Your agent has one identity across its face, voice, instructions, knowledge, and tools. The character is a look that the agent wears; changing the look does not create a second agent.
The first meeting
A visitor sees a greeting caption right away. The agent names itself as an AI, asks the page's useful first question, and lets the visitor know they can interrupt. Sound starts only when the browser permits it.
The visitor stays in control
The visitor can use the agent's one menu to change language, mute sound, move or resize the character, ask for the camera, or end the session. Camera and microphone permission come from the browser and require consent.
What the agent can do
An agent can answer from its instructions and connected knowledge, navigate to a page, point to a control, or use an allowed tool. A tool that writes or acts still follows its permission and approval rules.
Next: Create your own agent.