AutoIT Swarm

A fully autonomous, agent-swarm-based IT company system powered by multiple LLM providers (Ollama, OpenAI, Claude, OpenRouter, DeepSeek, Qwen). The system simulates a real IT company with specialized AI agents collaborating to deliver IT services end-to-end, with file generation, MCP tools, and Human-in-the-Loop (HITL) escalation.

🚀 Quick Start

# Set environment and start
$env:OLLAMA_URL="http://localhost:11434"  # Windows
python main.py --mode web

# Access dashboard at http://localhost:5000
# Configure LLM and agents in Settings

📖 Documentation

Document	Description
Getting Started	Quick start and usage examples
Features	Key features and capabilities
Architecture	System design and agent architecture
Configuration	Complete configuration reference
Agent Tools	NEW MCP, web search, and tools
Dashboard & API	Web dashboard and API documentation

Additional Guides

Document	Description
Installation Guide	Detailed setup instructions
Agents Guide	Agent roles and capabilities
Task Management	Task control, progress, and pause/resume
HITL Guide	Human-in-the-Loop escalation system
API Reference	REST API reference
Troubleshooting	Common issues and solutions

🎯 What is AutoIT Swarm?

AutoIT Swarm simulates a complete IT company with 12 specialized AI agents that collaborate to deliver IT services:

• Leadership: CEO, Project Manager
• Development: Senior/Junior Developers, Solutions Architect
• Quality: QA Testing, Security Expert
• Operations: DevOps Engineer, Database Administrator
• Support: Support Specialist, Billing Agent
• Marketing: Marketing Agent (NEW)

Plus dynamic sub-agents that can be created on-the-fly for specialized tasks.

✨ Key Features (v2.0)

🧠 Multi-LLM Provider Support

• Ollama - Local LLM (free, private)
• OpenAI - GPT-4, GPT-3.5 Turbo
• Claude - Anthropic Claude 3 family
• OpenRouter - Multi-provider gateway
• DeepSeek - DeepSeek AI models
• Qwen - Alibaba Qwen models

🔌 MCP & External Tools

• Web Search - Research via Tavily/Serper
• Web Fetch - Get content from URLs
• Document Search - Internal knowledge base
• Code Execution - Safe Python execution
• File Read/Write - Access allowed directories

📤 File Generation (15+ Formats)

• Code: HTML, CSS, JavaScript, Python
• Data: JSON, CSV, Excel (XLSX)
• Documents: Markdown, PDF, Word (DOCX)
• Images: PNG, JPEG
• Auto-saved to ./output/{project_id}/

🎛️ Enhanced Dashboard

• Projects Page - Card/list view, deliverables download
• Settings Page - 6 tabs (LLM, HITL, Output, Webhooks, Agents, System)
• Agent Editor - Configure capabilities per agent
• Real-time Updates - Socket.IO integration

🔄 Task Management

• Pause/Resume controls
• Progress percentage tracking (0-100%)
• Task timeout (auto-fail old tasks)
• Infinite loop prevention
• Retry logic with max retries

🔀 Human-in-the-Loop (HITL)

• 8 escalation triggers
• Web-based review panel
• Decision tracking and audit
• Email/Webhook notifications

📦 Installation

# Clone the repository
git clone <repository-url>
cd KimiAS

# Install dependencies
pip install -r requirements.txt

# Configure your LLM provider
# Option 1: Local Ollama (no API key needed)
$env:OLLAMA_URL="http://localhost:11434"

# Option 2: Cloud providers (set API key in Settings)
# OpenAI, Claude, OpenRouter, DeepSeek, Qwen

See Installation Guide for detailed instructions.

🏃 Run the System

# Web dashboard mode (recommended)
python main.py --mode web

# Access dashboard at http://localhost:5000
# Configure LLM provider in Settings → LLM Providers

# Single request mode
python main.py --request "Build me a REST API for user authentication"

# Check system status
python main.py --status

🌐 Web Dashboard

Access the dashboard at http://localhost:5000 for:

Page	Description
Dashboard	Overview with stats, health, performance
Tasks	Task board with pause/resume controls
Agents	Agent status and activity
Projects	Projects with card/list view, deliverables
HITL	Human review panel for escalations
Settings	LLM, Output, Webhooks, Agents configuration

Dashboard Features

• 4 Stats Cards: Total, In Progress, Completed, Pending HITL
• Task Status Chart: Visual breakdown by status
• Recent Tasks: Last 5 tasks with assignee
• Agent Activity: Active agents with task counts
• System Health: LLM, Task Board, Message Queue, Memory
• Active Projects: Project progress bars
• Performance: Completion rate, escalation rate, avg confidence

🔌 API & Webhooks

Submit Request via API

curl -X POST http://localhost:5000/api/submit-request \
  -H "Content-Type: application/json" \
  -d '{"request": "Build a user authentication system"}'

Create Task via Webhook

curl -X POST http://localhost:5000/api/webhook/incoming \
  -H "Authorization: Bearer your-token" \
  -H "Content-Type: application/json" \
  -d '{"description": "Build a login page", "agent": "SeniorDeveloper"}'

Execute MCP Tool

curl -X POST http://localhost:5000/api/mcp/execute \
  -H "Content-Type: application/json" \
  -d '{"tool": "web_search", "arguments": {"query": "Python best practices"}}'

Configure Agent Capabilities

curl -X PUT http://localhost:5000/api/agents/config/SeniorDeveloper/capabilities \
  -H "Content-Type: application/json" \
  -d '{"internet_access": true, "mcp": true, "document_access": false}'

⚙️ Configuration

Configure via Settings page in the dashboard or config.yaml:

# LLM Provider
llm:
  provider: "ollama"  # or openai, claude, openrouter, deepseek, qwen
  url: "http://localhost:11434"
  model: "lfm2.5-thinking:latest"
  temperature: 0.7
  max_tokens: 4096

# Web Search (for agent internet access)
web_search:
  enabled: true
  provider: "tavily"
  api_key: "your-api-key"

# Output Destinations
output:
  file:
    enabled: true
    directory: "./output"
    format: "json"

# Webhooks
webhooks:
  outgoing:
    enabled: true
    url: "https://your-system.com/webhook"
  incoming:
    enabled: true
    auth_token: "your-secret-token"

# MCP Servers
mcp:
  enabled: true
  servers:
    filesystem:
      url: ""
      api_key: ""

See Configuration Guide for all options.

🎯 Example Workflows

1. Generate HTML Page with File Output

# Submit via dashboard
# "Create a responsive login page with HTML and CSS"

# Files generated:
# ./output/{project_id}/login_page.html
# ./output/{project_id}/styles.css

# Download from: Projects → Your Project → Deliverables

2. Agent with Internet Research

1. Settings → Agents → Edit SeniorDeveloper
2. Enable "Internet Access"
3. Submit: "Research best Python web frameworks 2026"
4. Agent searches web and generates report

3. External Integration

# Configure webhook in Settings
# External system creates task:
curl -X POST http://localhost:5000/api/webhook/incoming \
  -H "Authorization: Bearer token" \
  -d '{"description": "Fix login bug", "agent": "SeniorDeveloper"}'

# Receive completion events at your endpoint

🔗 Quick Links

• Documentation Index
• Getting Started
• Agent Tools Guide - MCP and tools
• Configuration Guide
• Dashboard Guide
• Ollama Documentation
• OpenAI Documentation
• Tavily Web Search

---

Need Help? Check the Troubleshooting Guide or open an issue.

Version: 2.0 | Last Updated: March 2026