Privacy Policy

9. Automatic Error Reporting

To improve service stability and quickly resolve technical issues, MANULINK automatically collects error reports when problems occur. By agreeing to our Terms of Service and Privacy Policy during account creation, you consent to this automatic error reporting.

Data Collected

• Error type and stack trace (code location)
• Page URL and browser information
• Console log entries (technical diagnostics only)
• Anonymized session identifier

Data NOT Collected

• Personal messages or user content
• Form data, passwords, or credentials
• Screenshots (unless you manually provide them)
• Financial or payment information

How We Use This Data

Error reports are used exclusively for debugging, diagnosing technical issues, and improving service reliability. Reports are retained for 90 days and then automatically deleted.

Your Control

You may opt-out of automatic error reporting at any time by visiting your User Settings and navigating to "Privacy & Data". Opting out will not affect your ability to use MANULINK.

9A. AI Model Training Data

MANULINK uses artificial intelligence to power features including drawing classification, component detection, takeoff element recognition, document processing, email management, and ERP workflow automation. To continuously improve these features, we passively collect anonymized, PII-scrubbed signals from your interactions with the platform to train and refine our AI models.

Passive Workflow Signals Collected

When you perform normal ERP actions, MANULINK records anonymized structural signals about your choices. These signals are used to improve AI predictions and recommendations across the platform. Specifically, we collect:

• Drawing classification decisions — document type assignments, corrections, and takeoff detection approvals or rejections
• Component detection feedback — approved, corrected, or rejected element detections with geometry data
• Quote creation patterns — document type chosen, whether linked to a job or customer (no pricing data)
• Work order completion signals — department transitions, task completion ratios, actual vs. estimated time (no worker identity)
• Email record linking choices — which record type a user linked an email to, and the action tag applied
• Document classification confirmations — when a user confirms or corrects an AI-suggested document type in the Paper Shredder
• Purchase order structure — line item categories, linkage to jobs/work orders, payment term selections (no prices or vendor names)
• BOM re-explosion outcomes — component counts and price source resolution results

Data NOT Used for Training

• Personal names, email addresses, phone numbers, or user credentials
• Company names, customer names, or vendor names
• Dollar amounts, prices, costs, or any financial figures
• Project addresses, locations, or identifying descriptions
• The content of emails, notes, or uploaded documents
• Any field that can identify a specific person, company, or project

How Training Data Is Processed

Before any signal is stored, it passes through our PII scrubbing pipeline which replaces all identifying field values with neutral placeholders (e.g., [SCRUBBED], [DOLLAR_AMOUNT_SCRUBBED]). Storage references are cryptographically hashed so original file paths cannot be reconstructed. Data is aggregated across tenants so no individual contribution can be traced back to its source. Trained models are deployed exclusively within the MANULINK platform.

Image Data for Vision Model Training

To train vision AI models, anonymized cropped regions of architectural drawings may be included in training datasets. These regions contain only technical drawing content (lines, dimensions, symbols) and are stripped of all identifying information including title blocks, company names, and project names. Full uploaded files are never used directly.

Legal Basis (GDPR)

For users in the European Economic Area and United Kingdom, our legal basis for processing anonymized workflow signals for AI training is Legitimate Interest under GDPR Article 6(1)(f). We have assessed that this interest is not overridden by your rights because: (a) all data is anonymized before storage, (b) the processing directly improves the service you receive, and (c) you have a meaningful opt-out mechanism described below. A full Legitimate Interest Assessment is available on request at privacy@manulink.ai.

Your Control — Opt Out of Cross-Tenant Sharing

You can opt your organization out of contributing anonymized workflow patterns to the shared AI knowledge base at any time. Navigate to Company Settings → AI Usage & Billing and toggle off "Contributing anonymized patterns to global AI knowledge." This setting takes effect immediately for new signals. Note that local agent learning (used only within your organization's tenant) cannot be disabled, as it is required for the service to function correctly.

Your Consent

By accepting the Terms of Service and Privacy Policy during account creation, you consent to the use of anonymized AI interaction and workflow signal data for model training as described above. This practice is consistent with standard industry approaches used by AI-powered SaaS platforms including OpenAI, Anthropic, and similar services.