Glossary
The terms that come up most often in automation and AI projects, defined in a sentence or two. Where a detailed article exists, the link is given.
- AI agent
A software system combining an artificial intelligence model with data, tools and rules to carry out a task or a sequence of tasks.
Read the article → - API
An interface a program provides so that other programs can exchange with it in a stable way, without going through its screen.
- Automation
Software carrying out steps a person used to do by hand, following rules or a model.
Read the article → - Cloud
Running software on a provider's infrastructure, reachable over the network, rather than on hardware the company owns.
- CRM
Customer relationship management software: contacts, opportunities, history of exchanges.
- Embedding
A numerical representation of a text that lets a machine compare content by closeness of meaning rather than by identical words.
Read the article → - ERP
Integrated management software bringing together functions such as purchasing, stock, production, invoicing and accounting.
- Fine-tuning
Partial retraining of a model on examples specific to one use, in order to adjust its behaviour or style.
Read the article → - Human-in-the-loop
An automated process in which a person validates or corrects at a defined point before the system acts.
Read the article → - Inference
Running an already trained model to produce a result from an input. This is what consumes resources at every use.
- LLM
Large language model: a model trained on vast text corpora, able to produce and interpret natural language.
- MCP
Model Context Protocol: an open protocol that standardises how a model reaches external tools and data sources.
- OCR
Optical character recognition: converting an image or scanned document into text a machine can work with.
- On-premise
Said of software running on the company's own hardware, in its own premises or its own dedicated hosting.
Read the article → - Prompt
The instruction and context given to a language model to obtain a particular result.
- RAG
Retrieval-augmented generation: a method that retrieves the relevant passages of existing documents and hands them to the model so it answers from those sources.
Read the article → - Vector database
A database that stores embeddings and can quickly retrieve the content closest to a question.
Read the article → - Webhook
An automatic call sent by a program to a defined address when an event occurs, to trigger processing elsewhere.
- Workflow
An ordered sequence of steps describing how a task travels through an organisation, from its trigger to its result.
Read the article →