What is GenAI?

Generative AI (GenAI) refers to a category of artificial intelligence models capable of creating new content based on learned patterns from vast datasets. Unlike traditional AI, which primarily classifies or predicts outcomes based on existing data, generative AI can generate text, images, code, and other forms of content dynamically.

These models, typically based on large language models (LLMs) such as but not limited to Mistral, GPT, or LLaMA, process input data, understand context, and produce human-like responses. The core mechanism behind generative AI involves deep learning architectures, particularly transformers, which enable advanced natural language understanding and content generation.

Utilizing Matrix42 Local GenAI in M42 Pro Platform

Matrix42 local GenAI is a secure and fully self-contained AI model designed to enhance IT Service Management (ITSM) operations. By integrating with key enterprise AI tools, it enables knowledge delivery, automation, and agent assistance, all while ensuring data privacy and compliance.  

AI Knowledge Discovery (RAG) – Delivering Information to End Users

Matrix42 GenAI can power AI Knowledge Discovery by leveraging Retrieval-Augmented Generation (RAG). This allows end users to receive accurate, context-aware answers based on the organization's internal knowledge, such as:

• IT service documentation.
• Knowledge bases.
• Company policies.
• Internal repositories (SharePoint, KB articles, wikis, PDFs, Confluence, and structured data sources).

The model ensures that only authorized and relevant information is retrieved and shared, enhancing self-service capabilities while reducing the workload on IT support teams.

Actions – Instant AI Insights, Configured for Your Needs

Actions generate on-demand, exploratory content that helps agents understand tickets and make faster decisions—without cluttering permanent fields. With configurable AI, customers can tailor Actions to their specific workflows and priorities. Key capabilities include:

• Contextual summaries of ticket state, history, or issue progression—shareable with customers or stakeholders.
• Quick translations of descriptions and communications.
• Guided troubleshooting based on ticket details and known solutions.
• Smart recommendations for next steps and workflow triggers.

Agents can easily copy and repurpose generated insights wherever needed—sending summaries to customers, sharing status reports with managers, or leveraging analysis in other contexts. By configuring Actions to match your service model, organizations unlock immediate intelligence that keeps agents focused on resolution while AI handles analysis and exploration.

Writing Assistance – Empowering Agents to Create Better Content

Writing Assistance leverages Matrix42 GenAI to help service agents draft professional, contextually relevant content across the platform—from ticket emails and resolution texts to implementation plans and test documentation. Key benefits include:

• Consistent, professional communication across all content types and channels.
• Faster content creation by auto-generating initial drafts tailored to context.
• Personalized, contextual outputs that adapt to user requests, ticket details, and historical interactions.
• Quality assurance with grammar, style, and tone corrections built in.

By using Writing Assistance, agents can focus on problem-solving and decision-making while AI handles drafting and refinement—improving both productivity and content quality across the platform.

Conclusion

Matrix42 local GenAI integrates seamlessly into AI Knowledge (RAG), AI Actions, and AI Email Assist, providing:

• Secure and private AI-driven knowledge access.
• Automated IT workflows to support agents.
• Enhanced communication tools for faster, more professional email responses.

With full data sovereignty and local deployment, organizations can leverage AI’s power without compromising security or compliance, making Matrix42 GenAI a trusted AI solution for ITSM operations. 

Key benefits

• Optimized for ITSM – The model is specifically fine-tuned for service desk applications.
• Multi-Language Support – English, Finnish, Swedish, German.
• RAG Support (with supported products) – Capable of integrating with knowledge bases, websites, and document repositories.
• Scalability – runs on PUNA inference server, handling multiple output texts at comfortable speed even on cost-effective datacenter GPUs.
• Security & Data Privacy – Designed for regulatory compliance (e.g., GDPR) and robust data protection.

Technical details

• Base Model: PHI-4 14b (Open Model by Microsoft)
• Quantization: Q6_K, Q8 (reducing memory requirements by half while maintaining 95-98% accuracy).
• Context Length: 16k tokens
• Inference Engine: llama.cpp 
• Hardware Requirements:
  • Minimum: NVIDIA CUDA GPU (Ampere or later) with at least 20GB VRAM (NVIDIA L4, RTX4000 ADA)
  • Recommended (moderate traffic, Knowledge Discovery /RAG/): NVIDIA L40S (48GB VRAM), NVIDIA RTX 6000 ADA
• Training Data Sources:
  • Public datasets (licensed appropriately).
  • Synthetic data generated via high-quality commercial models (only those permitted by the license).
  • Manually curated and validated training datasets.

Data privacy

• Local Hosting Available – The model can be deployed on-premises, ensuring full control over sensitive data.
• We can also provide a model from our infrastructure, two options are available:
  • Equinix (Finland)
  • Hetzner (Germany)
• Data Masking & Anonymization – No personally identifiable information (PII) is stored during processing.
• Access Control Mechanisms:
  • Firewall-based protections limit unauthorized access.
  • No External API Calls Required – Unlike cloud-based LLMs, this model does not send queries to external OpenAI APIs, ensuring data residency and privacy compliance.
• Full Data Ownership by the Customer – The client retains exclusive ownership of data throughout the entire process.
• No Data Used for Model Training – Customer data is never used for model training or fine-tuning.
• Data Used Solely for Responses – Information provided to the model is only used for generating responses and is not stored or processed further.
• No Sensitive or Authentication Data is Transmitted – The model does not process or transmit authentication credentials or sensitive data, ensuring security.
• No Data Retention – The model does not store any user input or generated responses.
• Service logs are stored locally - Only admins have access, they are not shared or processed.
• It is possible to disable all logs - but this will reduce debugging capabilities
• Customer-Controlled Data Hosting – Data at rest (knowledge base, logs, settings, CAI platform) will be hosted from servers where customer's ITMS is running:
  
  • Equinix (Finland)
  • Noris (Germany)
  • On-premise / private cloud 

These mechanisms ensure that Matrix42 GenAI provides full data privacy, regulatory compliance, and secure deployment for enterprise IT environments.

Compliance

• GDPR Compliant – Fully adheres to EU General Data Protection Regulation, with on-premise options to ensure data sovereignty.
• ITSM Integration Security – Ensures secure authentication and API token management for integrating with Matrix42 and third-party ITSM tools.
• Secure Model Training – All training datasets were sourced under proper licensing and subjected to manual review for compliance.
• Auditable Logs & Monitoring – Logs do not store sensitive user inputs but enable security audits.

Matrix42 GenAI and EU AI Act compliance

Matrix42 GenAI has been designed and developed in alignment with the EU AI Act's regulatory principles, ensuring trustworthy, transparent, and secure AI usage within enterprise environments, particularly in IT Service Management (ITSM). The key compliance measures include:

1. Risk Classification & AI Transparency
   1. Matrix42 GenAI falls under the "limited risk" category, meaning it does not pose significant safety or fundamental rights concerns.
   2. It is a business-support AI, primarily assisting IT service operations without autonomous decision-making that affects individuals' legal rights or freedoms.
2. Human Oversight & Explainability
   1. The model does not operate autonomously in critical decision-making. Human operators, such as IT service agents, remain in control of AI-assisted outputs.
   2. Responses can be audited, reviewed, and modified by human agents before being deployed in critical workflows.
3. Bias & Fairness
   1. The model has been trained using carefully curated datasets, ensuring it does not propagate unfair biases in its responses.
   2. A continuous monitoring mechanism allows customers to validate outputs and fine-tune responses according to internal compliance policies.
4. Data Protection & Privacy (GDPR Alignment)
   1. Matrix42 GenAI follows strict data privacy measures, ensuring that:
      1. No customer data is stored or used for model retraining.
      2. It does not retain or transfer sensitive information outside the designated customer environment.
      3. It supports on-premise and private cloud deployment, enabling full data residency control.
5. Security & Robustness
   1. Matrix42 GenAI is resistant to adversarial manipulations such as prompt injection attacks, ensuring trustworthy and secure AI-generated responses.
6. Auditability & Compliance Reporting
   1. The system provides logging and monitoring for organizations that require audit trails for compliance with AI governance frameworks.
   2. Upon request, all logs can be disabled, aligning with strict privacy-by-design principles.