AssistantsPlanner class
A Planner that uses the OpenAI Assistants API to generate plans for the AI system.
Remarks
This planner manages conversations through OpenAI's thread-based system, handling:
- Thread creation and management
- Message submission
- Tool/function calling
- Response processing
Constructors
| Assistants |
Creates a new |
Methods
| begin |
Starts a new task. |
| continue |
Continues the current task. |
| create |
Creates a new assistant using the OpenAI Assistants API. |
Constructor Details
AssistantsPlanner<TState>(AssistantsPlannerOptions)
Creates a new AssistantsPlanner instance.
new AssistantsPlanner(options: AssistantsPlannerOptions)Parameters
- options
- AssistantsPlannerOptions
Options for configuring the AssistantsPlanner.
Method Details
beginTask(TurnContext, TState, AI<TState>)
Starts a new task.
function beginTask(context: TurnContext, state: TState, ai: AI<TState>): Promise<Plan>Parameters
- context
-
TurnContext
Context for the current turn of conversation.
- state
-
TState
Application state for the current turn of conversation.
- ai
-
AI<TState>
The AI system that is generating the plan.
Returns
Promise<Plan>
The plan that was generated.
Remarks
This method is called when the AI system is ready to start a new task. It delegates the
task handling to the continueTask method. The planner should
generate a plan that the AI system will execute. Returning an empty plan signals that
there is no work to be performed.
The planner should take the users input from state.temp.input.
continueTask(TurnContext, TState, AI<TState>)
Continues the current task.
function continueTask(context: TurnContext, state: TState, ai: AI<TState>): Promise<Plan>Parameters
- context
-
TurnContext
Context for the current turn of conversation.
- state
-
TState
Application state for the current turn of conversation.
- ai
-
AI<TState>
The AI system that is generating the plan.
Returns
Promise<Plan>
The plan that was generated.
Remarks
This method is called when the AI system is ready to continue the current task. It handles:
- Creating a new thread if one doesn't exist
- Submitting tool outputs if required
- Waiting for any in-progress runs to complete
- Submitting user input and creating a new run The method generates a plan that the AI system will execute. Returning an empty plan signals that the task is completed and there is no work to be performed.
The output from the last plan step that was executed is passed to the planner via state.temp.input.
createAssistant(string, AssistantCreateParams, string, AzureClientOptions)
Creates a new assistant using the OpenAI Assistants API.
static function createAssistant(apiKey: string, request: AssistantCreateParams, endpoint?: string, azureClientOptions?: AzureClientOptions): Promise<Assistant>Parameters
- apiKey
-
string
OpenAI API key.
- request
-
AssistantCreateParams
Definition of the assistant to create.
- endpoint
-
string
Optional. The Azure OpenAI resource endpoint. Required when using Azure OpenAI.
- azureClientOptions
-
AzureClientOptions
Optional. The Azure OpenAI client options.
Returns
Promise<Assistant>
The created assistant.