Three software layers. One environment for working with AI
From scattered chatbots, API keys and unsupervised agents to one environment where you know who is working with which data and model — and at what cost
AI is spreading through organisations from the bottom up — and outside their control
Teams connect directly to different providers' APIs, while employees take company data into public chatbots. This is shadow AI: nobody knows who uses which tools with which data. Three risks grow together — security, costs (an agent stuck in a loop can generate a runaway bill) and dependence on a single provider
Where AI is already available, it is often just a chat window. Each conversation starts from scratch, one employee's knowledge never reaches the team, and the chat remains disconnected from calendars, documents and the systems where work actually happens. Employees return to their old tools
Regulatory pressure is growing too: the AI Act, GDPR, DORA, NIS2/Polish KSC, KRI and ISO/IEC 42001 require demonstrable oversight of AI, backed by evidence. With fragmented access, collecting that evidence is practically impossible
An organisation cannot protect, account for or demonstrate what it cannot seeControl, predictable costs and compliance require one gateway for all AI traffic — and one environment where work takes place
Each layer answers a different question
First “what can I improve?”, then “is it safe?”, and finally “is it compliant?” — this is the order in which you make decisions and the order in which SAIE works
TWIN:DESK — an everyday workspace for documents, knowledge and AI assistants
A browser application that turns a basic chat into a complete workspace for employees, teams and entire organisations — with institutional memory, data control and integration with administrative processes
- AI interaction — switching models, controlling context, proactive suggestions
- Institutional knowledge — documents and knowledge bases (RAG) with source citations
- Operational work — notes, tasks, automations, reports and calendars
- Oversight — RBAC, audits, data classification, integration and model control
PROXY:AI — the central layer of control, security and corporate governance for AI
One gateway between users, applications and AI agents and over 100 models — cloud and local. It eliminates shadow AI, protects sensitive data and keeps costs under control — from cloud to on-premises and air-gap deployments
- AI Gateway — an intelligent router through which all model traffic passes
- AI Firewall — a real-time firewall: content analysis, policies and data masking
- Governance — immutable WORM records and evidence packs for auditors
- Cost Control — FinOps for AI: budgets, limits and predictable costs
CDF — AI decisions with context, an owner and evidence
The Cognitive Deployment Framework organises the AI lifecycle, accountability, risks and measurable operational quality. Compliance is built in before solution development starts, not afterwards
- Assessment — ACE selects legal requirements and controls for the organisation's profile — different for a bank and a public authority
- Pilot — Cognitive SLA measures decision quality, hallucinations and knowledge freshness
- Gate — “scale or stop” — an end to never-ending pilots
- Scale — continuously maintained AIMS evidence: AI system register, DPIA
One employee question passes through all three layers
For organisations accountable for their data
SAIE is for organisations that need to know where their data goes — because of regulatory requirements, contractual obligations or a straightforward risk assessment. You decide which model works with which data: your own locally hosted model, or an external one within boundaries you define
Three people who need to say “yes”
Each typically says “no” for a different reason. SAIE answers all three at once
If you need a simple website chatbot, there are cheaper solutions and we will gladly point you towards them. SAIE makes sense when your organisation's policy determines model selection, rather than a provider's default, and when risk assessment precedes the first visible results
What SAIE makes possible
Fourteen areas where SAIE changes how your organisation works with AI. Each shows which layer is responsible
Company knowledge in projects, workspaces and knowledge bases, available to those who need it
Folders, email, calendars and Teams as context — without loading data into a central RAG database
Documents, prompts, context and responses on-premises, in colocation or in the cloud — according to the customer's decision
Built-in mechanisms supporting the AI Act, GDPR, DORA, NIS2 and cybersecurity requirements, rather than additions after deployment
All model communication through a central gateway: ABAC/RBAC rules, MCP tool control and auditing at user, role, team and tenant level
Create advanced assistants using natural language
The system produces compliance evidence, while the products and CDF methodology are covered by ISO/IEC 42001 certification
Ready-to-use software — deployment in no more than 10 working days in a prepared on-premises architecture
Our own RAG database, OCR, metadata filtering and source previews. When context is missing, the system identifies a knowledge gap instead of guessing
Persistent user memory and work profile, a learning log, memory export, email and calendar synchronisation — with explicit editing and deletion
Over 100 providers, switching models during a conversation without losing context, a model council, comparison view and automatic fallback models
The cost of every request by user, team, project and model; hierarchical budgets, alerts, internal currency and denial-of-wallet protection
Schedules, natural-language “if–then” rules, task delegation between agents and undoing agent actions with a visible plan
Microsoft 365, Google, Outlook, Teams, Drive, OneDrive, GitHub, Jira, n8n, Langflow, Codex, Claude Code, Kubernetes, Helm, GitOps/IaC, VSCode/Cursor
128 featuresin 13 searchable categories
All SAIE capabilities · 128How SAIE differs from open source and hyperscalers
Full cost control from user to model, with multiple planning levels
Total cost is hard to predict: infrastructure, integration, maintenance and, above all, expertise
Cost grows with usage: tokens, storage, transfer and currency risk; control at subscription level, not process level
You choose the data location: on-premises, in the EEA or globally
Control is possible, but you build the entire environment and its security from scratch
Data in the provider's regions; metadata, telemetry and CLOUD Act jurisdiction on its side; no on-premises option
Your processes deployed in two weeks
A quick PoC, a distant production launch — moving from a demo to a working process usually takes months
Fast access to models and generic assistants; an integrator still has to add the business process
Software and services at business / enterprise standard, with full support
Split between communities, without a single accountable party or guaranteed SLA
Paid, standardised platform SLAs, not deployment SLAs; English-language tickets and lengthy escalation
Processes deployed on-premises — your unique know-how stays with you
Code is available, but protecting extensions, configuration and licences is your responsibility
Prompts, configurations and workflows live with the provider; access to platform mechanisms is limited
CDF methodology guides you step by step towards compliance with applicable regulations
You must build risk assessment, documentation, oversight and procedures yourself, without a methodology
Tools support the process, but accountability and documentation remain with the customer
Service tailored to your needs and capabilities
Versions, security patches and availability require your own team or multiple contracts
Automatic updates outside your control; customised SLAs increase costs
Bespoke features — commission what you need
Every non-standard feature must be designed, developed and maintained
The provider owns the roadmap; bespoke features are practically unavailable
Source code and training delivered to the customer — continuity without vendor lock-in
Less dependence on a vendor, more on frameworks, maintainers and your own team
Deep lock-in: proprietary APIs, embeddings, identity and formats; exit costs grow with every process
Switch local and hosted models without restrictions
Changing a model requires infrastructure adjustments, prompt changes and revalidation
The provider determines models, versions, limits and retirement dates; no local models
A local agent works with personal sources while preserving confidentiality
Data can stay local — if you build isolation, key management, DLP and auditing yourself
Contractual assurances (no training, encryption), but data leaves the organisation
The digital twin learns working patterns that can be shared with the team
You must build the mechanism that learns working patterns yourself
Knowledge of a specialist's work remains their personal profile, not an organisational asset
An integrated AI and automation environment — connect technologies without restrictions
Managing orchestration, monitoring and versions yourself increases complexity and risk
Strong integration within the provider's ecosystem, weak or expensive integration outside it