Core concepts

VerticalAI has a small vocabulary. Learn these six ideas and everything else in the product reads cleanly.

Workspace

A workspace is your tenancy — one team, one billing account, one set of resources. Phone numbers, knowledge bases, members, and credits all belong to the workspace, not to any single agent. Everything you build lives inside it.

Agent

An agent is the thing that answers the phone. Each workspace has one agent slot. The agent is a single language model driven by a system prompt and a flat list of tools — there is no diagram to draw and no graph to wire up. The model reads the prompt on every turn and decides what to say and which tool to call. The agent is inbound only: it answers calls, it never places them.

System prompt

The prompt is the persona and the rules — who the agent is, how it speaks, what it must and must not do. The model reads it on every turn, so it is the single biggest lever over how a call feels. You edit it on the Prompt page.

Tools

Tools are how the agent does real work: look up an order, book a meeting, search a knowledge base, transfer to a human, end the call. Each tool is a self-documenting unit — a name, a plain-language description, and a set of parameters. The model reads the description and decides when to call it. A tool's description is the highest-leverage prompt surface you have, so write it carefully. Tools come in two kinds: builtin actions the platform ships, and webhook tools that call your own HTTP endpoint. See the Tools guide.

Versions and publishing

Your edits land in a draft. Publishing promotes the draft to the live configuration that answers real calls. Exactly one version is published per agent at a time, and a published version is read-only — to change a live agent you create a fresh draft, edit it, and publish again. Publishing runs the validator first; any blocking issue stops the publish until you fix it.

Calls

A call is one conversation between a caller and the agent. It starts when the caller connects and ends when someone hangs up or the agent ends the call. Every call is transcribed and scored, and the results flow into transcripts and analytics so you can see how real calls are going and keep improving. Calls are also the unit you are billed on.

How it fits together

You build an agent inside a workspace by writing its prompt and tools, test it in the test panel until it behaves reliably, publish it, attach a phone number, and let it take calls. The getting started guide walks that path end to end; the guides below go deep on each surface.