Agents
What a voice AI agent is in Intrlume, what it's made of, and how it fits with the rest of the platform.
An agent is the AI that holds a phone or web conversation on your behalf. You configure it once — its voice, languages, conversation flow, knowledge, and tools — and it then handles calls at scale, around the clock. Every agent belongs to your organization.
What an agent is made of
Conversation (workflow)
The flow it follows — its prompt, greeting, and the steps and branches of the conversation.
Voice & language
The synthesized voice it speaks with, plus a primary language and an optional second language it can switch to.
Knowledge bases
Optional documents the agent can search mid-call to answer questions accurately.
Tools
Actions it can take during a call — look up a record, book a slot — through built-in or custom integrations.
Evaluation
Optional automatic scoring of each call for faithfulness, hallucination, and topic adherence.
Behavior
How it sounds and reacts — natural turn-taking, interruption handling, and optional ambient background sound.
Agent types
An agent works on one of three channels:
The agent calls your contacts — for reminders, follow-ups, surveys, or lead qualification. Driven by a campaign over an audience.
The agent answers calls to your number. Requires a phone number (DID) linked to the agent.
The agent talks to visitors through a click-to-talk widget embedded on your website — no phone number needed.
How you create one
The fastest way is to describe the agent in plain language — Intrlume generates the prompt, greeting, voice, and conversation flow for you in moments. You can then refine any part in the visual workflow editor.
See the Quickstart for a step-by-step walkthrough of creating and testing your first agent.
Where an agent fits
The agent sits at the center of the Intrlume loop:
- It runs a workflow — its conversation logic.
- You point it at an audience and launch a campaign to make outbound calls, or link a number for inbound.
- Each conversation becomes a call, with a recording, transcript, outcome (disposition), and extracted data.
- Evaluation scores call quality so you can refine the agent and improve results.
An agent's status moves from generating (while Intrlume builds it) to ready once it can take calls. You can stop an active agent at any time.