Glossary
Plain-language definitions of every term used across the platform — no jargon required.
Every term you will meet in the product, explained in plain language. If a word in another guide sends you here, this is the short answer.
Agent
The thing that answers the phone. One per workspace. It is a language model following your instructions, deciding what to say and what to do on each call.
Workspace
Your account space — one team, one billing account, one set of phone numbers and agents. Everything you build lives inside a workspace.
System prompt
The written instructions that tell the agent who it is and how to behave. The agent reads them on every turn. Think of it as the job description and the rule book combined.
Tool
A specific thing the agent can do during a call — look up an order, book a meeting, transfer to a person, hang up. You describe each tool in plain language and the agent decides when to use it.
Builtin tool
A tool the platform provides ready-made: transfer to a human, book a meeting, search the knowledge base, get the current time, end the call.
Webhook tool
A tool that calls one of your own web addresses to fetch or send information — how the agent talks to your own systems.
Variable
A reusable named value, written as {{like_this}}, that gets filled into a
tool when the agent uses it. Lets you store a setting once and reuse it.
Secret
A variable whose value is sensitive — an API key or password. It is stored encrypted and never shown again after you save it.
Knowledge base
The reference material you give the agent — documents, web pages, pasted text — so it can answer questions that are not in the prompt.
Filler
A short natural phrase ("one moment", "let me check") the agent says while it works, so the caller is not left in silence.
Voice
How the agent sounds — the chosen voice, its speed, its tone, and the language it speaks.
Model
The language model that powers the agent's thinking. Faster models feel more responsive; stronger ones reason better. You pick the balance.
Draft
Your work-in-progress version of the agent. Changes go here first and do not affect live callers until you publish.
Publish
The action that makes your draft the live version that real callers reach. The platform checks for blocking problems before it lets you.
Version
A snapshot of the agent's configuration. One version is live (published) at a time; the rest are history or drafts.
Validator
The automatic check that runs before publishing. It flags errors (which block publishing) and risks (which warn but do not block).
Call
One conversation between a caller and the agent, from connect to hang-up.
Resolved call
A call that reached an end and that the automatic judge agrees actually helped the caller. Resolved calls are what you are billed for.
Eval
A scripted test conversation, graded automatically, that you run repeatedly to check the agent behaves reliably before you ship a change.
Judge
An automatic reviewer (itself a language model) that decides whether a call or an eval scenario actually helped the caller.
Transcript
The full written record of a call — every line spoken and every tool the agent used.
Builder Chat
The built-in assistant that edits the agent for you when you describe what you want in plain language.
Credits
What you spend to use the platform. Pay-as-you-go — you are charged on outcomes, like a resolved call, rather than on raw usage.
Skill
A reusable capability you build once at the workspace level and refine in its own editor.
MCP
A connector that plugs an external system into your workspace so the agent can use its capabilities.
Phone number
The inbound number callers dial to reach your agent. Bought and managed at the workspace level, then routed to the agent.
Outcome
A label a call can be tagged with when it ends, used to group calls in your reporting.