Data security
On-premises deployment · AI automatic redaction · Knowledge-point-level permissions · End-to-end auditing
Security Principle: Your Data Never Leaves Your Premises
The raw material of a knowledge base is an enterprise's most critical assets: contracts, medical records, financial reports, official documents. From day one, UGLYPEAR AI was architected so that data never leaves the intranet: the entire base platform, including the AI detection models, is deployed in your own data center or private cloud, and document parsing, redaction, chunking, and storage are all completed locally, with no reliance on any external cloud service and full support for offline operation.
Fully On-Premises Deployment
The base platform is delivered as a Docker image or on bare metal. The OCR and layout analysis models (ONNX) ship with the package and run inference locally, while LLM summarization enhancement can connect to any OpenAI-compatible endpoint deployed on your intranet. From document intake to knowledge ingestion, the entire chain can operate with physical isolation: unplug the external network cable and the system keeps running.
Automatic Redaction of 13 Categories of Sensitive Information
Before ingestion, AI detectors automatically scan the images embedded in documents for sensitive content and blur it, producing a redaction report (recording type, location, and handling method) that is incorporated into the audit log. Two inverse optimizations (sky, bokeh) enhance background compression, balancing security against file size.
Knowledge-Point-Level Permission Control
Permissions are not implemented at the application layer and left to good intentions; they are implemented at the data layer. Each knowledge point (chunk) can carry tags for department, position, role, region, and confidentiality level. During retrieval, permission filtering happens before recall, so content outside a user's authorization never enters the candidate set. The multi-tenant architecture isolates different customers and business units by design, and supports preview verification of permission filtering results.
End-to-End Audit Logs
Who uploaded which document, when the system performed parsing, cleaning, redaction, or chunking, which knowledge points were produced, and which account retrieved what content and at what time: every link in the chain is traced and can be queried with one click from the admin console, meeting the requirements of internal audit and regulatory inspection.
Software Licensing and Tamper Protection
The SDK uses enterprise-grade machine-code licensing (hardware binding plus RSA signatures) with built-in anti-debugging protection to keep licenses from being cracked or misused. The server supports license revocation and quota management, and API keys are isolated by tenant.
Indigenous Technology Stack and Compliance
Full compatibility with China's indigenous technology stack: Kylin and UOS operating systems, Phytium and Kunpeng CPUs, Dameng and Kingbase databases, and TongWeb middleware. The architecture aligns with the principles of traceable processing and minimum necessity set out in the Data Security Law and the Personal Information Protection Law, supports deployment in MLPS environments, and offers an archival mode for government scenarios (full lossless plus compressed copy on a dual-track basis, with SHA-256 integrity verification).
One Line for Decision Makers
Security is not a list of features; it is an architectural choice. UGLYPEAR AI took a path that is bad for sales and good for customers: your data stays in your hands from beginning to end. What we sell is software, not your data.
Bring your security team and try to poke holes in it
Architecture, deployment, and audit plans, all open to review