Class AgentSession<UserData>

constructor

• new AgentSession<UserData>(opts): AgentSession<UserData>

• Type Parameters

  • UserData = unknown

  Parameters

  • opts: AgentSessionOptions<UserData>

  Returns AgentSession<UserData>

Optional llm

Readonly options

Optional stt

Optional tts

Optional turnDetection

Optional vad

agentState

• get agentState(): AgentState

• Returns AgentState

chatCtx

• get chatCtx(): ChatContext

• Returns ChatContext

connOptions

• get connOptions(): ResolvedSessionConnectOptions

• Connection options for STT, LLM, and TTS.

  Returns ResolvedSessionConnectOptions

currentAgent

• get currentAgent(): voice.Agent<any>

• Returns voice.Agent<any>

history

• get history(): ChatContext

• Returns ChatContext

input

• get input(): AgentInput

• Returns AgentInput

output

• get output(): AgentOutput

• Returns AgentOutput

userData

• get userData(): UserData

• Returns UserData

• set userData(value): void

• Parameters

  • value: UserData

  Returns void

clearUserTurn

• clearUserTurn(): void

• Returns void

close

• close(): Promise<void>

• Returns Promise<void>

commitUserTurn

• commitUserTurn(): void

• Returns void

emit

• emit<K>(event, ...args): boolean

• Type Parameters

  • K extends keyof AgentSessionCallbacks

  Parameters

  • event: K
  • Rest ...args: Parameters<AgentSessionCallbacks[K]>

  Returns boolean

generateReply

• generateReply(options?): SpeechHandle

• Parameters

  • Optional options: {
        allowInterruptions?: boolean;
        instructions?: string;
        toolChoice?: llm.ToolChoice;
        userInput?: string;
    }

    • Optional allowInterruptions?: boolean

    • Optional instructions?: string

    • Optional toolChoice?: llm.ToolChoice

    • Optional userInput?: string

  Returns SpeechHandle

interrupt

• interrupt(): Future<void>

• Returns Future<void>

run

• run(options): RunResult<unknown>

• Run a test with user input and return a result for assertions.

  This method is primarily used for testing agent behavior without requiring a real room connection.

  Parameters

  • options: {
        userInput: string;
    }

    Run options including user input

    • userInput: string

  Returns RunResult<unknown>

  A RunResult that resolves when the agent finishes responding

  TODO: Add outputType parameter for typed outputs (parity with Python)

  Example

  const result = await session.run({ userInput: 'Hello' });
  result.expect.nextEvent().isMessage({ role: 'assistant' });
  result.expect.noMoreEvents();
  Copy

say

• say(text, options?): SpeechHandle

• Parameters

  • text: string | ReadableStream<string>
  • Optional options: {
        addToChatCtx?: boolean;
        allowInterruptions?: boolean;
        audio?: ReadableStream<AudioFrame>;
    }

    • Optional addToChatCtx?: boolean

    • Optional allowInterruptions?: boolean

    • Optional audio?: ReadableStream<AudioFrame>

  Returns SpeechHandle

shutdown

• shutdown(options?): void

• Parameters

  • Optional options: {
        drain?: boolean;
        reason?: string;
    }

    • Optional drain?: boolean

    • Optional reason?: string

  Returns void

start

• start(__namedParameters): Promise<void>

• Parameters

  • __namedParameters: {
        agent: voice.Agent<any>;
        inputOptions?: Partial<RoomInputOptions>;
        outputOptions?: Partial<RoomOutputOptions>;
        record?: boolean;
        room?: Room;
    }

    • agent: voice.Agent<any>

    • Optional inputOptions?: Partial<RoomInputOptions>

    • Optional outputOptions?: Partial<RoomOutputOptions>

    • Optional record?: boolean

    • Optional room?: Room

  Returns Promise<void>

updateAgent

• updateAgent(agent): void

• Parameters

  • agent: voice.Agent<any>

  Returns void