otai-hf-local

Hugging Face local text classification plugin for Open Ticket AI - on-premise ML ticket categorization.

Overview

otai-hf-local enables local, privacy-preserving text classification using Hugging Face Transformers. Run state-of-the-art NLP models entirely on-premise for ticket categorization, priority detection, and sentiment analysis without sending data to external services.

Features

• 🤖 Local inference - Run models entirely on your infrastructure
• 🔒 Privacy-first - No data leaves your environment
• 🎯 Pre-trained models - Use any Hugging Face text classification model
• ⚡ GPU acceleration - Optional CUDA support for faster inference
• 🎨 Flexible pipelines - Support for multi-label and single-label classification
• 📊 Confidence scores - Get prediction probabilities

Installation

pip install otai-hf-local

For GPU support:

pip install otai-hf-local[gpu]

Configuration

Add the plugin to your Open Ticket AI configuration:

pipes:
  - type: hf_local
    params:
      model_name: "distilbert-base-uncased-finetuned-sst-2-english"
      device: "cpu"  # or "cuda" for GPU
      batch_size: 8

Usage

Programmatic Usage

from otai_hf_local import HFLocalClassifier

# Initialize classifier
classifier = HFLocalClassifier(
    model_name="distilbert-base-uncased-finetuned-sst-2-english",
    device="cpu"
)

# Classify text
result = await classifier.classify(
    text="The server is down and users cannot login"
)

print(result.label)  # e.g., "urgent"
print(result.confidence)  # e.g., 0.95

With Open Ticket AI

from open_ticket_ai import OpenTicketAI

# Load configuration with HF Local pipe
app = OpenTicketAI.from_yaml("config.yml")

# Process ticket through classification pipeline
result = await app.process_ticket(ticket)

Supported Models

Any Hugging Face model for text classification, including:

• Sentiment: distilbert-base-uncased-finetuned-sst-2-english
• Zero-shot: facebook/bart-large-mnli
• Multi-label: joeddav/xlm-roberta-large-xnli
• Custom models: Upload your fine-tuned models to Hugging Face

Requirements

• Python 3.13 or higher
• PyTorch 2.0+
• transformers 4.52.4+
• CUDA (optional, for GPU acceleration)

Performance

• CPU: ~50-200ms per classification (depends on model)
• GPU: ~10-50ms per classification
• Batch processing: Up to 10x faster with batching

Documentation

• Full docs: open-ticket-ai.com
• Hugging Face: huggingface.co/transformers

Contributing

Contributions welcome! See the main repository for guidelines.

License

LGPL-2.1-only - See LICENSE.

Related Packages

• open-ticket-ai - Core application
• otai-base - Base plugin framework
• otai-zammad - Zammad integration
• otai-otobo-znuny - OTOBO/Znuny integration

Links

• Homepage: open-ticket-ai.com
• Repository: GitHub
• Issue Tracker: GitHub Issues
• PyPI: pypi.org/project/otai-hf-local

otai-zammad

Zammad ticket system integration plugin for Open Ticket AI - automated helpdesk and ITSM workflows.

Overview

otai-zammad provides seamless integration between Open Ticket AI and Zammad ticket systems. It enables automated ticket management, AI-powered classification, and intelligent workflow automation for Zammad instances.

Features

• 🎫 Full CRUD operations - Create, read, update, and delete tickets
• 🔍 Advanced search - Query tickets with flexible search criteria
• 📝 Article management - Add notes and articles to tickets
• 🔄 Real-time sync - Keep ticket data synchronized
• 🔐 Secure authentication - Token-based API authentication
• 🎯 Type-safe - Full Pydantic v2 validation

Installation

pip install otai-zammad

Configuration

Add the plugin to your Open Ticket AI configuration:

ticketsystem_service:
  type: zammad
  params:
    base_url: "https://your-zammad.example.com"
    access_token: "your-api-token"
    timeout: 30.0
    verify: true

Usage

Programmatic Usage

from otai_zammad import ZammadTicketsystemService
from open_ticket_ai.models import UnifiedTicket, UnifiedEntity

# Initialize service
service = ZammadTicketsystemService(
    base_url="https://your-zammad.example.com",
    access_token="your-api-token"
)

# Create a ticket
ticket = await service.create_ticket(
    UnifiedTicket(
        subject="Issue with VPN",
        body="Cannot connect to VPN after update",
        queue=UnifiedEntity(name="IT Support"),
        priority=UnifiedEntity(name="2 normal"),
        customer=UnifiedEntity(name="user@example.com")
    )
)

With Open Ticket AI

from open_ticket_ai import OpenTicketAI

# Load configuration
app = OpenTicketAI.from_yaml("config.yml")

# Use the Zammad service
tickets = await app.ticketsystem.find_tickets(limit=10)

Requirements

• Python 3.13 or higher
• Zammad instance with API access
• Valid Zammad API token

Documentation

• Full docs: open-ticket-ai.com
• Zammad API: docs.zammad.org

Contributing

Contributions welcome! See the main repository for guidelines.

License

LGPL-2.1-only - See LICENSE.

Related Packages

• open-ticket-ai - Core application
• otai-base - Base plugin framework
• otai-otobo-znuny - OTOBO/Znuny integration
• otai-hf-local - Local AI model integration

Links

• Homepage: open-ticket-ai.com
• Repository: GitHub
• Issue Tracker: GitHub Issues
• PyPI: pypi.org/project/otai-zammad