The ITool & IToolProvider Interfaces

At the heart of AgentB's ability to perform actions and interact with external systems are Tools. This guide explains the fundamental interfaces: ITool for defining individual tools and IToolProvider for supplying them to agents.

The ITool Interface: Defining a Capability

An ITool represents a single, distinct capability that an agent can use. To create a custom tool, you implement this interface.

Core Methods of ITool:

  1. getDefinition(): Promise<IToolDefinition> | IToolDefinition

    • Purpose: This method is crucial. It returns an IToolDefinition object that describes your tool to the Large Language Model (LLM). The LLM uses this information to understand what the tool does and how to call it.

    • IToolDefinition Structure:

      • name: string: The unique, programmatic name of the tool (e.g., searchFlights, getWeatherForecast). Must be LLM-friendly (e.g., alphanumeric, underscores, hyphens, typically max 64 chars).

      • description: string: A clear, natural language description of what the tool does, its purpose, and when it should be used. This is very important for the LLM's decision-making process.

      • parameters: IToolParameter[]: An array describing the input parameters the tool expects.

        • IToolParameter Structure:

          • name: string: Name of the parameter (e.g., destinationCity, date).

          • type: string: The basic data type (e.g., string, number, boolean, object, array).

          • description: string: Natural language description of this parameter for the LLM.

          • required: boolean: Whether this parameter is mandatory.

          • schema?: any: (Highly Recommended) A JSON Schema object defining the structure, constraints (enums, min/max, pattern), and detailed type for this parameter. This gives the LLM precise instructions for formatting arguments.

  2. execute(input: Input, agentContext?: IAgentContext): Promise<IToolResult>

    • Purpose: This method contains the actual logic that your tool performs.

    • input: Input: An object containing the arguments for the tool. The LLM constructs this object based on your IToolDefinition's parameters. The shape of Input should match the properties defined in your parameters. For example, if you define parameters city and date, input would be { city: "Paris", date: "2024-12-25" }.

    • agentContext?: IAgentContext (Optional): If your tool needs access to broader runtime information (like runConfig, threadId, or other services available during the agent's run), the IAgentContext can be passed here.

    • Promise<IToolResult>: The method must return a Promise that resolves to an IToolResult object.

      • IToolResult Structure:

        • success: boolean: true if the tool's operation was successful, false otherwise.

        • data: any: The data payload resulting from the tool's successful execution. This is what gets passed back to the LLM.

        • error?: string: A descriptive error message if success is false.

        • metadata?: Record<string, any>: Optional additional information about the tool's execution (e.g., API call duration, units of data processed).

Example (Conceptual ITool):

import { ITool, IToolDefinition, IToolParameter, IToolResult, IAgentContext } from '@ulifeai/agentb';

class MySimpleTool implements ITool<{ query: string }, string> { // Input type, Output data type
  async getDefinition(): Promise<IToolDefinition> {
    return {
      name: "myCustomSearch",
      description: "Performs a custom search using an internal engine.",
      parameters: [
        {
          name: "query",
          type: "string",
          description: "The search term or question.",
          required: true,
          schema: { type: "string", minLength: 3 }
        }
      ]
    };
  }

  async execute(input: { query: string }, agentContext?: IAgentContext): Promise<IToolResult<string>> {
    console.log(`[MySimpleTool] Executing with query: ${input.query}`);
    // ... actual search logic here ...
    if (input.query === "magic") {
      return { success: true, data: "Found the magic keyword!" };
    }
    if (input.query === "fail") {
      return { success: false, data: null, error: "Simulated search failure." };
    }
    return { success: true, data: `Search results for "${input.query}".` };
  }
}

(See the Custom Tool Tutorial for a runnable example).

The IToolProvider Interface: Supplying Tools

An agent doesn't directly manage a list of ITool instances. Instead, it receives an IToolProvider via its IAgentContext. The IToolProvider is responsible for making tools available.

Core Methods of IToolProvider:

  1. getTools(): Promise<ITool[]>:

    • Returns a Promise that resolves to an array of all ITool instances this provider can supply.

  2. getTool(toolName: string): Promise<ITool | undefined>:

    • Returns a Promise that resolves to a specific ITool instance by its name, or undefined if not found.

  3. ensureInitialized?(): Promise<void> (Optional):

    • Some providers might need an asynchronous initialization step (e.g., an OpenAPIConnector needs to fetch and parse its spec). This method allows the system (like ApiInteractionManager) to ensure the provider is ready before attempting to get tools from it.

Why IToolProvider?

  • Dynamic Tool Loading: Providers can dynamically load or generate tools (e.g., OpenAPIConnector creates tools based on an API spec at runtime).

  • Abstraction: Agents interact with the IToolProvider interface, not concrete tool sources.

  • Composition: Providers like AggregatedToolProvider can combine tools from multiple other providers.

  • Specialization: You can have providers specialized for certain types of tools (e.g., database tools, file system tools).

Common Implementations in AgentB:

  • OpenAPIConnector: Implements IToolProvider to serve tools derived from an OpenAPI specification.

  • AggregatedToolProvider: Implements IToolProvider by taking an array of other IToolProviders and presenting their tools as a single collection (handles de-duplication by name).

  • Custom Providers: You can implement IToolProvider to supply your custom tools, as shown in the Custom Tool Tutorial.

By implementing ITool and making them available via an IToolProvider, you give your AgentB agents the capabilities they need to perform meaningful actions and interact with the world. The quality of your IToolDefinitions (especially the descriptions) directly impacts how well the LLM can understand and utilize your tools.

Previous04-toolsNextOpenAPIConnector

Last updated