Configure an AI Customer Service Agent in 3CX

Introduction

This guide explains how to configure the 3CX AI Customer Service Agent and integrate it into your customer-facing workflow. The Customer Service Agent answers calls professionally, identifies the issue, searches knowledge sources, responds briefly, asks one question at a time, and escalates only when necessary.

Before You Start

To help ensure the Customer Service Agent performs as intended from the start:

  • Plan and prepare the knowledge sources the agent should use: documentation, FAQs, help center articles, policies, product guides, or internal support notes.
  • You can find more guidance on how to structure and format these here.
  • Avoid trying to create a very complex AI Agent to handle all scenarios; split your solution across multiple, more targeted, agents; this will yield better end results and make agents more responsive.
  • Decide where unresolved calls should go:
  • A question might be too complex, or outside the AI agent's scope.
  • These calls will need to be correctly forwarded to keep callers satisfied.
  • For most customer service deployments, the default human handoff destination is typically a support queue.

Enable AI Agent Integration and Add Knowledge Sources

Follow our guides to set up AI Agent integration with OpenAI.

Well-structured Knowledge Sources are key to a successful deployment. More can be added later on, as needs change, but an agent's core knowledge should be ready before activating the agent.

To add Knowledge Sources, navigate to Admin > Integrations > AI > Knowledge Sources.

Create The Agent

Navigate to Admin > Integrations > AI and expand Add Agent. Choose Customer Service.

Agent Info

Configure the agent's basic information. It is recommended to keep the Maximum Call Duration under 15 minutes to avoid exhausting the AI Agent's context window.

Agent Type

In this section, we assign the Knowledge Sources this Agent will have access to. You can also use the Add button to quickly go to the Knowledge Source management section (make sure that any changes to the Agent setup are saved first). Similar to other AI Agent types, you can set Language options and tune the First Message.

Call Routing

Customer Service agents have two unique and very significant configuration sections in the Call Routing tab:

Human Handoff

When an AI agent fails to resolve a query, or the caller demands to be connected to a human agent without giving enough details to permit a more precise routing, this is the Handoff destination that the call will be transferred to.

Escalation Topics

Escalation topics define specific targets that can offer specialised service to the caller based on their needs. This can be human agents, queues, or even other AI Customer Service Agents with more specialized knowledge.

  • You can add multiple Escalation Topics manually using the Add button, or via CSV Import.
  • Already defined Escalation Topics can be Exported in CSV format, for reuse in other AI Customer Service Agents.
  • Follow the main principles of AI prompting: keep topic descriptions short and unambiguous.

Spam Filter

Like our previous AI Agents, you can define certain call types as Spam, letting the agent end the call quickly and not waste time unnecessarily for both the Agent, and the Handoff destinations.

Options

This tab defines:

  • The action the AI Agent takes in cases of Abusive/Frustrated callers.
  • The action the AI Agent takes in cases of Non Responsive/Non Collaborative callers.
  • The routing destination for the call in case the AI engine is not reachable.
  • Whether the Agent checks availability before attempting call transfers.

AI Prompt

These are the core instructions defining the AI Customer Service Agent's behaviour and role. If you would like to adjust the prompt to better fit your needs, including possibly to adjust the agent's personality and tone, remember to always make small changes, one at a time, and test incrementally.

Check here for guidance on correct AI prompt construction:

See also

Last Updated
This document was last updated on 30 July 2026

https://www.3cx.com/docs/ai-customer-service-agent/