Build your first AI agent
What you'll build
Before you start
3
Next steps
Last updated
A step-by-step walkthrough of building a working AI agent, end to end, using one example: a complaints assistant for a financial institution. Each step links to the reference page for that feature.
This is a how-to article. For what each feature does in detail, follow the links to the module pages at each step.
An AI agent that greets a person, recognises a complaint, collects its details, logs a ticket, and hands off to a live agent when needed.
A workspace with the AI Agents module enabled.
Permission to manage AI agents, set in Roles and permissions settings.
The persona defines how the agent speaks and behaves. In the Agent tab, set a name, then write a persona that gives the agent a role, a tone, and any limits.
For a complaints assistant, the persona might describe a calm, empathetic handler that acknowledges frustration and never offers financial advice.
See Manage AI agents for the full Agent tab.
Knowledge is the content the agent draws on to answer questions. In the Knowledge tab, add a training source — a document, a URL, or a pasted text snippet — such as your complaints policy.
See Knowledge for the training source types and limits.
The workflow is the scripted flow of triggers and actions. In the Workflows tab, build the complaint flow:
Message received: "I want to complain"
├─ Send message: "I can help with that. What's it about?"
├─ Ask survey → {financial_institution}, {date}, {complaint_details}
├─ Create ticket → Subject: "Complaint about {financial_institution}"
└─ Live agent handoff → Complaints teamReference pages: Message received · Send message · Ask survey · Create ticket · Live agent handoff
See Workflows and actions for the full list.
Use the AI agent tester beside the workflow to chat with the agent and check each step fires as expected. Switch the tester's language to confirm any localised content.
See Manage AI agents for the tester.
Connect the agent to a channel — such as webchat, WhatsApp, or Messenger — from the Publish tab, and make it live.
See Manage AI agents for publishing.
Add a Branch to route complaints by type.
Use LLM processing to extract structured details from the complaint.
Add more languages in the Agent tab and localise the workflow.
Last updated