Member Junction
    Preparing search index...

    Interface IAgentSettings

    App-scoped agent configuration, stored as JSON in the AgentSettings column of MJ: Applications. Every field is optional — an app opts into exactly what it needs.

    interface IAgentSettings {
        ClientTools?: {
            ClientToolDefinitionID?: string;
            Name?: string;
            Priority?: number;
        }[];
        DefaultAgentID?: string;
        Realtime?: {
            Channels?: {
                Config?: { [channelKey: string]: { [key: string]: unknown } };
                DisplayPolicy?: {
                    [channelKey: string]: "open-on-start" | "on-demand" | "headless";
                };
                Exclude?: string[];
                Include?: string[];
                RequireZeroDataRetentionFor?: ("state" | "pixels")[];
            };
            Disclosure?: "silent"
            | "mention"
            | "hand-voice";
            ModelPreference?: string;
            Persona?: { SpeakingStyle?: string; Tone?: string };
        };
        RelevantAgents?: {
            Advertised?: boolean;
            AgentID: string;
            Disclosure?: "silent"
            | "mention"
            | "hand-voice";
            Label?: string;
        }[];
    }
    Index

    Properties

    ClientTools? DefaultAgentID? Realtime? RelevantAgents?

    Properties

    ClientTools?: {
        ClientToolDefinitionID?: string;
        Name?: string;
        Priority?: number;
    }[]

    App-scoped static client tools, by tool-definition reference. Surfaced to every agent acting in this app and resolved by the unified client-tool resolver.

    Type Declaration

    • OptionalClientToolDefinitionID?: string

      References MJ: AI Client Tool Definitions by ID (preferred) or Name.

    • OptionalName?: string
    • OptionalPriority?: number

      Optional app-level priority for first-match-wins resolution.

    DefaultAgentID?: string

    The app's default/lead agent (conversational default AND realtime lead identity). Agent ID.

    Realtime?: {
        Channels?: {
            Config?: { [channelKey: string]: { [key: string]: unknown } };
            DisplayPolicy?: {
                [channelKey: string]: "open-on-start" | "on-demand" | "headless";
            };
            Exclude?: string[];
            Include?: string[];
            RequireZeroDataRetentionFor?: ("state" | "pixels")[];
        };
        Disclosure?: "silent"
        | "mention"
        | "hand-voice";
        ModelPreference?: string;
        Persona?: { SpeakingStyle?: string; Tone?: string };
    }

    Realtime co-agent overrides that layer into the config cascade above the agent's own TypeConfiguration (and below runtime overrides). Only keys set here override; unset keys fall through to the agent / type defaults.

    Type Declaration

    • OptionalChannels?: {
          Config?: { [channelKey: string]: { [key: string]: unknown } };
          DisplayPolicy?: {
              [channelKey: string]: "open-on-start" | "on-demand" | "headless";
          };
          Exclude?: string[];
          Include?: string[];
          RequireZeroDataRetentionFor?: ("state" | "pixels")[];
      }

      Which interactive realtime channels (Whiteboard, Remote Browser, an Open App's own widget, …) this app's sessions get. Layers into the same per-channel scoping cascade as the agent's own realtime.channels configuration (see @memberjunction/ai-core-plus ResolveRealtimeChannelScope): a layer's Include turns a channel on (the only way to get an opt-in channel), its Exclude turns one off, and the MOST SPECIFIC layer to mention a channel wins; within one layer Exclude beats Include. A channel's registry row IsActive = false remains a master kill switch that nothing here can override. Channel keys match the channel's name, case-insensitively.

      • OptionalConfig?: { [channelKey: string]: { [key: string]: unknown } }

        Per-channel opaque configuration, keyed by channel key; delivered to the channel at session start.

      • OptionalDisplayPolicy?: { [channelKey: string]: "open-on-start" | "on-demand" | "headless" }

        Per-channel display override, keyed by channel key.

      • OptionalExclude?: string[]

        Channel keys to turn OFF. Beats Include within the same layer.

      • OptionalInclude?: string[]

        Channel keys to turn ON (required for opt-in channels; a no-op for all-sessions ones).

      • OptionalRequireZeroDataRetentionFor?: ("state" | "pixels")[]

        Exposure levels that need a zero-data-retention model. When the session model's configuration does not declare Privacy.ZeroDataRetention: true, exposure is lowered to below the lowest level listed, and the agent is told why. Applies to every channel in the session.

    • OptionalDisclosure?: "silent" | "mention" | "hand-voice"

      Default delegation disclosure for this app's co-agent.

    • OptionalModelPreference?: string

      Model preference override (AI Models Name or ID).

    • OptionalPersona?: { SpeakingStyle?: string; Tone?: string }

      Persona override folded into the session system prompt at mint.

    RelevantAgents?: {
        Advertised?: boolean;
        AgentID: string;
        Disclosure?: "silent" | "mention" | "hand-voice";
        Label?: string;
    }[]

    Agents relevant to this app — the static allowed-target set for the co-agent. Union-accumulated at runtime with agent-type defaults and dynamic (channel) registrations.

    Type Declaration

    • OptionalAdvertised?: boolean

      If true, surfaced in pickers / proactively offered; if false, available but not advertised.

    • AgentID: string

      Agent ID (loop or flow — transparent to the co-agent).

    • OptionalDisclosure?: "silent" | "mention" | "hand-voice"

      Per-target disclosure override; falls back to the effective default disclosure.

    • OptionalLabel?: string

      Optional friendly label for the capability manifest / disclosure ("Skip", "Query Builder").

    Settings

    Member Visibility

    On This Page

    Properties