Glossary
Most misunderstandings in AI procurement come from neighbouring terms rather than missing definitions: fine-tuning and RAG, storage location and processing location, agent and agentic. So each entry says what the term is not — and what it means for a project that has to go into production.
Terms explained
RAG (Retrieval-Augmented Generation)
RAG is an architecture in which a language model looks something up in a defined body of documents before answering, and cites where it found the answer.
ReadLLM (Large Language Model)
A large language model is a system trained on very large amounts of text to predict the most likely continuation of a sequence.
ReadSelf-hosted LLM
A self-hosted LLM runs on hardware you control rather than through a vendor API.
ReadHallucination
A hallucination is a fabricated statement delivered in convincing form — a court decision that never existed, a contract clause that is not in the document.
ReadAI agent
An AI agent is a system that pursues a goal over several steps — planning, calling tools, checking intermediate results — rather than returning a single answer to a single prompt.
ReadAgentic AI
Agentic AI describes the property of a system that plans, acts and evaluates across multiple steps; an AI agent is a system that has that property.
ReadAI red teaming
AI red teaming is adversarial testing of an AI system: deliberately trying to make it produce harmful, leaked or prohibited output before someone else does.
ReadAI governance
AI governance is the standing arrangement around AI use: who is accountable, how use cases are classified, how evidence is produced, and how a new tool gets approved.
ReadShadow AI
Shadow AI is the use of AI tools inside an organisation without approval, oversight or a contract — typically a personal account on a public chat service, used with work documents.
ReadGrounding
Grounding means an answer is tied to specific supplied material rather than produced from model memory — and, in a usable system, that the tie is visible as a citation.
ReadDocument intelligence
Document intelligence is the step after text extraction: deciding what the extracted text means — which number is the total, which date is the due date, whether a required field is missing.
ReadEvaluation (evals)
Evaluation is the practice of measuring an AI system against a fixed set of tasks so that every subsequent change is a measured delta rather than an impression.
ReadData residency
Data residency is the question of where data is physically processed and stored — which, for AI systems, means where inference runs rather than where the vendor is incorporated or where the web interface is served.
ReadThree questions that carry a vendor conversation
RAG — does the system answer from your corpus or from model memory? That determines whether statements can be traced. Hallucination — is the system permitted to say it does not know? One that must always answer will invent. Inference location — where does the computation run? Not where the company is registered, but where your inputs are processed.
Three questions produce a different conversation than a list of desired features — and they show quickly which supplier has no answer.
The German glossary covers a partly different set of terms, chosen from German-language demand:twenty entries, including accounting, contracts and works-council topics.
Frequently asked
Which terms does a decision-maker actually need?+
Three carry a vendor conversation: RAG (whether answers come from your corpus or from model memory), hallucination (whether the system is permitted to say it does not know), and where inference runs (not where the company is registered, but where your inputs are processed). Everything else is implementation detail.
Why does every entry name what the term is confused with?+
Because most misunderstandings come from neighbouring terms rather than missing definitions — fine-tuning and RAG, storage location and processing location, agent and agentic. Separating those pairs changes the questions you ask suppliers.
Is this a translation of the German glossary?+
No. The terms were selected from English-language demand, which differs: self-hosting and red teaming carry weight here and almost none in German, while the German glossary covers accounting and works-council topics that barely register in English. Where a term appears in both, the entry was written separately for each audience.
Will more terms be added?+
Along the questions that come up in actual projects. Terms that appear only in vendor decks and never in a conversation do not get an entry.