Core concepts

The objects Callab AI is built from, and how they relate: agents, versions, tools, outcomes, phone numbers, campaigns, deployments, contacts and workspaces.

Everything in the dashboard is one of a dozen things. Learn these and every screen makes sense.

Agent

The thing that talks. An agent has a modality, voice or text, chosen at creation and fixed for its life. It has a prompt type: a single prompt, or a flow made of connected nodes. It has a language, a voice (voice agents) or a model (text agents), a prompt, tools, outcomes and advanced settings.

An agent lives in the agent workspace, reached from AI Agents. The workspace has four groups of sections: Build (Settings, Advanced, Flow, Tools, Outcomes, Memory), Testing (Chat, Call, Evaluation), Improve (Recommendations, Change Requests) and Release (Publish, Widget, Deployments, Versions). Which sections appear depends on the agent: only flow agents have Flow, only voice agents have Call, only text agents have Chat and Deployments, and Memory appears once data retention and memory are on.

Version

Every save creates a version. Versions are numbered v<major>.<revision>, listed newest first under Versions, and each one records exactly what changed. You can compare two versions and roll back to any of them. Saving does not take a version live: Publish does, once the launch checks pass.

Prompt

The agent’s brief, written in three fields: Agent Identity (who it is), Tasks (what it does) and Guardrails (what it must and must not do). A fourth setting, First message, is what it says when it speaks first. The prompt can be generated from a description, a website or a document with Generate With AI.

Tool

An action the agent may take during a conversation. Built-in tools are End Call, Transfer Call, Redirect to Agent, Press Keys (DTMF), Schedule Callback, Live Webhook, Send Email, Cal.com, Zapier, Knowledge Base and MCP. Tools have their own section, Tools under Build, where Add tool adds one and each configured tool is listed with its description. The description is what the model reads to decide when to use it.

Dynamic variable

A placeholder in the prompt, filled per conversation: from a contact’s metadata in a campaign, or from a default value. Managed from the Variables button next to the prompt.

Outcome

A structured field the agent fills once a conversation ends, such as customer_intent with a fixed list of possible values. Outcomes appear as columns in Call Logs and on each call’s page. Defined under Outcomes in the workspace.

Phone number

A number an agent can answer on or call from. Numbers come from a SIP trunk, your own Twilio account or a Go Auto Dial extension. Managed under Phone Numbers.

Campaign

How voice agents reach people. An outbound campaign calls a list of contacts from a phone number with an agent, within calling hours and with a retry policy. An inbound campaign answers calls to a number with an agent. Managed under Launch AI Voice.

Contact

A person with a phone number, a category, a tag and optional metadata. Contacts are the audience of outbound campaigns and are managed under Contacts, one at a time or by importing a file.

Deployment

Where a text agent is live: a Slack workspace, a Telegram bot, a Microsoft Teams app, a WhatsApp Business number or a Discord server. Managed under Launch AI Text, or from the agent’s own Deployments section.

A chat or call interface for a web page, styled in the agent’s Widget section and embedded with one script tag.

Call and conversation

A call is one phone or browser conversation with a voice agent, with a recording, a transcript, a summary, quality scores and outcomes. A conversation is the same for a text agent: turns, tokens, tool calls. Found under Call Logs and Conversations.

Workspace and company

A company is your account: members, billing, storage, phone numbers. A workspace is a partition inside it that holds agents, campaigns and contacts. Every company starts with a Default Workspace; switch with the selector at the bottom of the sidebar.