The agent is doing nothing and is waiting for the user — the conversational turn has ended.
The agent is working on an answer that has not started streaming yet.
Expressive (V4) only: set when a turn starts and when the user's speech has just been transcribed. Use it for a typing or thinking indicator.
The agent's video has started; the agent is speaking.
The agent is running one or more tool calls.
Expressive (V4) only: set when a tool call starts, and returned to after the agent finishes speaking while any call is still outstanding (for example a client tool waiting on the user). The individual calls are reported by onRunningToolCallsChange.
What the agent is doing right now, for driving the application's UI.
The argument of onAgentActivityStateChange. Expressive (V4) agents move through all four states; Talks (V2) and Clips (V3) agents only alternate between TALKING and IDLE, and only on fluent streams.