Gemini Enterprise for Customer Experience v1beta API - Class SessionService.SessionServiceBase (1.0.0-beta10)

[BindServiceMethod(typeof(SessionService), "BindService")]
public abstract class SessionService.SessionServiceBase

Reference documentation and code samples for the Gemini Enterprise for Customer Experience v1beta API class SessionService.SessionServiceBase.

Base class for server-side implementations of SessionService

Inheritance

object > SessionService.SessionServiceBase

Namespace

Google.Cloud.Ces.V1Beta

Assembly

Google.Cloud.Ces.V1Beta.dll

Methods

BidiRunSession(IAsyncStreamReader<BidiSessionClientMessage>, IServerStreamWriter<BidiSessionServerMessage>, ServerCallContext)

public virtual Task BidiRunSession(IAsyncStreamReader<BidiSessionClientMessage> requestStream, IServerStreamWriter<BidiSessionServerMessage> responseStream, ServerCallContext context)

Establishes a bidirectional streaming connection with the CES agent. The agent processes continuous multimodal inputs (e.g., text, audio) and generates real-time multimodal output streams.

--- Client Request Stream --- The client streams requests in the following order:

  1. Initialization: The first message must contain [SessionConfig][google.cloud.ces.v1beta.BidiSessionClientMessage.config]. For audio sessions, this should also include [InputAudioConfig][google.cloud.ces.v1beta.SessionConfig.input_audio_config] and [OutputAudioConfig][google.cloud.ces.v1beta.SessionConfig.output_audio_config] to define audio processing and synthesis parameters.

  2. Interaction: Subsequent messages stream [SessionInput][google.cloud.ces.v1beta.BidiSessionClientMessage.realtime_input] containing real-time user input data.

  3. Termination: The client should half-close the stream when there is no more user input. It should also half-close upon receiving [EndSession][google.cloud.ces.v1beta.BidiSessionServerMessage.end_session] or [GoAway][google.cloud.ces.v1beta.BidiSessionServerMessage.go_away] from the agent.

--- Server Response Stream --- For each interaction turn, the agent streams messages in the following sequence:

  1. Speech Recognition (First N messages): Contains [RecognitionResult][google.cloud.ces.v1beta.BidiSessionServerMessage.recognition_result] representing the concatenated user speech segments captured so far. This is only populated for audio sessions.

  2. Response (Next M messages): Contains [SessionOutput][google.cloud.ces.v1beta.BidiSessionServerMessage.session_output] delivering the agent's response in various modalities (e.g., text, audio).

  3. Turn Completion (Final message of the turn): Contains [SessionOutput][google.cloud.ces.v1beta.BidiSessionServerMessage.session_output] with [turn_completed][google.cloud.ces.v1beta.SessionOutput.turn_completed] set to true. This signals the end of the current turn and includes [DiagnosticInfo][google.cloud.ces.v1beta.SessionOutput.diagnostic_info] with execution details.

--- Audio Best Practices ---

  1. Streaming: Stream [audio data][google.cloud.ces.v1beta.SessionInput.audio] CONTINUOUSLY, even during silence. Recommended chunk size: 40-120ms (balances latency vs. efficiency).

  2. Playback & Interruption: Play [audio responses][google.cloud.ces.v1beta.SessionOutput.audio] upon receipt. Stop playback immediately if an [InterruptionSignal][google.cloud.ces.v1beta.BidiSessionServerMessage.interruption_signal] is received (e.g., user barge-in or new agent response).

Parameters
Name Description
requestStream IAsyncStreamReaderBidiSessionClientMessage

Used for reading requests from the client.

responseStream IServerStreamWriterBidiSessionServerMessage

Used for sending responses back to the client.

context ServerCallContext

The context of the server-side call handler being invoked.

Returns
Type Description
Task

A task indicating completion of the handler.

CancelAssistantTurn(CancelAssistantTurnRequest, ServerCallContext)

public virtual Task<CancelAssistantTurnResponse> CancelAssistantTurn(CancelAssistantTurnRequest request, ServerCallContext context)

Cancels the assistant session's in-flight turn, if any: the explicit user stop for a turn running detached from any stream. A cancel landing on a task that does not host the turn preempts it through the session turn epoch instead, ending it within one lease renewal.

Parameters
Name Description
request CancelAssistantTurnRequest

The request received from the client.

context ServerCallContext

The context of the server-side call handler being invoked.

Returns
Type Description
TaskCancelAssistantTurnResponse

The response to send back to the client (wrapped by a task).

GenerateOnboardingSuggestions(GenerateOnboardingSuggestionsRequest, ServerCallContext)

public virtual Task<GenerateOnboardingSuggestionsResponse> GenerateOnboardingSuggestions(GenerateOnboardingSuggestionsRequest request, ServerCallContext context)

Generates personalized onboarding suggestions for the Arc zero state: classifies the requesting user (new / exploring / returning) from their assistant-session history in the app and returns suggestion chips (resume a session, continue work, or start something new) to render before any message is sent.

Parameters
Name Description
request GenerateOnboardingSuggestionsRequest

The request received from the client.

context ServerCallContext

The context of the server-side call handler being invoked.

Returns
Type Description
TaskGenerateOnboardingSuggestionsResponse

The response to send back to the client (wrapped by a task).

RunSession(RunSessionRequest, ServerCallContext)

public virtual Task<RunSessionResponse> RunSession(RunSessionRequest request, ServerCallContext context)

Initiates a single-turn interaction with the CES agent within a session.

Parameters
Name Description
request RunSessionRequest

The request received from the client.

context ServerCallContext

The context of the server-side call handler being invoked.

Returns
Type Description
TaskRunSessionResponse

The response to send back to the client (wrapped by a task).

StreamChatAiAssistant(StreamChatAiAssistantRequest, IServerStreamWriter<StreamChatAiAssistantResponse>, ServerCallContext)

public virtual Task StreamChatAiAssistant(StreamChatAiAssistantRequest request, IServerStreamWriter<StreamChatAiAssistantResponse> responseStream, ServerCallContext context)

Runs Arc, the AI assistant, for the specified assistant session in a streaming fashion.

Parameters
Name Description
request StreamChatAiAssistantRequest

The request received from the client.

responseStream IServerStreamWriterStreamChatAiAssistantResponse

Used for sending responses back to the client.

context ServerCallContext

The context of the server-side call handler being invoked.

Returns
Type Description
Task

A task indicating completion of the handler.

StreamRunSession(RunSessionRequest, IServerStreamWriter<RunSessionResponse>, ServerCallContext)

public virtual Task StreamRunSession(RunSessionRequest request, IServerStreamWriter<RunSessionResponse> responseStream, ServerCallContext context)

Initiates a single-turn interaction with the CES agent. Uses server-side streaming to deliver incremental results and partial responses as they are generated.

By default, complete responses (e.g., messages from callbacks or full LLM responses) are sent to the client as soon as they are available. To enable streaming individual text chunks directly from the model, set [enable_text_streaming][google.cloud.ces.v1beta.SessionConfig.enable_text_streaming] to true.

Parameters
Name Description
request RunSessionRequest

The request received from the client.

responseStream IServerStreamWriterRunSessionResponse

Used for sending responses back to the client.

context ServerCallContext

The context of the server-side call handler being invoked.

Returns
Type Description
Task

A task indicating completion of the handler.