New customer story: LeadsIt, the B2B prospecting app we built with AI. Read the customer story →

Vocabulary

AI vocabulary, without the jargon

AI words come up everywhere and are rarely explained. Here are the 40 terms you will meet in an AI project, each in one or two sentences, with an example when it helps.

The basics

The words you hear everywhere, explained in one sentence.

Artificial intelligence (AI)

A set of techniques that let software do tasks that used to need a person: understanding a text, recognising an image, making a simple decision.

Generative AI

AI that produces new content (text, images, code) from a request. Example: drafting a letter from a three-line instruction.

Language model (LLM)

The engine behind generative AI: a program trained on huge amounts of text that predicts the most likely next words. Claude, GPT, Mistral and Llama are language models.

Prompt

The message you write to tell the AI what to do. The more precise it is (context, expected format, examples), the better the answer.

Chat assistant (chatbot)

An interface where you ask questions in everyday language and the AI answers. A consumer chatbot does not know your company: it cannot see your documents.

AI agent

An assistant that does more than answer: it chains several steps and uses tools (searching your documents, filling in software, preparing an e-mail) to reach a goal.

Agentic AI

A way of building AI around agents that act, not just answer. At Open Intelligence, the agent proposes and a person approves before any important action.

Token

The unit of text a model reads and writes: a piece of a word, about three to four characters. Model providers often charge by the number of tokens.

Context window

How much text a model can take into account at once: the question, the documents provided and the answer. Beyond that, the useful passages must be chosen.

Hallucination

A statement the AI makes up and presents confidently, with no basis in its sources. Making the AI answer from your documents, with citations, greatly reduces this risk.

AI and your documents

How an AI finds the right information in what your company has already written.

RAG (retrieval-augmented generation)

The method that makes an AI answer from your documents: it first finds the useful passages, then writes its answer based on them and cites them.

Knowledge base

The set of reference documents the AI is allowed to consult: procedures, contracts, product sheets, internal FAQs.

Indexing

Preparing your documents so they can be searched quickly: they are read, split and filed, a bit like the index at the back of a book.

Chunking

Splitting a long document into short, coherent passages so the AI finds the right paragraph rather than a whole document.

Embedding

A translation of a text into a series of numbers that represents its meaning. Two texts about the same thing have close embeddings, even with different words.

Vector database

The database that stores embeddings and instantly finds the passages whose meaning is closest to a question.

Reranking

A second sorting pass: among the passages found, the AI puts first those that really answer the question.

Cited answer

An answer that shows the documents and passages it relies on, so anyone can check them in one click.

Connecting AI to your tools

What lets AI work inside the software you already use.

API

A doorway that lets two pieces of software exchange information automatically, without retyping. Example: the agent reads an order in your ERP through its API.

MCP (Model Context Protocol)

An open standard, published by Anthropic in late 2024, that connects an AI assistant to tools and data in a uniform way, a bit like a universal plug.

Connector (integration)

The link between the AI and a specific tool: your e-mail, your CRM, your document space. It respects the tool’s access rights.

Automation (workflow)

A chain of tasks that runs by itself. Example: each invoice received is read, matched with its order, and discrepancies are flagged.

Fine-tuning

Retraining a model on your own examples to teach it a style or a task. To answer from documents that change often, RAG is usually simpler.

Open model (open source)

A model whose parameters are published and that you can run on your own servers, such as Llama or some Mistral models.

On-premises hosting

The AI and your data stay on servers installed on your premises. It is the most controlled option, often requested for sensitive data.

Private cloud

Servers dedicated to your company at a hosting provider, for example in Europe, not shared with other customers.

Copilot

An assistant built into software you already use, helping with your task without doing it for you. Example: a CRM copilot that summarises an account before a meeting.

OpenOS

The engine built by Open Intelligence: every piece of information keeps its source and date, and checks confirm that each action the AI proposes is allowed before it happens.

Trust, security and compliance

The ideas that let you use AI without losing control.

Guardrail

A rule that limits what the AI can say or do. Example: never give legal advice, or never send an e-mail without approval.

Human approval

A step where a person reviews and approves what the AI prepared before it goes out. The tool prepares, the human decides.

Inherited access rights

The AI applies the same rights as your tools: an employee only sees, in answers, the documents they are already allowed to open.

Traceability

The ability to find out afterwards what the AI consulted, what it answered and who approved what.

Evaluation

Measuring the quality of an AI’s answers on a list of questions whose correct answer is known, repeated after every change.

Test set

The list of reference questions, with their correct answers, used to evaluate the AI and to check that an update broke nothing.

Prompt injection

An attack where instructions hidden in a document or web page try to hijack the AI. Guardrails and human approval limit its effect.

GDPR

The European regulation on personal data protection. It applies as soon as an AI processes data about people: customers, employees, candidates.

DPIA (impact assessment)

An analysis of the risks a processing operation poses to people, mandatory when that risk is high. It is prepared before the project starts.

AI Act

The European regulation on artificial intelligence, in force since 1 August 2024 and applied in stages. It classifies AI uses by their level of risk.

Missing a word? Ask us

We explain every technical choice in plain words before proposing a solution and a quote.