Build Your First AI Agent

Sign in to the workspace that will own the agent. Have your public website URL ready, along with the support questions you want the agent to answer.

Your workspace plan determines how many agents and how much training data you can use. Review the available capacity if Create new chatbot opens the plans page.

1. 1From the workspace, choose Create new chatbot.
2. 2Enter your business website in Scan Website and select the objective for your agent.
3. 3Continue to Data Sources and review the discovered pages. Keep pages that explain your product, service, policies, and support processes.
4. 4Create the agent and open Build → Data sources to follow training progress.
5. 5Open the website source to review individual URLs. Resolve failed pages and confirm the pages you need are Trained.

Open Playground and review the business instructions generated during setup. Adapt the role and examples to your actual business, then save your changes.

1. 1Ask a common question that your source pages answer.
2. 2Check the answer against the original page, including any links or steps.
3. 3Ask an unclear question and a question your sources cannot answer.
4. 4Verify that the agent asks for clarification or acknowledges missing information.
5. 5Configure and test an action or human handoff if your support process needs one.

For a software support agent, useful questions include how to embed the widget, how to add a knowledge source, and how a teammate can take over a conversation.

1. 1Open Channels or the Playground Deploy menu and select the chat bubble.
2. 2Review the widget appearance, AI Chat, Visibility, and allowed-domain settings.
3. 3Copy your agent's embed code into a page you control.
4. 4Open that page as a visitor, open the widget, and send a message.
5. 5Confirm the reply and check that the conversation appears in the correct workspace.

Client-side actions must also be tested from the embedded page, where your registered website functions are available.