Agent SDK

The Agent SDK provides a framework for building AI agents that can interact with various services and APIs. It implements the ReAct (Reasoning, Acting, Observing) prompting technique to enable LLM-powered agents to solve complex tasks through iterative reasoning and action.

Architecture

BaseAgent

The BaseAgent class is the foundation of the agent architecture, providing a CLI interface and core functionality for all agent implementations. It handles common tasks like command registration, logging, and context management.

Key Features:

• CLI interface with standard commands:
  • skills: Lists agent capabilities and available skills
  • about: Provides information about what the agent does
  • terms: Shows domain-specific terminology the agent understands
  • ask: Executes the agent with a user message and streams responses
• Logging system using Pino
• Version tracking from package.json
• Context management for maintaining state between invocations

Abstract Methods:

• ask(): Main method that processes user input and returns responses
• skills(): Returns available skills/capabilities
• terms(): Returns domain-specific terminology
• about(): Returns agent purpose description

ReactBaseAgent

The ReactBaseAgent extends BaseAgent to implement the ReAct (Reasoning, Acting, Observing) prompting technique. This approach enables an LLM-powered agent to reason through multi-step tasks by iteratively:

1. Thinking about how to approach a problem
2. Deciding on an action to take
3. Observing the result
4. Repeating until the task is complete

BaseSkills and APIBaseSkills

The SDK provides base classes for implementing agent capabilities:

• BaseSkills: Abstract class for defining agent skills
• APIBaseSkills: Extension of BaseSkills specifically for API-based integrations

Extending ReactBaseAgent

You can create custom integrations by extending the ReactBaseAgent class. Here's a simplified example:

import { ReactBaseAgent } from "@tegonhq/agent-sdk";
import { GithubSkills } from "./github-skills";
import { JARGON } from "./jargon";

export class GithubAgent extends ReactBaseAgent {
  skills(): Array<any> {
    const githubSkills = new GithubSkills({});
    const skills = githubSkills.skills();
    return Object.keys(skills).map((key) => ({ ...skills[key], name: key }));
  }

  // Provide domain-specific terminology
  terms(): string {
    return JARGON;
  }

  // Implementation of the runSkill method that executes actions
  async runSkill(
    skillName: string,
    parameters: any,
    integrationConfig: Record<string, string>
  ): Promise<string> {
    const githuSkills = new GithubSkills(integrationConfig);
    return githuSkills.runSkill(skillName, parameters);
  }
}

BaseSkills

The BaseSkills class provides an abstraction for implementing skill capabilities that agents can use. It separates skill definitions from the agent implementation.

Key Features:

• Constructor that takes integration configuration
• Abstract methods for defining and executing skills

Abstract Methods:

• skills(): Returns a map of available skills with their definitions
• runSkill(): Executes a specific skill with provided parameters

APIBaseSkills

The APIBaseSkills class extends BaseSkills to provide a foundation for API-based integrations. It handles common API concerns like headers and base URLs.

Key Features:

• Constructor that initializes headers and base URL
• Abstract methods for API configuration

Abstract Methods:

• getBaseURL(): Determines the base URL for API requests
• getHeaders(): Generates headers for API requests using integration configuration