Outputs: Server Events

Server Events are messages sent from the Deepgram server to the client over the WebSocket connection during a voice agent interaction. These events provide real-time updates about the conversation's status, including user and agent actions, as well as any processing that's occurring.

• Welcome: Confirms that the WebSocket has opened successfully.
• Settings Applied: Confirms that the configuration settings have been applied.
• Conversation Text: Provides the text of what was spoken by either the user or the agent.
• User Started Speaking: Notifies that the user has begun speaking.
• Agent Thinking: Informs the client that the agent is processing information.
• Function Call Request: Sent when the agent needs to make a function call.
• Function Call Response: Sent to provide information about a function call.
• Function Call Cancelled: Sent when a function call the client already received is cancelled because the user started speaking again, either inside the speculative window or after the turn was confirmed.
• PromptUpdated / SpeakUpdated / ThinkUpdated: Server confirms that an Update* message has been applied.
• Agent Audio Done: Indicates that the server has finished sending the final audio segment to the client.
• Error / Warning: Server reports a fatal error (Error) or non-fatal issue (Warning).
• Latency Report: Provides a detailed STT, LLM, and TTS latency breakdown for each turn.

Each of these events serves a specific purpose in managing the flow of the conversation and keeping the client informed about the state of the interaction. They allow for a dynamic and responsive experience when using Deepgram's Voice Agent API.