Agent configuration

Click Configure on the agent card to open the settings. Each row opens one panel. A row with a warning sign needs your attention before the agent can be activated. If you have Can view or Can use access, the panels open read only.

Agent Profile

Agent Icon, Agent Name and Description. The description is shown to people in chat, so write it for them.

Instructions & Behavior

• System Prompt defines how the agent behaves and what it knows. Be specific. It is required for activation.
• Response Mode: Fast Mode for quick answers to simple questions, Detailed Mode for thorough answers with more context.
• Capabilities: Web Search, Image Generation (with Standard or High Quality mode), Graph Generation for charts, and Next Questions Helper for suggested follow-ups.
• Security Controls: PII Protection, Advice Restriction (medical, legal and financial advice) and Content Restriction. All three are on for a new agent.

Model

Pick a Model Provider, then a Model. A model is required for activation. Some providers need your organization's own API key; they show API Key Required and an Add API Key button. On a free trial, some models show Not available in Free Trial.

Add to Agent

The Add to Agent row, and the + button on the canvas, attach what the agent works with:

• Add connection: a data source the agent reads from.
• Add Tool: an integration the agent can act in.
• Add Training: a training set.

You only see items shared with you at Can use or higher. Anything else is hidden from the list. See Knowledge sources and How integrations work.

Skills

Select the skills this agent should apply. Each shows a badge: System, Custom or Modified.

Output Channels

The channels are independent: one agent can serve web chat, the API and the widget at the same time. With API Access on, Create Key makes an API key; see REST API. For the widget, Configure the widget opens the widget settings: brand, content, welcome screen, theme, live support, layout and install instructions, with a live preview. You change the widget there, with no code changes on your site. See Web SDK.

Conversation Starters

Up to 8 ready-made prompts shown when someone opens a new chat with the agent. Keep each one short and grounded in what the agent really does.

Feedback

Answers people rated Bad Response in chat, with the reason they chose and the answer itself. Give each item a status (To Do, In Progress, Completed or Ignored) to track what you have fixed. Search and filter by status.

Agent Management

Activate or deactivate the agent, and delete it. Delete Agent is shown only to the owner. See Agents for the lifecycle.