every code found and its replacement The result is a complete document with all real data restored.

Question 18

Can I draw regions manually?

Accepted Answer

Yes. If auto-detection missed something, press D or click the Draw tool, then click and drag on the document to create a box. Choose a sensitive data type from the picker that appears. Manually drawn regions work exactly like auto-detected ones. you can encode, redact, or clear them.

Question 19

How does OCR work for scanned documents?

Accepted Answer

promptShield automatically runs OCR (Optical Character Recognition) on scanned PDF pages and image files (JPG, PNG, TIFF, BMP). The extracted text is used for sensitive data detection, and regions are drawn over the original scan so you can see exactly what was found. OCR quality depends on scan resolution. 150 DPI or higher is recommended for best results.

Question 20

What metadata does promptShield strip?

Accepted Answer

promptShield deep-strips all metadata from exported PDFs: • Standard document properties. Author, Title, Subject, Keywords, Creator, Producer • XMP metadata streams • Table of contents and bookmarks • Annotations and form fields • Embedded file attachments • Page thumbnails • AcroForm field values • Document-level JavaScript For input files, it also strips: Word author info and tracked changes, Excel creator info and formulas, and image EXIF/IPTC/XMP/ICC metadata.

Question 21

Can I control the exported file size?

Accepted Answer

Yes. In the export panel, you can set a maximum file size (in MB) and a maximum number of files. When a document exceeds the size limit, promptShield automatically splits it into parts that fit. This is especially useful for AI tools that have file size limits.

Question 22

What are AI Instructions in exported files?

Accepted Answer

When exporting, you can insert an instruction page that tells the AI to preserve codes in its response. For example: "This document contains codes like [PHKR92]. Preserve them exactly as-is in your response." This ensures the AI doesn't paraphrase or modify the codes, making Decode Files work reliably.

Question 23

How do I verify codes before exporting?

Accepted Answer

Use the Verify Codes button in the export panel. It shows every code across all selected documents in one table. Use it to: • Catch misclassified regions before they reach the AI • Confirm all sensitive data has been encoded or redacted • Spot regions accidentally cleared instead of encoded • Review codes across linked files for consistency

Question 24

What keyboard shortcuts are available?

Accepted Answer

Navigation: ←/→ for pages, +/− for zoom, 0 to reset zoom, Tab to cycle regions Tools: V for Pointer, S for Lasso, D for Draw, Space to toggle tool Selection: Ctrl+A to select all, Ctrl+Click to multi-select, Escape to deselect Actions: T to toggle encode, D to toggle redact, Delete to clear Clipboard: Ctrl+C to copy, Ctrl+V to paste regions History: Ctrl+Z to undo, Ctrl+Y to redo Other: Ctrl+F for the auto-detect panel

Question 25

What detection methods does promptShield use?

Accepted Answer

promptShield chains four complementary layers: 1. Patterns. built-in and custom rules for structured data like emails, phone numbers, national IDs, credit cards, IBAN. Fast and fully offline. Language-aware. 2. Expressions. enter specific terms directly or paste from Excel/CSV. Flags every exact occurrence in the document. 3. Entities. a local AI model that identifies person names, organizations, and locations even when they don't follow a predictable format. 4. Deep Analysis. a language model reads the full text to understand context. Catches sensitive data that all other methods miss — indirect references, implicit identifiers, and uncommon formats.

Question 26

Can I create custom detection patterns?

Accepted Answer

Yes. Custom patterns let you define your own detection rules for data that built-in patterns don't cover — internal reference numbers, employee IDs, project codes. You can create patterns in Simple mode (visual block builder) or Advanced mode. Each pattern has a name, sensitive data type, case sensitivity toggle, and confidence score. Test patterns against sample text to see matches before saving.

Question 27

How does Deep Analysis (AI Engine) work?

Accepted Answer

The AI Engine powers Deep Analysis using a local LLM on your machine or a remote OpenAI-compatible API. For local processing, you need a GPU with 4+ GB VRAM or 16+ GB system RAM. Download compatible GGUF model files and place them in the models directory. For remote API connections, configure the API Base URL, API Key, and Model Name in Settings. Compatible with OpenAI, Azure, and local servers.

Question 28

Can I adjust how regions are grouped?

Accepted Answer

Yes. The Region Grouping slider in Detection Settings (0–100%) controls how aggressively nearby words merge into a single sensitive data highlight. Lower values keep regions separate; higher values merge adjacent words into one region.

Question 29

How do I handle text the detection missed?

Accepted Answer

You have several options: • Draw a manual region (D key) over the missed text • Add the missed term to your Expression list so it's caught everywhere • Create a custom pattern for recurring formats • Use the 'Create Regions from Selection' feature to find every matching occurrence in the document Detection improves when you combine layers. Patterns + Entities + Deep Analysis catches the most.

Question 30

Does promptShield work on multi-language documents?

Accepted Answer

Yes. You can set detection to auto-detect mode, which analyzes each page independently and applies relevant language models. promptShield supports English, French, German, Spanish, Italian, Dutch, and Portuguese. For documents mixing multiple languages, auto-detect is recommended.

Question 31

How accurate is the detection?

Accepted Answer

Accuracy depends on the layers enabled. Using all four together provides the highest recall: • Patterns handle structured data like emails, IDs, and numbers with near-perfect precision • Entities catch names, organizations, and locations with high accuracy • Deep Analysis catches subtle, context-dependent sensitive data that other methods miss We always recommend reviewing detected regions before exporting. promptShield is a tool that assists — the final decision is yours.

Question 32

Can I choose which entity detection model to use?

Accepted Answer

Yes. When Entities is enabled, you can choose from multiple models: Auto, spaCy, Piiranha, WikiNEuRal, XLM-RoBERTa, and various language-specific models. The Auto option selects the best model for the detected language.

Question 33

What format are exported files?

Accepted Answer

Exported files are always clean PDFs, regardless of the original input format (Word, Excel, PowerPoint, or images). This ensures consistent, platform-independent output with all metadata stripped.

Question 34

Can I export multiple documents at once?

Accepted Answer

Yes. promptShield supports batch export of up to 50 documents. Files can be exported as individual PDFs or as a single ZIP archive. When documents are linked (in the same folder), codes are consistent across all exported files.

Question 35

What happens to metadata when I export?

Accepted Answer

All metadata is deep-stripped from exported PDFs. This includes document properties, XMP metadata, bookmarks, annotations, form fields, JavaScript, embedded attachments, and page thumbnails. No hidden data survives the export.

Question 36

How does file splitting work?

Accepted Answer

In the export panel, set a maximum file size (in MB). When a document exceeds this limit, promptShield automatically splits it into multiple parts that fit within the limit. This is useful for AI tools with file upload restrictions.

Question 37

What file types can I upload for Decode?

Accepted Answer

Decode Files accepts TXT, CSV, PDF, DOCX, XLSX, and PPTX files. You can also paste text directly. promptShield scans the content for codes and replaces each one with the original value from your code registry.

Question 38

What if the AI modified or paraphrased a code?

Accepted Answer

If the AI changes a code, Decode Files will flag it as unresolved. To prevent this, use the AI Instructions feature during export. it inserts a page telling the AI to preserve codes exactly as-is. If codes are still modified, you can manually review the unresolved items in the summary table.

Question 39

Can I back up and restore my code registry?

Accepted Answer

Yes. In Settings > Codes, you can export your entire code registry as a backup file and restore it later. This is useful for migrating between machines or maintaining a shared team registry.

Question 40

What is the Trust Badge?

Accepted Answer

The Trust Badge is a verifiable certification that proves your organization uses promptShield to protect sensitive data before sharing documents with AI tools. It demonstrates your commitment to privacy and data protection.

Question 41

How do I get a Trust Badge?

Accepted Answer

Go to Dashboard > Trust Badge. Your badge is automatically generated based on your active promptShield subscription. You'll receive a unique badge URL, QR code, and embeddable HTML snippet.

Question 42

Where can I display the Trust Badge?

Accepted Answer

You can display the Trust Badge on your website, in email signatures, on proposals, in compliance documentation, or anywhere you want to demonstrate your data protection practices. Use the provided HTML embed code, or link to your unique verification URL.

Question 43

Can anyone verify my Trust Badge?

Accepted Answer

Yes. Each Trust Badge has a unique verification URL that anyone can visit to confirm your organization's active promptShield usage. The verification page shows your organization name and badge status.

Question 44

Is the Trust Badge included in all plans?

Accepted Answer

The Trust Badge is available on all paid plans (Week Pass, Pro, Business, and Enterprise). It remains active as long as your subscription is current.

Question 45

How does promptShield help with GDPR compliance?

Accepted Answer

Because PII is replaced with meaningless codes before leaving your machine: • No Data Processing Agreement (DPA) is required with AI providers • No cross-border transfer justification needed. no SCCs or adequacy decisions • Data minimization is satisfied. only encoded placeholders are exposed • Right to erasure is simple. deleting the local code registry removes all mappings • The architecture is privacy-by-design at its core

Question 46

Does promptShield address HIPAA requirements?

Accepted Answer

Yes. Since promptShield strips Protected Health Information (PHI) before it reaches the AI: • No Business Associate Agreement (BAA) is needed with AI providers • promptShield's detection covers all 18 HIPAA Safe Harbor categories • The Minimum Necessary Rule is satisfied. only de-identified codes reach the AI • Full audit trail records all detection, encoding, and export actions

Question 47

What about CCPA/CPRA (California)?

Accepted Answer

promptShield addresses CCPA requirements because: • No sale or sharing of personal information occurs. encoded exports contain no PI • The AI provider doesn't qualify as a CCPA 'service provider' since it receives no personal information • Consumer rights (delete, know, opt out) are satisfied locally

Question 48

Is promptShield compliant with PIPEDA (Canada)?

Accepted Answer

Yes. promptShield aligns with PIPEDA principles: • Principle 3 (Consent). no personal information is disclosed to AI providers • Principle 5 (Limiting Use). personal information is used only for the stated purpose • Principle 7 (Safeguards). local-only processing ensures the strongest possible safeguard

Question 49

Can promptShield be used in government and defense contexts?

Accepted Answer

Yes. promptShield's fully offline architecture makes it suitable for classified and restricted environments. Air-gapped activation is supported, and no document data ever leaves the machine. Contact us for Enterprise deployment options tailored to government requirements.

Question 50

Does promptShield provide an audit trail?

Accepted Answer

Yes, on Business and Enterprise plans. The audit log records all user activity including document loads, detections, encoding actions, exports, and account changes. The audit log is accessible from Dashboard > Audit Log.

Question 51

Do I need a DPA with promptShield?

Accepted Answer

promptShield processes documents entirely on your local machine. No personal data is transmitted to promptShield's servers. The only data exchanged is license validation and software update checks, neither of which involves document content or PII. If you require a DPA for regulatory reasons, contact our Enterprise team.

Question 52

Where is my data stored?

Accepted Answer

All data stays on your local machine. Documents, detection results, and the code registry are stored in your user data directory. Nothing is uploaded to any server. On Windows: %LOCALAPPDATA%\promptShield On macOS: ~/Library/Application Support/promptShield

Question 53

What network connections does promptShield make?

Accepted Answer

promptShield makes only two types of network calls: • License validation: confirms your subscription status • Update checks: looks for new software versions Neither call transmits document content, detection results, sensitive data, or any file data. You can verify this with any network monitor.

Question 54

Does promptShield collect telemetry or usage analytics?

Accepted Answer

Yes — anonymous usage statistics, and never document content. What that means depends on the version. The free web app always sends anonymous counts: which features get used, how long the detection engine takes to load, and a random per-browser identifier. There is no account and no opt-out — that is the trade for a free tool. The desktop app sends the same kind of counts and you can switch them off at any time in Settings → Privacy & Legal, on any plan; crash reports are separate and can also be turned off. On a Pro licence the desktop runs fully air-gapped through offline activation: no licence checks, no crash reports, no usage statistics — no outbound request at all. In every version, your documents, their file names, and anything detected inside them are never transmitted.

Question 55

How do I migrate my data to a new machine?

Accepted Answer

To migrate your code registry and settings to a new machine: 1. Export your code registry from Settings > Codes > Export 2. Transfer the backup file to the new machine (USB, network share, etc.) 3. Install promptShield on the new machine and activate your license 4. Import the code registry from Settings > Codes > Import Your codes, mappings, and custom patterns are fully restored.

Question 56

Is there a CLI or API available?

Accepted Answer

Yes. promptShield is available as a CLI tool (installable via PyPI) and as a self-hosted Docker API. Both are included with Business and Enterprise plans. Install the CLI: pip install promptshield-app Deploy the API: docker pull promptshield/promptshield-api:latest

Question 57

What can I do with the CLI?

Accepted Answer

The CLI lets you: • Detect PII in documents from the terminal • Anonymize and export documents programmatically • Restore original data (decode) • Start a local API server • Process documents in batch pipelines All operations work offline just like the desktop app.

Question 58

How does the self-hosted Docker API work?

Accepted Answer

Deploy the promptShield API on your own infrastructure using Docker Compose. All processing happens inside the container. zero-trust architecture means no data ever leaves your environment. The API provides REST endpoints for detection, anonymization, and decoding. See the API documentation at /developers/docs.

Question 59

What document formats does the API support?

Accepted Answer

The API supports the same formats as the desktop app: PDF, DOCX, XLSX, PPTX, and images (PNG, JPEG, TIFF). All formats are processed uniformly through the same detection pipeline.

Question 60

How do I authenticate API requests?

Accepted Answer

API requests are authenticated using your license key (starts with ps_live_). Include it in the Authorization header of your requests. Manage API keys at Dashboard > Developers > API Keys.

Question 61

Is there a Python SDK?

Accepted Answer

Yes. The promptShield Python SDK (promptshield-app on PyPI) lets you import the core engine directly in your Python scripts. It provides a promptShield client class with detect(), anonymize(), and detokenize() methods. Install: pip install promptshield-app With Office format support: pip install "promptshield-app[office]"

Question 62

How is the Settings screen organized?

Accepted Answer

Settings has five tabs: • Detection. Sensitive data detection methods, entity model selection, language, region grouping, custom patterns, AI engine, and review behavior • Codes. code registry statistics, backup/restore • Directories. default folders for uploads, downloads, and backups • Updates. check for new versions, download, install, or apply offline packages • Account. license management, appearance theme, UI language, help tips, and legal information

Question 63

Can I change the app theme?

Accepted Answer

Yes. Go to Settings > Account > Appearance. promptShield offers Dark, Light, and System themes. The System option follows your operating system preference.

Question 64

How do I update promptShield?

Accepted Answer

Go to Settings > Updates. Click Check for Updates to see if a new version is available. You can download and install it directly, or if your machine has no internet access, download the update package on another device and apply it using the offline installer.

Frequently Asked Questions

General

Desktop App

Detection & Privacy

Export & Decode

Licensing & Account

Trust Badge

Compliance & Regulations

Security & Data

Developers & API

Settings & Configuration

Billing & Payments

Troubleshooting

Still have questions?