SAIE products / Deployment and security

Deployment and security

For IT and security teams: where SAIE runs, how it is built, how it scales and what its certificates confirm

01 · DEPLOYMENT OPTIONS

You choose where your data is stored

Every deployment is a separate, fully isolated environment. For On-Premises and Air-Gap, selecting GPU infrastructure for local models is part of deployment. The platform integrates with DLP and SIEM solutions, supports custom data classifiers and industry-specific extensions

You choose the data location — from a managed service to an isolated environment

SaaS · managed service, fastest launch · use cases: evaluations, organisations · without their own infrastructure · DATA: PROVIDER'S DATA CENTRE (POLAND OR EU) · On-Premises · installation on your own infrastructure · use cases: regulated sectors, · public administration · DATA: THE ORGANISATION'S SERVER ROOM · Air-Gap · environment isolated from public networks, · local models only · use cases: defence, critical · infrastructure · DATA: ISOLATED INFRASTRUCTURE · FASTEST LAUNCH · HIGHEST DATA SOVEREIGNTY

02 · ARCHITECTURE AND PLATFORM

26 platform features

Architecture, security, performance and interoperability — shared by TWIN:DESK and PROXY:AI

Architecture and deployment11

Delivery as Docker container applications deployable in Kubernetes
Integration with a relational database such as PostgreSQL
Operation on-premises or in the customer's private cloud with full data control
Use of a high-performance vector database as the main RAG engine
Support and indexing for at least 1 million vectors with at least 1024 dimensions
Launch of a basic workspace through Docker in approximately 60 seconds
Per-tier GPU isolation for higher service levels
Declarative data-sovereignty control with an EU-only switch for the tenant
Offline operation with local models and tool integrations
Isolation of the AI environment from public networks for critical sectors
A multi-tenant model supporting B2C, B2B and B2B2C, local payments and white-label delivery

Security and privacy6

Restricting the pasting of secrets, keys, passwords and complete personal data without organizational consent
Role- and permission-based data access without automatic administrator visibility into private user content
Detecting and masking personal data before sending it to a model
A complete cryptographically signed event log with retention up to 2 years
An immutable log of every call compliant with EU AI Act requirements
A real-time visual map of request flows

Performance and scalability5

Support for long-running contexts and complex research tasks
Scaling to 1,000 users
Fallback mechanisms for Neo4j graph connection problems and other backend components
Native OpenTelemetry support for monitoring latency, errors and request rerouting
Predictable latency and availability under load through health checks and retries with backoff

Interoperability4

Integration with calendars, external tools and different AI providers
GitOps/IaC compatibility with config.yaml as the source of truth and a Git versioning workflow
Integration with mail, calendars, files and Google, Outlook, Drive, OneDrive, GitHub and Jira environments
Integration with n8n, Codex and Claude Code without rebuilding the entire environment

03 · CERTIFICATES

INTEGRATED MANAGEMENT SYSTEM

One management system. Seven ISO standards. One auditor

All certificates cover the same integrated system, rather than separate “projects”. ISO 42001 certification covers software design and development, methodology, consulting and training

Bureau Veritascertification body
all seven standards

0 nonconformitiesin the ISO/IEC 42001 certification audit (22 May 2026)

ISO/IEC42001

ARTIFICIAL INTELLIGENCE MANAGEMENT