Publish and share your agent

Before publishing

Complete a final review:

• The name and description identify the agent correctly.
• Instructions define what it can and cannot do.
• Knowledge bases are selected and documents are indexed.
• Only necessary tools are authorized.
• Main questions and edge cases have been tested.
• Behavior for unknown information is safe and useful.

Activate the agent

In Studio, review Agent status. An active agent can respond in Studio and published channels. An inactive agent does not accept new requests.

Choose a channel

Open the agent's Channels tab. Available options depend on your organization and may include Web Widget, API, Microsoft Teams, or WhatsApp.

The Web Widget is usually the most direct first public channel. You can configure the assistant name, primary color, welcome message, and page position.

Limit source exposure

Before publishing, review Sources in responses. You can choose one of these maximum levels:

• Do not show sources.
• Show references.
• Show references and excerpts.
• Allow document access.

Each channel can apply a stricter setting. Internal or confidential documents must not be exposed publicly.

Run an external test

After enabling the channel:

1. Open the experience as a user would.
2. Ask known and unknown questions.
3. Check the welcome message and visual identity.
4. Review the mobile experience.
5. Confirm that unauthorized sources and actions are not shown.

Monitor the result

The Dashboard and Audit areas help you review activity and relevant changes. Regularly check conversations, usage, tool errors, and integration status.

Publishing is not the end of the work: use real queries to improve instructions and knowledge without changing the agent's URL or channel.