Skip to content

kosmos_architecture.md

Latest commit

 

History

History
121 lines (96 loc) · 3.9 KB

kosmos_architecture.md

File metadata and controls

121 lines (96 loc) · 3.9 KB

KOSMOS Software Architecture

graph LR
    %% External Systems
    KSP[Kerbal Space Program]
    kRPC[kRPC Mod]
    OpenAI[OpenAI API]
    
    %% Main Loop Flow
    subgraph "KOSMOS Execution Loop"
        direction TB
        
        %% Telemetry Collection
        TelemetryCollector[📊 Telemetry Collector]
        
        %% Agent Communication Loop
        subgraph "Agent Communication Loop"
            direction LR
            MissionControl[🎯 Mission Control<br/>Plans next mission]
            FlightAgent[🚀 Flight Agent<br/>Generates & executes code]
            AuditAgent[🔍 Audit Agent<br/>Monitors progress]
        end
        
        %% Environment Bridge
        KSPEnv[🌉 KSP Environment<br/>Bridge to KSP]
        
        %% Data Storage
        VectorDB[💾 Vector Database<br/>Skills & QA Cache]
        Checkpoint[💾 Checkpoint System<br/>Mission State]
    end
    
    %% External Connections
    KSPEnv --> kRPC
    kRPC --> KSP
    
    %% Data Flow - Telemetry Loop
    KSPEnv --> TelemetryCollector
    TelemetryCollector --> MissionControl
    TelemetryCollector --> FlightAgent
    TelemetryCollector --> AuditAgent
    
    %% Agent Communication Loop
    MissionControl -->|"1. Propose Mission"| FlightAgent
    FlightAgent -->|"2. Execute Code"| KSPEnv
    KSPEnv -->|"3. Get Results"| AuditAgent
    AuditAgent -->|"4. Provide Feedback"| MissionControl
    
    %% AI Processing
    MissionControl --> OpenAI
    FlightAgent --> OpenAI
    AuditAgent --> OpenAI
    
    %% Data Persistence
    MissionControl --> VectorDB
    FlightAgent --> Checkpoint
    AuditAgent --> VectorDB
    
    %% Control Flow
    FlightAgent -->|"Time-sliced execution<br/>2-3 second loops"| KSPEnv
    AuditAgent -->|"Real-time monitoring<br/>Intervention capability"| FlightAgent
    
    %% Styling
    classDef external fill:#e1f5fe,stroke:#01579b,stroke-width:3px
    classDef telemetry fill:#fff3e0,stroke:#e65100,stroke-width:2px
    classDef agent fill:#e8f5e8,stroke:#1b5e20,stroke-width:2px
    classDef bridge fill:#f3e5f5,stroke:#4a148c,stroke-width:2px
    classDef storage fill:#fce4ec,stroke:#880e4f,stroke-width:2px
    
    class KSP,kRPC,OpenAI external
    class TelemetryCollector telemetry
    class MissionControl,FlightAgent,AuditAgent agent
    class KSPEnv bridge
    class VectorDB,Checkpoint storage
Loading

Architecture Overview

Core Components

  1. Kosmos (Main Orchestrator)

    • Central coordination hub
    • Manages agent lifecycle
    • Handles mission execution flow
    • Coordinates between all agents

  2. Environment Layer

    • KSPEnv: Bridge between KOSMOS and Kerbal Space Program
    • TelemetryCollector: Real-time data collection from KSP vessel

  3. AI Agent Layer

    • MissionControlAgent: Plans missions and proposes next steps
    • FlightAgent: Generates and executes kRPC Python code
    • AuditAgent: Monitors progress and provides critiques
    • ManeuverAgent: Manages reusable maneuver library

  4. Control Primitives

    • Executable functions for common operations
    • Context examples for AI training
    • Reusable code snippets

  5. Data Storage

    • Checkpoint system for mission state
    • Vector database for QA caching and skills
    • Mission history tracking

Data Flow

  1. Telemetry Collection: TelemetryCollector → All Agents
  2. Mission Planning: MissionControlAgent → FlightAgent
  3. Code Execution: FlightAgent → KSPEnv → kRPC → KSP
  4. Progress Monitoring: AuditAgent → MissionControlAgent
  5. Skill Learning: ManeuverAgent → Vector Database

Key Features

  • Time-sliced Execution: Short loops for real-time responsiveness
  • Shared Control Bus: Real-time communication between agents
  • Skill Library: Persistent storage of successful maneuvers
  • Checkpoint System: Mission state persistence
  • Multi-agent Coordination: Specialized agents for different tasks