Module agent 

Voice Agent (/v1/agent/converse) types.

It contains:

• endpoint::Endpoint — custom LLM/TTS endpoint URL + headers.
• aws_credentials::AwsCredentials — credential block shared by AWS Bedrock (Think) and AWS Polly (Speak).
• audio — AudioConfig, AudioInput, AudioOutput, encodings, and container for the audio block on Settings.
• listen — AgentListenSettings and the V1/V2 Deepgram STT provider sub-types.
• think — ThinkSettings and the five Think provider variants (OpenAI, Anthropic, AWS Bedrock, Google, Groq).
• speak — SpeakSettings and the five Speak provider variants (Deepgram, ElevenLabs, Cartesia, OpenAI, AWS Polly).
• history — HistoryMessage (conversation + function-call history) used by agent.context.messages[] and the server-emitted History event.
• settings — top-level SettingsMessage and the AgentConfig oneOf (Inline(InlineAgentConfig) vs. Saved(Uuid)).
• messages — the remaining client-to-server messages (UpdateListen, UpdateSpeak, UpdateThink, UpdatePrompt, InjectUserMessage, InjectAgentMessage, FunctionCallResponse, KeepAlive, ForceEndTurn) plus the ClientMessage discriminated union over all ten.
• response — every server-emitted JSON event (Welcome, SettingsApplied, ConversationText, UserStartedSpeaking, AgentThinking, FunctionCallRequest, FunctionCallCancelled, AgentStartedSpeaking, AgentAudioDone, Error, Warning, History, LatencyReport, ListenUpdated, PromptUpdated, SpeakUpdated, ThinkUpdated, InjectionRefused, FunctionCallResponse) plus an Unknown catch-all for forward compatibility, all unified under the response::AgentResponse enum. Dispatch is on the event’s type, so AgentResponse::Unknown holds exactly the events this SDK does not model; a recognized event with a malformed payload is an error on the stream, never a silent Unknown. A recognized event carrying an unrecognized value is neither: it stays on the happy path through that value’s own catch-all. A role this release does not name, on ConversationText or History, arrives as ConversationRole::Unknown(String) holding the wire string verbatim, and a history entry matching neither modeled shape as HistoryMessage::Unknown(serde_json::Value). So a new server enum value never ends a session.
• websocket — the Agent sub-client and live-session primitives (AgentHandle, AgentEventStream, AgentEvent) that connect to wss://agent.deepgram.com/v1/agent/converse. start_at_url requires TLS (wss://) except for loopback hosts.

Debug output for AwsCredentials, Endpoint, and FunctionEndpoint redacts secrets (AWS keys/tokens and header values), so a Settings value can be logged safely.

Wire format matches the AsyncAPI schemas in deepgram-docs under api/specs/asyncapi/schemas/agent/. See the simple_agent and function_calling examples for end-to-end usage.

Re-exports

pub use audio::AudioConfig;

pub use audio::AudioContainer;

pub use audio::AudioInput;

pub use audio::AudioInputEncoding;

pub use audio::AudioOutput;

pub use audio::AudioOutputEncoding;

pub use aws_credentials::AwsCredentials;

pub use aws_credentials::AwsCredentialsType;

pub use endpoint::Endpoint;

pub use history::ConversationHistoryMessage;

pub use history::ConversationRole;

pub use history::FunctionCallHistoryMessage;

pub use history::HistoryFunctionCall;

pub use history::HistoryMessage;

pub use history::HistoryMessageType;

pub use listen::AgentListenProvider;

pub use listen::AgentListenSettings;

pub use listen::DeepgramListenV1Provider;

pub use listen::DeepgramListenV1Version;

pub use listen::DeepgramListenV2Provider;

pub use listen::DeepgramListenV2Version;

pub use listen::DeepgramProviderType;

pub use messages::ClientMessage;

pub use messages::ForceEndTurnMessage;

pub use messages::ForceEndTurnType;

pub use messages::FunctionCallResponseMessage;

pub use messages::FunctionCallResponseType;

pub use messages::InjectAgentBehavior;

pub use messages::InjectAgentMessageMessage;

pub use messages::InjectAgentMessageType;

pub use messages::InjectUserMessageMessage;

pub use messages::InjectUserMessageType;

pub use messages::KeepAliveMessage;

pub use messages::KeepAliveType;

pub use messages::UpdateListenMessage;

pub use messages::UpdateListenType;

pub use messages::UpdatePromptMessage;

pub use messages::UpdatePromptType;

pub use messages::UpdateSpeakMessage;

pub use messages::UpdateSpeakType;

pub use messages::UpdateThinkMessage;

pub use messages::UpdateThinkType;

pub use response::AgentAudioDoneEvent;

pub use response::AgentAudioDoneType;

pub use response::AgentFunctionCall;

pub use response::AgentResponse;

pub use response::AgentStartedSpeakingEvent;

pub use response::AgentStartedSpeakingType;

pub use response::AgentThinkingEvent;

pub use response::AgentThinkingType;

pub use response::CancelledFunctionCall;

pub use response::ConversationTextEvent;

pub use response::ConversationTextType;

pub use response::ErrorEvent;

pub use response::ErrorType;

pub use response::FunctionCallCancelledEvent;

pub use response::FunctionCallCancelledType;

pub use response::FunctionCallRequestEvent;

pub use response::FunctionCallRequestType;

pub use response::InjectionRefusedEvent;

pub use response::InjectionRefusedType;

pub use response::LatencyReportEvent;

pub use response::LatencyReportType;

pub use response::ListenUpdatedEvent;

pub use response::ListenUpdatedType;

pub use response::PromptUpdatedEvent;

pub use response::PromptUpdatedType;

pub use response::SettingsAppliedEvent;

pub use response::SettingsAppliedType;

pub use response::SpeakUpdatedEvent;

pub use response::SpeakUpdatedType;

pub use response::ThinkUpdatedEvent;

pub use response::ThinkUpdatedType;

pub use response::UserStartedSpeakingEvent;

pub use response::UserStartedSpeakingType;

pub use response::WarningEvent;

pub use response::WarningType;

pub use response::WelcomeEvent;

pub use response::WelcomeType;

pub use settings::AgentConfig;

pub use settings::AgentContext;

pub use settings::InlineAgentConfig;

pub use settings::SettingsFlags;

pub use settings::SettingsMessage;

pub use settings::SettingsMessageType;

pub use speak::SpeakProvider;

pub use speak::SpeakSettings;

pub use think::ContextLength;

pub use think::FunctionEndpoint;

pub use think::ThinkFunction;

pub use think::ThinkProvider;

pub use think::ThinkSettings;

pub use websocket::Agent;

pub use websocket::AgentEvent;

pub use websocket::AgentEventStream;

pub use websocket::AgentHandle;

pub use websocket::InsecureAgentUrl;

Modules

audio

Audio I/O configuration for the Voice Agent Settings message.

aws_credentials

AWS credentials shared by AWS Bedrock (Think) and AWS Polly (Speak) providers.

endpoint

Custom endpoint configuration for non-Deepgram Think and Speak providers.

history

Conversation history messages used by agent.context.messages[] on the Voice Agent Settings message and emitted by the server as History events during a session.

listen

Agent Listen (speech-to-text) provider settings.

messages

Client-to-server message types for the Voice Agent WebSocket.

response

Server-to-client event types for the Voice Agent WebSocket.

settings

Voice Agent Settings message — the full client-to-server payload that configures an agent session.

speak

Voice Agent Speak (TTS) settings.

think

Voice Agent Think (LLM) settings.

websocket

Voice Agent WebSocket connection — the live session at wss://agent.deepgram.com/v1/agent/converse.