Quickstart: your first text agent

Create a text agent, choose its model, chat with it, and get the one-line embed code for your website. About five minutes.

A text agent answers in writing: on a web page through the widget, or in Slack, Telegram, Microsoft Teams, WhatsApp and Discord through a deployment. This page creates one and ends with the embed code for a website.

Create the agent

Open the wizard

In the sidebar choose AI Agents, then Create Agent.

Choose text, then a single prompt

Pick Text Agent, then Single Prompt Agent.

Start from scratch

Choose Start from scratch. The agent is created at once, named New Text Agent, and opens in its workspace. There is no voice step: a text agent has a model instead of a voice.

Creating a text agent takes three choices.

Name it and pick a model

Rename it

Under General, replace the name with your own and add a one-line Description. Set Language to the language your customers write in.

Choose the model

Under Model, pick the LLM that answers, for example GPT-5.4 Mini. The choice can be changed at any time.

Write the prompt and save

Fill Agent Identity, Tasks and Guardrails as in the voice quickstart, then Save.

The Model picker in a text agent's settings — The Model section of a text agent.

Chat with it

Under Testing, choose Chat. Type a message and send it. The reply streams in, and each turn shows its response time.

The Chat test.

Get the embed code

Open the widget

Under Release, choose Widget. The configurator shows a live preview on the right; the presets and the Layout, Avatar, Text, Colors and Features panels change it.

Copy the code

Choose Embed Code, then Copy. It is one custom element and one script tag. Paste it before the closing body tag of any page.

Make the agent public

An embedded agent has to be reachable without signing in. Choose Share agent at the top of the workspace, then turn on Agent Visibility.

The Embed Code dialog with the callab-chat element and script tag — The embed code for a text agent.

Verify

Conversations in the main sidebar lists the chat within a few seconds: the agent, the channel (Web), turns, tokens and status.
Open it to read the thread, the model that answered and the token usage per turn.

Troubleshooting

The widget loads but the agent does not answer. The agent is still private. Turn on Agent Visibility under Share agent.
The reply says the model does not exist. The agent still has the model it was created with. Pick one under Model and save.
The reply is in the wrong language. Set Language under General, and say so in Guardrails as well.

What next

Deploy to Slack or WhatsApp

Put the same agent in the apps your customers already use.

Style the widget

Presets, colours, labels and placement.

Give it knowledge

Answer from your own documents.

Capture outcomes

Structured fields after every conversation.