AI class

Creates a new AI system.

new AI(options: AIOptions<TState>)

Parameters

An action that is called to DO an action.

static DoCommandActionName: "___DO___" = "___DO___"

Property Value

Remarks

The action system is used to do other actions. Overriding this action lets you customize the execution of an individual action. You can use it to log actions being used or to prevent certain actions from being executed based on policy.

The default behavior is to simply execute the action handler passed in so you will need to perform that logic yourself should you override this action.

An action that will be called anytime an input is flagged by the moderator.

static FlaggedInputActionName: "___FlaggedInput___" = "___FlaggedInput___"

Property Value

Remarks

The default behavior is to simply log an error to the console. Override to send a custom message to the user.

An action that will be called anytime an output is flagged by the moderator.

static FlaggedOutputActionName: "___FlaggedOutput___" = "___FlaggedOutput___"

Property Value

Remarks

The default behavior is to simply log an error to the console. Override to send a custom message to the user.

An action that will be called anytime the planner encounters an HTTP response with status code >= 400.

static HttpErrorActionName: "___HttpError___" = "___HttpError___"

Property Value

Returns the moderator being used by the AI system.

Moderator<TState> moderator

Property Value

The AI's moderator

Planner<TState> planner

Property Value

Returns the planner being used by the AI system.

An action that will be called after the plan has been predicted by the planner and it has passed moderation.

static PlanReadyActionName: "___PlanReady___" = "___PlanReady___"

Property Value

Remarks

Overriding this action lets you customize the decision to execute a plan separately from the moderator. The default behavior is to proceed with the plans execution only with a plan contains one or more commands. Returning false from this action can be used to prevent the plan from being executed.

An action that is called to SAY something.

static SayCommandActionName: "___SAY___" = "___SAY___"

Property Value

Remarks

Overriding this action lets you customize the execution of the SAY command. You can use it to log the output being generated or to add support for sending certain types of output as message attachments.

The default behavior attempts to look for an Adaptive Card in the output and if found sends it as an attachment. If no Adaptive Card is found then the output is sent as a plain text message.

If you override this action and want to automatically send Adaptive Cards as attachments you will need to handle that yourself.

A text string that can be returned from an action to stop the AI system from continuing to execute the current plan.

static StopCommandName: "STOP" = "STOP"

Property Value

Remarks

This command is incompatible and should not be used with tools augmentation

The task either executed too many steps or timed out.

static TooManyStepsActionName: "___TooManySteps___" = "___TooManySteps___"

Property Value

An action that will be called anytime an unknown action is predicted by the planner.

static UnknownActionName: "___UnknownAction___" = "___UnknownAction___"

Property Value

Remarks

The default behavior is to simply log an error to the console. The plan is allowed to continue execution by default.

Registers a handler for a named action.

function action<TParameters>(name: string | string[], handler: ActionHandler<TState, TParameters>): AI<TState>

Parameters

Returns

The AI system instance for chaining purposes.

Remarks

The AI systems planner returns plans that are made up of a series of commands or actions that should be performed. Registering a handler lets you provide code that should be run in response to one of the predicted actions.

Plans support a DO command which specifies the name of an action to call and an optional set of entities that should be passed to the action. The internal plan executor will call the registered handler for the action passing in the current context, state, and entities.

Additionally, the AI system itself uses actions to handle things like unknown actions, flagged input, and flagged output. You can override these actions by registering your own handler for them. The names of the built-in actions are available as static properties on the AI class.

Registers the default handler for a named action.

function defaultAction<TParameters>(name: string | string[], handler: ActionHandler<TState, TParameters>): AI<TState>

Parameters

Returns

The AI system instance for chaining purposes.

Manually executes a named action.

function doAction<TParameters>(context: TurnContext, state: TState, action: string, parameters?: TParameters): Promise<string>

Parameters

Returns

The result of the action.

Checks to see if the AI system has a handler for a given action.

function hasAction(action: string): boolean

Parameters

Returns

True if the AI system has a handler for the given action.

Calls the configured planner to generate a plan and executes the plan that is returned.

function run(context: TurnContext, state: TState, start_time?: number, step_count?: number): Promise<boolean>

Parameters

Returns

True if the plan was completely executed, otherwise false.

Remarks

The moderator is called to review the input and output of the plan. If the moderator flags the input or output then the appropriate action is called. If the moderator allows the input and output then the plan is executed.