Installation

To install the package, run:

npm install @virtuals-protocol/game

Overview

Currently, this SDK allows you to develop your agents powered by the GAME architecture in its most fullest and most flexible form.

The SDK is made up of 3 main components (Agent, Worker, function), each with configurable arguments.

Agent (a.k.a. high level planner)

• Takes in a Goal
  • Drives the agent's behavior through the high-level plan which influences the thinking and creation of tasks that would contribute towards this goal
• Takes in a Description
  • Combination of what was previously known as World Info + Agent Description
  • This includes a description of the "world" the agent lives in, and the personality and background of the agent

Worker (a.k.a. low-level planner)

• Takes in a Description
  • Used to control which workers are called by the agent, based on the high-level plan and tasks created to contribute to the goal

Function

• Takes in a Description
  • Used to control which functions are called by the workers, based on each worker's low-level plan
  • This can be any executable

Core Features

1. Functions and Executables

Functions define available actions for the agent:

import {
  GameFunction,
  ExecutableGameFunctionResponse,
  ExecutableGameFunctionStatus,
} from "@virtuals-protocol/game";

const myFunction = new GameFunction({
  name: "action_name",
  description: "Description of action",
  args: [
    { name: "param", type: "type", description: "param description" },
  ] as const,
  executable: async (args) => {
    try {
      // Implement your function logic here
      return new ExecutableGameFunctionResponse(
        ExecutableGameFunctionStatus.Done,
        "Action completed successfully"
      );
    } catch (e) {
      return new ExecutableGameFunctionResponse(
        ExecutableGameFunctionStatus.Failed,
        "Action failed"
      );
    }
  },
});

Executable functions must return an instance of ExecutableGameFunctionResponse with:

• ExecutableGameFunctionStatus
• Feedback message

2. State Management

Easy and flexible way to define the state management, what the agent sees and how that changes.

async function getAgentState(): Promise<Record<string, any>> {
  return {
    health: 100,
    inventory: ["sword", "shield", "potion"],
  };
}

3. Workers

Workers are simple interactable agents that execute the tasks defined by the user. They can be specialized agents with defined capabilities:

import { GameWorker } from "@virtuals-protocol/game";

const worker = new GameWorker({
  id: "worker_id",
  name: "Worker Name",
  description: "Worker description",
  functions: [list_of_functions],
  getEnvironment: async () => {
    return {
      // environment details
    };
  },
});

Key features:

• Can be shared or unique per worker
• Processes function execution results to update state

4. Agents

Agents are used to autonomously function in an open-ended manner by just providing a general goal. Tasks are generated by the agent itself continuously, and the agent will attempt to complete them. You can provide many workers to the agent, and they will be used to execute the tasks.

import { GameAgent } from "@virtuals-protocol/game";

const agent = new GameAgent("your_api_key", {
  name: "Agent Name",
  goal: "Primary goal",
  description: "Description",
  getAgentState: agent_state_function,
  workers: [worker1, worker2],
});

// Compile and run
await agent.init();
await agent.run();

4.1. Running worker (Reactive)

To run a worker individually, you can initialize and run the worker without the main loop. Here's an example:

const agentWorker = agent.getWorkerById(worker1.id);

const task =
  "Gotten a message from user. Message content: hey! I will need help with my project, I need an image of a cat-hugging AI. Can you help me with that? Give me something that cool and cute!";

await agentWorker.runTask(task);

This allows you to run specific workers of an agent.

Whitepaper

For more detailed information, please refer to the Virtuals Game Framework Whitepaper.

Example Usage (Twitter)

Setting Up the Agent

1. Create the Twitter Functions

   Define the functions for posting tweets, searching tweets, and replying to tweets:

   import {
     ExecutableGameFunctionResponse,
     ExecutableGameFunctionStatus,
     GameFunction,
   } from "@virtuals-protocol/game";
   
   const postTweetFunction = new GameFunction({
     name: "post_tweet",
     description: "Post a tweet",
     args: [
       { name: "tweet", description: "The tweet content" },
       {
         name: "tweet_reasoning",
         description: "The reasoning behind the tweet",
       },
     ] as const,
     executable: async (args, logger) => {
       try {
         // TODO: Implement posting tweet
         logger(`Posting tweet: ${args.tweet}`);
         logger(`Reasoning: ${args.tweet_reasoning}`);
   
         return new ExecutableGameFunctionResponse(
           ExecutableGameFunctionStatus.Done,
           "Tweet posted"
         );
       } catch (e) {
         return new ExecutableGameFunctionResponse(
           ExecutableGameFunctionStatus.Failed,
           "Failed to post tweet"
         );
       }
     },
   });
   
   const searchTweetsFunction = new GameFunction({
     name: "search_tweets",
     description: "Search tweets and return results",
     args: [
       { name: "query", description: "The query to search for" },
       { name: "reasoning", description: "The reasoning behind the search" },
     ] as const,
     executable: async (args, logger) => {
       try {
         const query = args.query;
         // TODO: Implement searching of tweets based on query string
         logger(`Searching tweets for query: ${query}`);
   
         return new ExecutableGameFunctionResponse(
           ExecutableGameFunctionStatus.Done,
           "Tweets searched here are the results: [{tweetId: 1, content: 'Hello World'}, {tweetId: 2, content: 'Goodbye World'}]"
         );
       } catch (e) {
         return new ExecutableGameFunctionResponse(
           ExecutableGameFunctionStatus.Failed,
           "Failed to search tweets"
         );
       }
     },
   });
   
   const replyToTweetFunction = new GameFunction({
     name: "reply_to_tweet",
     description: "Reply to a tweet",
     args: [
       { name: "tweet_id", description: "The tweet id to reply to" },
       { name: "reply", description: "The reply content" },
     ] as const,
     executable: async (args, logger) => {
       try {
         const tweetId = args.tweet_id;
         const reply = args.reply;
   
         // TODO: Implement reply tweet
         logger(`Replying to tweet ${tweetId}`);
         logger(`Replying with ${reply}`);
   
         return new ExecutableGameFunctionResponse(
           ExecutableGameFunctionStatus.Done,
           `Replied to tweet ${tweetId} with ${reply}`
         );
       } catch (e) {
         return new ExecutableGameFunctionResponse(
           ExecutableGameFunctionStatus.Failed,
           "Failed to reply to tweet"
         );
       }
     },
   });

2. Create the Worker

   Define a worker that uses the functions in twitter.ts:

   import { GameWorker } from "@virtuals-protocol/game";
   
   const postTweetWorker = new GameWorker({
     id: "twitter_main_worker",
     name: "Twitter main worker",
     description: "Worker that posts tweets",
     functions: [searchTweetsFunction, replyToTweetFunction, postTweetFunction],
     // Optional: Provide environment to LLP
     getEnvironment: async () => {
       return {
         tweet_limit: 15,
       };
     },
   });

3. Create the Agent

   Define an agent that uses the worker in agent.ts:

   import { GameAgent } from "@virtuals-protocol/game";
   
   const agent = new GameAgent("YOUR_API_KEY", {
     name: "Twitter Bot",
     goal: "Search and reply to tweets",
     description: "A bot that searches for tweets and replies to them",
     workers: [postTweetWorker],
     // Optional: Provide state to HLP
     getAgentState: async () => {
       return {
         username: "twitter_bot",
         follower_count: 1000,
         tweet_count: 10,
       };
     },
   });

Defining a custom logger

To define a custom logger for the agent, you can use the setLogger method. This method allows you to specify how the agent's log messages should be handled. Here's an example of how to set up a custom logger:

agent.setLogger(agent, (msg) => {
  console.log(`-----[${agent.name}]-----`);
  console.log(msg);
  console.log("\n");
});

In this example, the custom logger will print the agent's name followed by the log message to the console. You can customize the logger function to handle log messages in any way you prefer, such as writing them to a file or sending them to a logging service.

Using the Logger in Custom Functions

You can use the logger within your custom functions to log messages. The logger is passed as an argument to the executable function. Here's an example of how to use the logger in a custom function:

const customFunction = new GameFunction({
  name: "custom_action",
  description: "A custom action with logging",
  args: [{ name: "param", description: "Parameter for the action" }] as const,
  executable: async (args, logger) => {
    try {
      logger(`Executing custom action with param: ${args.param}`);
      // Implement your function logic here
      return new ExecutableGameFunctionResponse(
        ExecutableGameFunctionStatus.Done,
        "Custom action completed successfully"
      );
    } catch (e) {
      logger(`Failed to execute custom action: ${e.message}`);
      return new ExecutableGameFunctionResponse(
        ExecutableGameFunctionStatus.Failed,
        "Custom action failed"
      );
    }
  },
});

In this example, the logger is used to log messages before and after the execution of the custom action. This helps in tracking the function's execution flow and any errors that occur.

Running the Agent

The agent will initialize and start running, performing actions such as posting tweets, searching for tweets, and replying to tweets at regular intervals.

await agent.init();
// running at a fix interval of 60 seconds
await agent.run(60, {
  /**
   * @property {boolean} verbose - A flag to enable or disable verbose logging.
   *
   * @description
   * The `verbose` property is used to control the verbosity of the logging output.
   * When set to `true`, detailed logs will be generated, which can be useful for
   * debugging and development purposes. When set to `false`, only essential logs
   * will be produced, reducing the amount of log output.
   */
  verbose: true | false,
});

Running Agent (without fix interval)

With the step function app has more control over in interval

await agent.step();

Example Usage (Telegram)

This example provides a use case where agents are reacting to a task.

Running the Agent (Telegram)

The agent will initialize and start running, performing actions such as generating images and replying to messages.

Defining the Worker (Telegram)

Define a worker that uses the functions in tg.ts:

import { GameWorker } from "@virtuals-protocol/game";

const telegramWorker = new GameWorker({
  id: "telegram",
  name: "telegram",
  description: "Telegram worker",
  functions: [.........],
});

await agent.init();

const agentTgWorker = agent.getWorkerById(telegramWorker.id);

const task =
  "Gotten a message from user. Message content: hey! I will need help with my project, I need an image of a cat-hugging AI. Can you help me with that? Give me something that cool and cute!";

await agentTgWorker.runTask(task, {
  /**
   * @property {boolean} verbose - A flag to enable or disable verbose logging.
   *
   * @description
   * The `verbose` property is used to control the verbosity of the logging output.
   * When set to `true`, detailed logs will be generated, which can be useful for
   * debugging and development purposes. When set to `false`, only essential logs
   * will be produced, reducing the amount of log output.
   */
  verbose: true | false,
});

License

This project is licensed under the MIT License.