Private AI glossary
Clear, citable definitions of terms around private AI, LLM architecture and European compliance (GDPR, NEN 7510, DORA, EU AI Act). Intended for IT, compliance and procurement professionals looking for the right word fast.
A
- AI Act (EU)(Regulation 2024/1689, EU AI Act)
- The first integral legal framework for AI in the EU. Requires risk classification, transparency and human oversight. High-risk systems (AI in HR, healthcare or critical infrastructure) must demonstrate audit trails, technical documentation and data-quality requirements. Amaii supports compliance through audit logging, model choice and data traceability.
C
- Company brain (AI Company Brain)
- A central AI layer that unlocks all company knowledge — documents, contracts, policies, files, emails — through natural language. Unlike public chatbots, a company brain operates only on internal data and preserves role-based access. Amaii positions itself as the private company brain under the tagline Build your AI company brain™.
D
- DORA(Digital Operational Resilience Act, Regulation 2022/2554)
- EU regulation (in force since 17 January 2025) mandating digital resilience in the financial sector. Banks, insurers and pension funds must manage ICT risks, report incidents and register third-party risks (including AI suppliers). Private AI within your own environment makes DORA compliance demonstrably easier than SaaS LLMs with external processing.
E
- Embedding
- Numerical vector representation of text (or image) that allows semantic similarity to be calculated. Embeddings sit at the heart of RAG systems: documents are stored in advance in a vector database so the LLM can retrieve the most relevant passages when a question comes in.
F
- Fine-tuning
- Further training of an existing LLM on your own documents and examples, so the model recognises the terminology, writing style and context of your organisation. Especially interesting for open source models, because the weights then remain the property of the customer.
G
- GDPR(Regulation 2016/679, General Data Protection Regulation)
- European privacy law (Regulation 2016/679) that governs the processing of personal data. For AI it means: lawful basis, data minimisation, purpose limitation and the rights of data subjects to access and erasure. Amaii is GDPR-compliant by design because all data processing takes place within the customer's own infrastructure and no prompts or documents are shared with external AI providers.
H
- Hallucination
- An LLM answer that sounds plausible but is factually incorrect. Hallucinations are the main compliance risk in legal, medical and financial contexts. RAG architecture with source references (as applied by Amaii) drastically reduces the risk because answers are traced back to verifiable source documents.
L
- LLM (Large Language Model)
- A large language model with billions of parameters that can understand and generate text. Well-known examples include GPT-4, Mistral and Claude. With Amaii, the customer decides which LLM runs in their own environment — open-source (Mistral) or EU-hosted commercial models.
M
- Model-agnostic
- An architecture that is not tied to one specific AI model. The customer can replace the underlying LLM without rebuilding integrations. Prevents vendor lock-in. Amaii is designed to be model-agnostic by default.
N
- NEN 7510
- Dutch standard for information security in healthcare, derived from ISO 27001 with healthcare-specific extensions. Mandatory for hospitals, mental-health institutions, GP practices and healthcare insurers. AI applications working with patient data must support NEN 7510. Amaii runs on-premise and keeps EHR data within the NEN 7510 boundary.
O
- On-premise deployment(on-prem, self-hosted)
- Installation and execution of software on the customer's own servers, behind their own firewall. Opposite of SaaS. For AI, on-premise means that no prompt or document leaves the organisation. Amaii supports both on-premise and private cloud (within the EU) deployment.
- Open source LLM
- A language model whose weights are publicly available, such as Mistral or DeepSeek. You can run the model in your own environment instead of consuming it as a service. That keeps you in control, prevents vendor lock-in and lets you fine-tune on your own data.
P
- Private AI
- AI solution where data and model processing stay within the organisation's own infrastructure. No processing by external providers such as OpenAI, Anthropic or Google. Opposite of public AI services. Amaii is a private AI platform.
- Private cloud LLM
- Private LLM that does not run on-premise but in a shielded EU cloud environment within the customer's tenant. Combines data sovereignty with cloud scalability, without prompts or documents being shared with a public AI provider.
- Private LLM
- A Large Language Model that runs exclusively within the customer's own environment — on-premise or in a private cloud. Prompts, answers and trained data never leave the organisation. Fundamentally different from API-based LLMs like ChatGPT or Copilot.
Read more: Secure private LLM for business: the complete guide
R
- RAG (Retrieval-Augmented Generation)
- Architecture where an LLM first retrieves relevant company documents for every question and passes them as context to the model. Benefits: up-to-date company data, source references and reduced hallucinations. Amaii's company brain is built around a RAG pipeline with a vector database.
- RBAC (Role-Based Access Control)
- Access control based on user roles. In an AI context RBAC means an employee only gets answers based on documents they have access to. Amaii applies RBAC at document level, integrated with Active Directory and Microsoft Entra ID.
S
- Sovereign AI(Data sovereignty)
- AI infrastructure that falls entirely under European jurisdiction and control. No dependence on US Big Tech or cloud providers that fall under the US CLOUD Act. Amaii was founded to offer European organisations a sovereign AI choice.
V
- Vector database
- Specialised database for storing and searching embeddings. Enables semantic search: finding on meaning rather than exact word match. Examples: pgvector, Qdrant, Weaviate. Amaii uses pgvector for consolidated deployment inside Postgres.
- Vendor lock-in
- Dependence on a single supplier that makes switching technically or contractually impossible. In AI this shows up as proprietary APIs, model-specific prompts and data-extraction obstacles. Amaii's open architecture and model-agnostic setup prevent vendor lock-in.
