Overview
The core concepts behind Agoralia — agents, numbers, calls, and campaigns — and the gated agent layer that keeps an autonomous AI safe to operate.
Overview
Agoralia is a gated agent layer for AI voice calling. You build everything programmatically over MCP, REST, or CLI; the platform enforces compliance and risk gating; the dashboard lets a human observe and approve.
Core concepts
Agents
An agent is a configured voice persona: a system prompt, a language, a voice, and call behavior (first message, max duration, recording, knowledge base, compliance config). You create an agent, then publish it to sync it to the voice provider — a freshly created agent cannot place calls until it has been published.
See Agents.
Numbers
A phone number is what your agent calls from. You can buy a number from Agoralia's pool or link a Bring-Your-Own (BYO) number from Telnyx, Twilio, Vonage, Sinch, Bird, Infobip, Bandwidth, or SIP. Each number is scoped to your workspace and registered with the voice provider before it can be used.
Calls
A call is a single outbound (or inbound) conversation. To place an outbound call you need a
published agent, a phone number, and a destination in E.164 format (e.g. +39...). The platform
runs a DNC pre-check and applies per-country recording/disclosure rules automatically.
Campaigns
A campaign runs an agent against many contacts. Before launch, the platform runs a pre-flight check (agent ready, contacts present, compliance OK, budget sufficient) and requires a human "go" because launching is an irreversible, costly action.
See Campaigns.
The gated agent layer
Agoralia is built to be handed to an autonomous AI. Three responsibilities are deliberately split:
| You / your agent (build) | The platform (gate) | A human (observe + approve) |
|---|---|---|
| MCP · REST · CLI | RBAC · compliance · pre-flight · budget · risk tiers | read-only data · audit log · approvals |
Each action is classified by risk. Reversible and cheap actions proceed automatically; irreversible, costly, or legally sensitive ones (launching a campaign, provisioning a number, placing live calls) wait for a human approval in the dashboard.
The dashboard is read-only by design — there is no configuration wizard. You configure by code and observe in the UI. This is what makes the platform safe to expose to an autonomous agent.
How surfaces relate
The same gated core powers all three surfaces, so the concepts above are identical whether you use MCP, REST, or CLI. Differences are only in how you authenticate and how you call — see Authentication.