Words explained

A handful of technical words are hard to avoid, mostly because they are printed on buttons in the product. Here is what each of them means in practice. One sentence, no background needed.

Agent

A helper set up once with standing instructions, then reused. Instead of explaining the same job every time, you write the instructions down, give them a name, and switch the agent on when you need it. See Projects, agents and tasks.

API key

A long secret string that lets a program sign in, the way a password lets a person sign in. You only need one if you connect a tool on your own computer to zGentic; see Model as a Service.

Assistant

The thing that answers you in chat. It is a model (see below) plus everything zGentic gives it: your documents, your memories, the web when you allow it, and the ability to run steps in order.

Blocked

The label on a step of a plan that was never attempted, because the step it depended on did not finish. The card names what it was waiting for. See when a step fails.

Connector

A link to another system — Jira, Confluence, a weather service — so the assistant can look things up there while it answers. See Connectors.

Context

Everything the model is shown before it writes an answer: your question, the earlier messages in the chat, and any documents or memories that were pulled in. Each model can only hold so much at once, which is why very long conversations eventually start to lose their beginning.

Model

The engine that writes the answers. Models differ in speed, price, how well they cope with difficult work, and whether they can look at pictures. Your organisation decides which ones you can choose from.

Plan

The ordered list of steps the assistant draws up when a request is really several jobs. Steps that need another step's result wait for it; the rest run at the same time. You watch it as a card in the chat and in the Tasks panel. See Projects, agents and tasks.

Project

A shared place for one topic, holding its chats, its files and its results together. Sharing the project shares all of it at once. See Projects, agents and tasks.

Retrieval, or RAG

Answering from your own documents. The assistant searches the files you and your projects have uploaded, uses the passages that fit, and tells you which file each part came from. RAG is short for retrieval-augmented generation, which is the same idea stated technically.

Sandbox

A sealed, temporary workspace where the assistant can run a small program it wrote itself — for example to turn a table into a spreadsheet. It cannot reach anything else while it runs, and the time it takes appears in your organisation's usage.

Thinking

An option that lets the assistant work a problem out step by step before answering. It gives better results on hard questions and costs more; the working-out is shown, folded up, beneath the answer.

Token

The unit models are measured and priced in. One token is a chunk of text a little shorter than an average word. Model prices are quoted per million tokens, and both the text you send and the text that comes back are counted.

Tools

In zGentic, the panel behind the + button beside the message box, where you decide what this chat may do: search the web, read your documents, think longer, call an agent, reach another system.

Turn

One exchange — your message and the answer to it. The word appears in the message you see if an answer is interrupted before it finishes. See if an answer disappears.