This document describes the system architecture and design of the AutoIT Swarm platform.
┌─────────────────────────────────────────────────────────────────────────┐
│ Client Layer │
│ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │
│ │ Web UI │ │ CLI Tool │ │ REST API │ │ Socket.IO │ │
│ └─────────────┘ └─────────────┘ └─────────────┘ └─────────────┘ │
└─────────────────────────────────────────────────────────────────────────┘
↓
┌─────────────────────────────────────────────────────────────────────────┐
│ Application Layer │
│ ┌─────────────────────────────────────────────────────────────────┐ │
│ │ Flask Web Server │ │
│ └─────────────────────────────────────────────────────────────────┘ │
│ ┌─────────────────────────────────────────────────────────────────┐ │
│ │ Task Management Core │ │
│ │ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ │ │
│ │ │ Queue │ │ Scheduler│ │ Progress │ │ State │ │ │
│ │ │ Manager │ │ │ │ Tracker │ │ Manager │ │ │
│ │ └──────────┘ └──────────┘ └──────────┘ └──────────┘ │ │
│ └─────────────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────────┘
↓
┌─────────────────────────────────────────────────────────────────────────┐
│ Agent Layer │
│ ┌─────────────────────────────────────────────────────────────────┐ │
│ │ Agent Orchestrator │ │
│ └─────────────────────────────────────────────────────────────────┘ │
│ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │
│ │ CEO Agent │ │ Project │ │ Senior │ │ Junior │ │
│ │ │ │ Manager │ │ Developer │ │ Developer │ │
│ └─────────────┘ └─────────────┘ └─────────────┘ └─────────────┘ │
│ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │
│ │ QA Agent │ │ DevOps │ │ Security │ │ DB │ │
│ │ │ │ Agent │ │ Agent │ │ Agent │ │
│ └─────────────┘ └─────────────┘ └─────────────┘ └─────────────┘ │
│ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │
│ │ UI/UX │ │ Technical │ │ Support │ │
│ │ Agent │ │ Writer │ │ Agent │ │
│ └─────────────┘ └─────────────┘ └─────────────┘ │
│ ┌─────────────────────────────────────────────────────────────────┐ │
│ │ Dynamic Sub-Agent Factory │ │
│ └─────────────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────────┘
↓
┌─────────────────────────────────────────────────────────────────────────┐
│ Intelligence Layer │
│ ┌─────────────────────────────────────────────────────────────────┐ │
│ │ Ollama LLM Backend │ │
│ │ (Local Model Inference) │ │
│ └─────────────────────────────────────────────────────────────────┘ │
│ ┌─────────────────────────────────────────────────────────────────┐ │
│ │ ChromaDB Vector Store │ │
│ │ (Memory & Context Storage) │ │
│ └─────────────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────────┘
↓
┌─────────────────────────────────────────────────────────────────────────┐
│ HITL Layer │
│ ┌─────────────────────────────────────────────────────────────────┐ │
│ │ Human-in-the-Loop Interface │ │
│ │ (Escalation, Review, Approval) │ │
│ └─────────────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────────┘
Client Request → CEO Agent → Task Delegation → Specialized Agents → Delivery
↓
Human-in-the-Loop (if needed)
-
Request Intake
- Client submits request via Web UI, CLI, or API
- Request is queued in Task Management Core
-
CEO Analysis
- CEO Agent receives the task
- Analyzes requirements and complexity
- Determines if dynamic sub-agents are needed
-
Task Delegation
- Project Manager creates task breakdown
- Tasks are assigned to appropriate agents
- Smart Agent Registry routes to best-fit agents
-
Execution
- Specialized agents execute their assigned tasks
- Progress is tracked in real-time (0-100%)
- Agents collaborate via shared context
-
Quality Gates
- QA Agent validates outputs
- Security Agent reviews for vulnerabilities
- Complex decisions escalate to HITL
-
Delivery
- Final output is compiled
- Task marked as completed
- Results delivered to client
| Component | Responsibility |
|---|---|
| Queue Manager | Task prioritization and ordering |
| Scheduler | Task timing and dependencies |
| Progress Tracker | Real-time progress monitoring (0-100%) |
| State Manager | Task state transitions |
| Agent | Role |
|---|---|
| CEO | Strategic decisions, task analysis, sub-agent creation |
| Project Manager | Task breakdown, scheduling, coordination |
| Senior Developer | Complex implementation, architecture |
| Junior Developer | Supporting development tasks |
| UI/UX Designer | Interface and experience design |
| QA Testing | Quality assurance and testing |
| DevOps | Deployment and infrastructure |
| Security | Security audits and reviews |
| Database Admin | Database design and optimization |
| Technical Writer | Documentation creation |
| Support | User support and issue resolution |
Creates specialized agents on-demand:
- Analyzes task requirements
- Generates appropriate agent capabilities
- Routes to existing agents when possible
- Creates temporary agents for unique needs
┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ Request │ → │ Context │ → │ Agent │
│ Input │ │ Building │ │ Processing │
└──────────────┘ └──────────────┘ └──────────────┘
↓ ↓
┌──────────────┐ ┌──────────────┐
│ ChromaDB │ │ Ollama │
│ Vector │ │ LLM │
│ Store │ │ Inference │
└──────────────┘ └──────────────┘
↓ ↓
┌──────────────┐ ┌──────────────┐
│ Memory │ │ Response │
│ Retrieval │ │ Generation │
└──────────────┘ └──────────────┘
- Local LLM inference
- Model configuration via
config.yaml - Supports all Ollama-compatible models
- No cloud dependencies
- Vector storage for agent memory
- Context retrieval for conversations
- Persistent task history
- Semantic search capabilities
- Real-time bidirectional communication
- Live task updates
- Progress synchronization
- Dashboard auto-refresh
- Local-First: All processing stays on local machine
- No External APIs: No data sent to cloud services
- HITL Controls: Human approval for sensitive operations
- Audit Trail: Complete logging of all actions
- Concurrent Processing: Multiple tasks executed in parallel
- Agent Pooling: Efficient resource utilization
- Queue Management: Intelligent task prioritization
- Stateless Design: Easy horizontal scaling (future)
- Getting Started - Quick start guide
- Features - Feature overview
- Configuration - Configuration options
- API Reference - REST API documentation
Want to learn more? Check out the Agents Guide for detailed agent capabilities.