Glossary
Glossary of AI assistants
The words you meet as soon as AI assistants come up, from language models to the GDPR, explained with concrete examples and what they change for an organization that puts one online.
A
- AI Act
- The AI Act is European Regulation (EU) 2024/1689 on artificial intelligence: it sets obligations proportionate to the risk of each use, including telling people when they are talking to an AI.
- AI agent
- An AI agent is a system that pursues a goal by chaining several steps, choosing for itself which actions to take and taking their results into account.
- AI assistant
- An AI assistant is software that answers questions in everyday language using a language model, often drawing on an organization’s documents.
C
- Chat widget
- A chat widget is a conversation window within a web page, often opened from a bubble, where visitors write to an assistant or team without leaving the site.
- Chatbot
- A chatbot is a program that converses with visitors in writing, following a predefined script, generating answers with a language model, or combining both.
- Contact memory
- Contact memory is what an AI assistant retains about a person between conversations (preferences, context, requests), for a limited time and in compliance with the GDPR.
- Context window
- The context window is the maximum amount of text, measured in tokens, that a language model can read in one go: prompt, history, documents and question included.
D
- Data processing agreement (DPA)
- A data processing agreement, or DPA, is the contract required by Article 28 of the GDPR between an organization and the provider that processes personal data on its behalf.
- Data retention
- The retention period is the length of time an organization keeps personal data before deleting or anonymizing it, depending on how useful the data really is.
E
- Embedding
- An embedding is a series of numbers that represents the meaning of a text, so that two texts with similar meaning get similar series.
F
- Fine-tuning
- Fine-tuning means continuing to train a language model on chosen examples to permanently change its style, its format or how well it performs a task.
G
- GDPR
- The GDPR, Regulation (EU) 2016/679, has set since 25 May 2018 the rules every organization must follow when it processes the personal data of people in the European Union.
- Generative AI
- Generative AI is the branch of artificial intelligence that produces new content, text, images, sound or code, from a request written in everyday language.
- Guardrails
- Guardrails are all the measures that keep an AI assistant in its role: answering from the right documents, not making things up, resisting manipulation and handing over.
H
- Hallucination
- An AI hallucination is a generated answer that presents invented, inaccurate or unsupported information as fact, such as a price or procedure that does not exist.
- Human handover
- Human handover is the transfer of a conversation from an AI assistant to a team member, with its full history.
I
- Intent
- An intent is what a person wants to achieve by writing a message, whatever words they use: finding opening hours, asking for a price, reporting a problem.
K
- Knowledge base
- A knowledge base is a structured collection of documents an AI assistant can consult to answer: website pages, FAQs, guides, procedures and product sheets.
L
- Lead qualification
- Lead qualification means gathering information that shows whether a prospect’s request fits the offer: their needs, timeframe, context and contact details.
- Live chat
- Live chat is a conversation window on a website in which a visitor exchanges written messages, in real time, with a person from the business.
- LLM (large language model)
- An LLM (large language model) is an artificial intelligence model trained on large amounts of text to understand a written question and write an answer.
N
- Natural language processing (NLP)
- Natural language processing, or NLP, covers the techniques that allow a computer to read, understand, analyse or produce human language.
P
- Prompt
- A prompt is the text of instructions a language model receives with every question: it sets its role, tone, sources and what to do when it does not know.
R
- RAG (retrieval-augmented generation)
- RAG (retrieval-augmented generation) combines a document search with an answer written by a language model from the retrieved passages.
- Resolution rate
- The resolution rate measures the share of conversations in which the visitor's request was settled, by the assistant alone or after a person took over.
S
- Semantic search
- Semantic search finds passages of text by their meaning rather than by exact words, so that a question worded differently can still be answered.
- Shared inbox
- A shared inbox is a common space where the conversations a team must handle arrive, with the assignment, history and status of each one.
- Source citation
- Source citation means an AI assistant shows the document or passage behind its answer alongside it, so the reader can check it.
T
- Token
- A token is a fragment of text, often a short word or a piece of a word, that language models use to read, write and bill their work.
- Tool calling
- Tool calling lets a language model request a specific action, with structured parameters, which the system runs before sending the result back to the model.
V
- Vector database
- A vector database stores embeddings and quickly finds the passages of text whose meaning is closest to a question.
W
- Webhook
- A webhook is a mechanism that automatically sends a message to a web address when an event occurs, such as a new contact, so another tool can respond.
See it on your own cases
Everything starts with a one-hour audit of your questions, your documents and your tools. Then we propose the level that fits.
Request a demo