Analyzing Project89: A New Modular High-Performance AI Agent Framework Design

Project89 has adopted an innovative approach to designing the Agent framework, which is a high-performance Agent framework for game development, more modular and with better performance compared to current Agent frameworks.

This article will provide a detailed introduction to the architectural upgrades of this framework compared to traditional Agent frameworks, to help readers better understand its innovations.

Developer Background

The founder of Project89 previously participated in the development of the Magick project, which is software that uses AI for programming. He ranked as the fourth developer in the project, demonstrating strong technical skills.

Why Choose ECS Architecture Design for Agent Framework

ECS(Entity-Component-System) architecture is widely used in game development and simulation systems. Currently, games that adopt the ECS architecture include:

• Blockchain games: Mud, Dojo
• Traditional games: Overwatch, Star Citizen, etc.
• Mainstream game engines like Unity are also evolving towards ECS.

The ECS architecture completely separates data from logic, which is beneficial for efficiently managing various entities and their behaviors in large-scale scalable scenarios.

Introduction to ECS Architecture

ECS includes three core concepts:

1. Entity( Entity): is just an ID, it does not contain data or logic.
2. Component(: Used to store specific data or status of entities.
3. System (System ): Responsible for executing logic related to certain components.

In ArgOS, each Agent is regarded as an Entity, capable of registering different components, such as:

• Agent Component: Stores basic information such as Agent name, model name, etc.
• Perception Component: stores perceived external data
• Memory Component: Stores the memory data of the Agent
• Action Component: Stores the Action data to be executed

Example of the System Workflow:

1. Perception of weapons, the Perception System updates the Agent's Perception Component.
2. The Memory System calls the Perception and Memory Component to persist the perception data.
3. The Action System calls the Memory and Action Component to execute the corresponding actions.
4. Obtain the updated Agent Entity

![Deconstructing Project89: A Modular, High-Performance Next-Generation AI Agent Framework Design])https://img-cdn.gateio.im/webp-social/moments-69e16ae3b5af8c39e1fc522358584cef.webp(

ArgOS System Architecture

ArgOS has designed various Components and Systems to enable Agents to perform more complex tasks.

Divide the System into three levels:

1. Conscious ) CONSCIOUS ( system: high update frequency, such as every 10 seconds.
2. 潜意识)SUBCONSCIOUS( system: update frequency is low, such as every 25 seconds
3. Unconscious ) UNCONSCIOUS ( System: The update frequency is the slowest, such as over 50 seconds.

The relationships between different systems are complex, mainly including:

• PerceptionSystem: Collect external stimuli, update Agent's Perception
• ExperienceSystem: Convert stimulation into abstract experiences stored in Memory
• ThinkingSystem: Generates thinking results based on the current state
• ActionSystem: Execute specific actions
• GoalPlanningSystem: Assess goal progress and generate new goals
• PlanningSystem: Generate execution plans for objectives
• RoomSystem: Handles room-related updates
• CleanupSystem: Recycle invalid entities

![Deconstructing Project89: A Modular, High-Performance Next-Generation AI Agent Framework Design])https://img-cdn.gateio.im/webp-social/moments-59f1984837d6636d7efc378c731a01eb.webp(

Analysis of ArgOS Overall Architecture

) Core Architecture Layers

Including layers such as Runtime, Systems, Components, Managers, etc.

![Deconstructing Project89: A Modular, High-Performance Next-Gen AI Agent Framework Design]###https://img-cdn.gateio.im/webp-social/moments-c37bd7442c3743f80a4fd9ba8afc123b.webp(

) Component Classification

1. Core Identity Types: Agent, PlayerProfile, etc.
2. Behavior and Status Categories: Action, Goal, Plan, etc.
3. Perception and Memory types: Perception, Memory, etc.
4. Environment and Space Types: Room, OccupiesRoom, etc.
5. Appearance and Interaction: Appearance, UIState, etc.
6. Auxiliary or Operations and Maintenance: Cleanup, DebugInfo, etc.

![Deconstructing Project89: A Modular, High-Performance Next-Generation AI Agent Framework Design]###https://img-cdn.gateio.im/webp-social/moments-19636582e09b473536b17c2de0c61fbc.webp(

) Manager Architecture

Provides system-level functions such as RoomManager, StateManager, EventBus, etc.

![Deconstructing Project89: A Modular, High-Performance Next-Generation AI Agent Framework Design]###https://img-cdn.gateio.im/webp-social/moments-b0180b99743a98bafb2b2d066900d65c.webp(

) database interaction

Unified management of interactions with the database through StateManager.

![Deconstructing Project89: A Modular, High-Performance Next-Generation AI Agent Framework Design]###https://img-cdn.gateio.im/webp-social/moments-6a3ba796aae66b8a9f977ec99b5bb1c8.webp(

Architectural Innovations

• Each System operates independently, with no inter-calling relationship.
• Easy to add or remove features, highly modular
• Performance is better than traditional object-oriented architecture
• The design of executing the System in layers is highly innovative.

![Deconstructing Project89: A Modular, High-Performance Next-Generation AI Agent Framework Design])https://img-cdn.gateio.im/webp-social/moments-aadac69287c878cdd84a72c6b8929ef2.webp(

Overall, this is a highly modular and high-performance framework that provides new architectural options for the gaming and DEFI fields.

![Deconstructing Project89: A Modular, High-Performance Next-Generation AI Agent Framework Design])https://img-cdn.gateio.im/webp-social/moments-4f9f7656a99298c7a027a5e4722ce4d5.webp(

![Deconstructing Project89: A Modular, High-Performance Next-Generation AI Agent Framework Design])https://img-cdn.gateio.im/webp-social/moments-62bbe0d65d3748c5bc0027c5ffa39288.webp(