Member Junction
    Preparing search index...

    Interface RealtimeSessionStartContext

    Context passed to the realtime-session factory when starting an agent room session.

    interface RealtimeSessionStartContext {
        AgentID?: string;
        AgentName?: string;
        AgentSessionID?: string;
        AvatarDelivery?: "room";
        ContextUser?: UserInfo;
        ConversationID?: string;
        HostFraming?: string;
        HostTools?: RealtimeToolDefinition[];
        MeetingMode?: boolean;
        MetadataProvider?: IMetadataProvider;
        PhoneCall?: boolean;
        PriorTranscript?: string;
        RealtimeModelID?: string;
        RealtimeVoice?: string;
        ResolveHostTools?: (
            resolved: {
                DriverClass?: string;
                ModelID?: string;
                ModelVendorID?: string;
            },
        ) => RealtimeToolDefinition[]
        | undefined;
        RoomName: string;
        SelfNames?: string[];
        TargetAgentID?: string;
    }
    Index

    Properties

    AgentID? AgentName? AgentSessionID? AvatarDelivery? ContextUser? ConversationID? HostFraming? HostTools? MeetingMode? MetadataProvider? PhoneCall? PriorTranscript? RealtimeModelID? RealtimeVoice? ResolveHostTools? RoomName SelfNames? TargetAgentID?

    Properties

    AgentID?: string

    The agent to voice in the room, when known.

    AgentName?: string

    The agent's display name (used for the bot name + turn-taking matcher).

    AgentSessionID?: string

    The MJ agent-session id — threaded into the co-agent observability run for session grouping.

    AvatarDelivery?: "room"

    'room' when this host's bot can publish the agent's avatar into the room (the native module answered describeAvatarVideo with support): the session may then ask the model to render it. Absent: audio only, as before.

    ContextUser?: UserInfo

    The user the session runs as.

    ConversationID?: string

    The MJ: Conversations row the session writes to — stamped on the co-agent observability run.

    HostFraming?: string

    Host-authored instructions appended to the system prompt (for example the phone framing).

    HostTools?: RealtimeToolDefinition[]

    Tools the host declares and executes itself (call control, handoff). Added to the model's tool set.

    MeetingMode?: boolean

    Multi-agent meeting mode — set when the agent joins a room that already has agents. Flows to the realtime session so its model's auto-response is disabled (speaks only when addressed).

    MetadataProvider?: IMetadataProvider

    The metadata provider for the session.

    PhoneCall?: boolean

    true when the agent is answering a phone call in the room (AgentRoomHostOptions.Channel 'phone'): the caller sees no video, so the session asks the model for no avatar and reports phone. Never set with AvatarDelivery.

    PriorTranscript?: string

    Role-tagged transcript so far, framed into the prompt when a lost model session is re-opened mid-call.

    RealtimeModelID?: string

    Optional per-session Realtime MODEL override (Name or ID).

    RealtimeVoice?: string

    Optional per-session VOICE override (provider-native voice id).

    ResolveHostTools?: (
        resolved: {
            DriverClass?: string;
            ModelID?: string;
            ModelVendorID?: string;
        },
    ) => RealtimeToolDefinition[]
    | undefined

    Optional callback allowing the coordinator to resolve host tools dynamically based on the model and driver class context actually resolved for the session, before session opening.

    RoomName: string

    The room being joined.

    SelfNames?: string[]

    The names the agent answers to (display name + aliases) — phrasing for the meeting prompt.

    TargetAgentID?: string

    The TARGET agent the co-agent voices via invoke-target-agent (the one being "called").

    Settings

    Member Visibility

    On This Page

    Properties