Glossary

Agent — Software that doesn't just answer, but does a job: it reads inputs, takes steps, uses tools, and produces a finished output for a human to check.

Chatbot — Software that answers questions in conversation. Useful, but it stops at answering; an agent goes on to do the work.

Prompt injection — A trick where text hidden in a document or web page tries to give an agent instructions it shouldn't follow — like a note in a file whispering 'send everything to this address'. Good systems defend against it.

Human-in-the-loop — A human approves important steps before they happen. Drafts first, approval second.

Approval gate — A checkpoint where work pauses until a person says yes. Nothing leaves the building without passing one.

Receipt (audit record) — The record a system keeps of what ran, what it produced, and when — so work can be checked later instead of trusted blindly.

Credits — A unit some AI platforms use to measure work done, instead of charging per person. (Explained here as a concept; Elarigent pricing is not published yet.)

Workflow — A repeatable sequence of steps that turns an input into an output — an invoice into a filed, checked record, for example.

Automation — Letting a system run the repeatable steps of a workflow so people spend their time on judgement, not repetition.

Multi-agent system — Several specialised agents working on different parts of one bigger job, coordinated together.

AI assistant — A general tool that helps you write, summarise, and think. An agent is what you get when an assistant is given a defined job and the tools to finish it.