Glossary

A short reference for the words used throughout Gabbex.

Assistant

A single AI configuration owned by a workspace. An assistant has its own knowledge sources, tools, channels, conversations, and leads. Most teams have one assistant. Larger teams have several — for example, a “Sales” assistant and a “Support” assistant — and switch between them in the dashboard.

Channel

A surface where the assistant runs. The website widget, Shopify, WhatsApp, and Facebook Messenger are all channels. The same assistant can run on multiple channels at once.

Chunk

A small slice of content from one of your sources. When Gabbex indexes a website page or a PDF, it splits the text into chunks of a few hundred words. The assistant retrieves chunks at answer time, not whole documents. You will see the term in technical articles about RAG; you do not need to interact with chunks directly.

Conversation

A single back-and-forth exchange between a visitor and the assistant. A conversation can be one message or fifty. Conversations themselves are not metered — only the AI replies inside them are, in credits. You can browse every conversation in the Conversations dashboard.

Credit

The unit Gabbex uses to meter AI replies. Every reply your assistant sends costs a flat number of credits, determined by the model — 1 credit for fast models, 2 for mid-tier, 3 for top-tier reasoning models. Each plan includes a monthly credit allowance for the entire workspace, which resets at the start of each billing period. See Credits and model costs for the full breakdown.

Embedding

A vector representation of a piece of text that captures its meaning. Used internally for semantic search. You do not configure embeddings — Gabbex creates them automatically when a source is indexed.

Escalation

A handoff from the assistant to a human. Triggered by the escalate-to-human tool when the assistant is not confident, the topic is out of scope, or the visitor explicitly asks for a real person. Escalations show up in the notification bell and in the Conversations dashboard.

Guardrails

The built-in safety rules that apply to every Gabbex assistant: stay in scope, do not leak account details, refuse unsafe requests, do not make things up, escalate when uncertain. Guardrails are always on and cannot be disabled.

Knowledge base

The combined set of all sources connected to an assistant — every website page, file, Notion page, and Q&A entry. The assistant answers from its knowledge base and nothing else.

Lead

A visitor who left contact details with the assistant. Created automatically by the lead capture tool or by an escalation. Visible in the Leads dashboard.

Member

A user account that belongs to a workspace. Members have a role (owner, admin, member, or viewer) that controls what they can do.

Operator

A team member who responds to escalations. Operators are configured per assistant on the escalation tool. They receive email and in-app notifications when an escalation fires.

RAG (Retrieval-Augmented Generation)

The technique Gabbex uses to ground answers in your content. When a visitor asks a question, Gabbex retrieves the most relevant chunks from the knowledge base and uses them to compose an answer. RAG is what stops the assistant from making things up.

Source

One unit of knowledge connected to an assistant — a website crawl, a file upload, a Notion connection, or a Q&A list. An assistant can have many sources. See Knowledge sources overview.

System instructions

The hidden rules the assistant follows in every conversation. Includes the built-in guardrails and any custom instructions you have added under General → Custom instructions.

Tool

An action the assistant can take during a conversation, beyond plain replies. Lead capture and escalation are tools. See Tools overview.

Workspace

The top-level container that owns assistants, members, and a subscription. You can be a member of more than one workspace. See Workspaces, members and roles.