Multi-Agent System for Intelligent Customer Experience Incident Management

Multi-agent system that automates and structures the customer incident management flow, from receiving the message to automatically creating tickets in Linear when necessary.

🎯 Description

This project implements a multi-agent system that processes Customer Experience (CX) incidents through six specialized agents that work in coordination:

• Interpreter: Analyzes customer text and extracts key information
• Classifier: Classifies the incident by type and affected area
• Validator: Evaluates if there is sufficient information to process
• Prioritizer: Calculates priority based on impact, urgency, and severity
• Decision Maker: Decides whether to escalate to development or manage from CX
• Linear Writer: Automatically creates tickets in Linear when necessary

🚀 Features

• ✅ Automatic incident processing with multi-agent system
• ✅ Complete web interface for ticket management
• ✅ Linear integration for automatic issue creation
• ✅ Support for multiple AI providers (Anthropic, OpenAI, Gemini, Mistral, Ollama)
• ✅ Flexible configuration: use AI or predefined rules per agent
• ✅ Real-time processing with visual feedback
• ✅ Batch processing of multiple tickets
• ✅ Complete traceability of all decisions made

📋 Requirements

• PHP 8.4.16 or higher
• Composer
• Node.js and NPM
• Database (MySQL, PostgreSQL, SQLite)
• (Optional) API keys for AI services
• (Optional) Linear API key

🔧 Installation

1. Clone the repository:

git clone <repository-url>
cd agents

2. Install dependencies:

composer install
npm install

3. Configure the environment:

cp .env.example .env
php artisan key:generate

4. Configure the database in .env:

DB_CONNECTION=mysql
DB_HOST=127.0.0.1
DB_PORT=3306
DB_DATABASE=your_database
DB_USERNAME=your_username
DB_PASSWORD=your_password

5. Run migrations:

php artisan migrate

6. (Optional) Configure AI services in .env:

Global AI Configuration:

# Enable/disable LLM processing globally (default: false)
AI_USE_LLM=true

# Select the default AI provider (anthropic, openai, gemini, mistral, ollama)
NEURON_AI_PROVIDER=anthropic

# Timeout and retry settings
NEURON_AI_TIMEOUT=30
NEURON_AI_RETRIES=2

Anthropic (Claude) Configuration:

ANTHROPIC_KEY=your_api_key
ANTHROPIC_MODEL=claude-3-5-sonnet-20241022

OpenAI Configuration:

OPENAI_KEY=your_api_key
OPENAI_MODEL=gpt-4

Google Gemini Configuration:

GEMINI_API_KEY=your_api_key
GEMINI_MODEL=gemini-2.5-flash

Mistral AI Configuration:

MISTRAL_KEY=your_api_key
MISTRAL_MODEL=mistral-small-latest

Ollama (Local) Configuration:

OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_MODEL=llama2

7. (Optional) Configure Linear in .env:

LINEAR_API_KEY=your_api_key
LINEAR_TEAM_ID=your_team_id

8. Build assets:

npm run build

9. Start the server:

php artisan serve

🎮 Usage

Accessing the application

Once the server is started, access the URL shown by Laravel (default http://localhost:8000) in your browser.

Ticket management

• View tickets: Navigate to /support to see the ticket list
• Process ticket: Click on a ticket and select "Process" to run the multi-agent workflow
• Real-time processing: Use the "Process with streaming" button to see progress in real-time
• Batch processing: Select multiple tickets and process them all at once

Configuration

Access /settings to configure:

Agent Activation:

• Enable/disable individual agents in the workflow
• Disabled agents will be bypassed during processing

LLM Configuration per Agent: Each agent can be configured individually to use:

• Global: Uses the default configuration from AI_USE_LLM environment variable
• LLM: Always uses AI processing for this agent (requires API keys)
• Heuristic: Always uses predefined rules (faster, no API keys needed)

This allows you to:

• Use AI only for complex agents while using heuristics for simpler ones
• Test different configurations without changing environment variables
• Optimize costs by using AI selectively

External Services:

• Configure AI provider API keys in .env (see Installation section)
• Configure Linear integration (API key and team ID)

🏗️ Architecture

The system uses a multi-agent architecture where:

• Each agent has a specific responsibility
• Agents work sequentially, passing information between them
• A central orchestrator coordinates the complete flow
• The entire process is recorded for traceability

🛠️ Technologies

• Laravel 12: PHP web framework
• Neuron AI: Language model integration
• Linear API: Automatic ticket creation
• Tailwind CSS: Interface styling
• Pest: Testing framework

📁 Project Structure

app/
├── AI/
│   ├── Agents/          # Agent implementations
│   ├── Neuron/          # AI-powered agents
│   ├── Orchestrator/     # Orchestrator and shared state
│   └── Prompts/          # Prompts for AI agents
├── Http/
│   └── Controllers/      # Application controllers
├── Integrations/
│   └── Linear/          # Linear integration
└── Models/              # Data models

resources/
└── views/               # Blade views

routes/
└── web.php              # Application routes

🧪 Testing

Run tests:

php artisan test

📚 Documentation

For more details about the project, see PROJECTE.md.

📝 License

This project is licensed under the MIT license.

👤 Author

Eduard Altimiras Duocastella