Analyzing Project89: A Modular, High-Performance Next-Generation AI Agent Framework

Project89 adopts a completely new approach to designing the Agent framework, which is a high-performance Agent framework for game development, more modular and with better performance compared to currently used Agent frameworks.

This article will provide a detailed introduction to the high-performance Agent framework ArgOS in Project89.

1. Why Use ECS to Design the Agent Framework

ECS(Entity-Component-System) is an architectural pattern commonly used in game development and simulation systems. It separates data from logic thoroughly to efficiently manage various entities and their behaviors in large-scale scalable scenarios.

1. Entity( Entity): is merely an ID, containing no data or logic.
2. Component( component): used to store the specific data or state of the entity.
3. System (System ): Responsible for executing logic related to certain components.

In ArgOS, each Agent is regarded as an Entity and can register different components, such as:

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

System workflow:

1. Perception System updates perception data
2. Memory System persists perceptual data to the database
3. Action System executes actions based on memory
4. Finally obtain an updated Agent Entity

2. ArgOS System Architecture

ArgOS has designed various Components and Systems, dividing the Systems into three levels (ConsciousnessLevel):

1. Conscious ( SYSTEM: High update frequency, executed every 10 seconds.
2. 潜意识)SUBCONSCIOUS( System: Update frequency is low, executed once every 25 seconds
3. UNCONSCIOUS ) system: updates at the slowest frequency, executing once every 50 seconds or more.

The main System includes:

• PerceptionSystem: Collect external stimuli and update to the Perception component
• ExperienceSystem: Converts stimulation into abstract experiences and stores them.
• ThinkingSystem: Engage in thinking and generate thinking results
• ActionSystem: Execute Action
• GoalPlanningSystem: Evaluate goal progress and generate new goals
• PlanningSystem: Generate execution plans for objectives
• RoomSystem: Handle room-related updates
• CleanupSystem: Remove useless entities

3. ArgOS Overall Architecture Analysis

1. Core Architecture Layering: Entity-Component-System

2. Component Classification:

   • Core Identity Types: Agent, PlayerProfile, etc.
   • Behavior and State Category: Action, Goal, Plan, etc.
   • Perception and Memory: Perception, Memory, etc.
   • Environment and Space Category: Room, OccupiesRoom, etc.
   • Appearance and Interaction: Appearance, UIState, etc.
   • Auxiliary or operation & maintenance types: Cleanup, DebugInfo, etc.

3. Manager Architecture:

   • EventBus: Event publishing and subscription
   • RoomManager: Manage room layout
   • StateManager: Synchronization between ECS and the database
   • ActionManager: Manage action execution
   • PromptManager: LLM Prompt Management

4. Interacting with the database:

   • Data loading and persistence through StateManager
   • Persist periodically or event-driven
   • Save state on exit

4. Architectural Innovations

1. Each System operates independently, with no calling relationships, making it easy to expand.
2. Modular design, flexible configuration of Agent capabilities
3. The ECS architecture has excellent performance and is suitable for complex scenarios.
4. Divide the System into layers of consciousness, simulating the human cognitive process.

Overall, ArgOS is a modular, high-performance, and easily expandable Agent framework that provides new architectural options for game development and AI Agent systems.