The digital landscape is shifting, and a quiet revolution is underway. For years, the cloud promised unparalleled flexibility and scalability, but a growing unease is leading many organizations to reconsider. Critical infrastructure, once thought secure in the vastness of the cloud, is now feeling the squeeze of heightened security threats and geopolitical tensions. Data sovereignty and compliance, once niche concerns, are now front and center, especially in heavily regulated industries. The wild west of cloud-based AI, while offering tempting capabilities, often comes at the cost of control and privacy, leaving many feeling exposed and vulnerable.

This is precisely where Fsas Technologies’ Private GPT steps in. As an on-premise Private AI solution, it offers a complete, self-contained stack that lets you reclaim ownership of your data and your AI. Imagine the power of advanced AI without the gnawing worry of data breaches, non-compliance with stringent regulations, or the unpredictable winds of global politics impacting your operations. Private GPT mitigates these issues by placing the entire solution—from your sensitive data to the AI models themselves—squarely within your infrastructure.

This empowers you with full control, uncompromised security, and the peace of mind of true data ownership, while seamlessly connecting all your data sources to make Private GPT a central hub for complete oversight of your AI processes.

And now, we’re thrilled to announce the upcoming release of Private GPT v1.5, an update that builds on this foundation to deliver even more power, precision, and privacy. Let’s dive into the exciting new features arriving in this landmark release!

Private GPT v1.5.0: A Leap Forward in On-Premises AI

The initial release of v1.5.0 brings a wave of enhancements designed to make your Private GPT experience smarter, faster, and more adaptable than ever.

Default LLM Upgrade: Mistral 3.1 Small (8-bit-quantized)

This is a game-changer! The default Large Language Model (LLM) has been upgraded from Mistral Nemo to the highly capable Mistral 3.1 Small (8-bit quantized). Mistral 3.1 Small is a 24-billion-parameter, open-source model renowned for its text fluency, multilingual understanding, and strong instruction-following capabilities. The 8-bit quantization ensures it runs efficiently on your hardware while delivering significantly improved performance and accuracy compared to its predecessor. You’ll experience faster response times and more intelligent outputs, all within the confines of your private network.

API 2.0

The new API 2.0 provides enhanced integration capabilities, allowing developers to seamlessly connect Private GPT with existing enterprise applications and workflows. We’ve refined the API to offer more granular control and flexibility, making it easier to leverage Private GPT’s power across your entire tech stack.

Hybrid RAG: Smarter Context Handling

Hybrid RAG (Retrieval-Augmented Generation) revolutionizes how Private GPT handles context. Instead of just pulling individual chunks of information, the system can now intelligently retrieve and utilize entire documents as context when they fit within the LLM’s input window. This significantly enhances answer quality, providing more comprehensive and relevant responses by feeding the model a richer, more complete understanding of your data.

XML Source Support

We’ve expanded our document parsing capabilities to include support for XML files. This means Private GPT can now extract and understand structured data within XML documents, broadening the range of enterprise data sources you can leverage for your AI applications. This is in addition to the parsing for HTML and other document types.

Summarization 2.0

Summarization 2.0 brings significant improvements to Private GPT’s ability to condense lengthy texts into concise, accurate summaries. With enhanced understanding of document structure and context, the model can now generate higher-quality summaries that retain key information and nuances, saving you valuable time and improving information accessibility.

Scenarios 2.0: Total Prompt Customization

Scenarios just got a massive upgrade! Scenarios 2.0 gives you unprecedented control over prompt content and model behavior. You can now customize system prompts with dynamic variables (like cutoff_date or user_name), adjust key parameters such as creativity and number of chunks, and define how the AI should respond before, during, and after a user’s message. This allows for highly tailored AI assistants, compliance bots, or customer service tools that perfectly adapt to your specific use case. Furthermore, Scenarios are now accessible and configurable via API for easy integration.

RAG History 1.0: Context That Follows the Conversation

Our new RAG History (Smart History Agent) scenario enables the system to understand follow-up questions across multiple chat turns. This is a limited release feature that automatically rewrites or decomposes user prompts (as also seen in the new Query rewriting and Query Decomposition features), into optimized search queries, filters out irrelevant instructions, and pulls relevant context from both the knowledge base and the ongoing chat. This is ideal for natural, multi-turn conversations in research, support, and internal knowledge use cases, providing a much more intuitive and human-like interaction. (limited release feature means this is only available for scenario-based prompts, expansion beyond this will be considered for future releases).

Beyond v1.5.0: What’s Coming Next!

The innovation doesn’t stop with the initial v1.5.0 release! We’re already hard at work on subsequent updates that will continue to push the boundaries of on-premise AI.

Context Precision Analysis (CPA)

This feature will enable a deeper understanding and evaluation of the relevance and accuracy of retrieved context. This means more precise answers and a reduced risk of the AI generating information based on irrelevant data points. (Note: This feature, along with knowledge graphs, is currently a paid additional feature not included in standard licensing models)

Multi-GPU Support, Default LLM Switching, and Unquantized Mistral Small 3.1

Multi-GPU support will allow you to leverage more powerful hardware configurations, accelerating processing times and enabling more complex AI tasks. The ability to switch default LLMs (from a predefined certified list that can be selected or specified as part of the solution) will provide greater choice and adaptability to your specific needs, allowing you to select the best model for different workloads. Additionally, an unquantized version of Mistral Small 3.1 will be available for those who require the absolute highest fidelity and performance, potentially at the cost of increased hardware requirements.

MCP Integration, Alliance Partner Support, and Advanced Function/Tool Calling

Model Context Protocol (MCP) integration will provide a standardized way for Private GPT to manage, share, and understand context across different AI models and components within your organization. This ensures consistency and efficiency in how context is handled, leading to more robust and reliable AI applications. We’re also thrilled to announce support for our alliance partner, NetApp, ensuring seamless integration with your existing data management and storage solutions. Finally, advanced Function/Tool calling will empower Private GPT to interact with external systems and applications, turning it into a truly intelligent agent capable of performing actions beyond just generating text.

Due to the rapid pace of development along with shifting priorities in the industry the above mentioned features are subject to change and the development cycle may adjust to meet the market and customer demands. For the latest information please check out our public roadmap.

The Future is Private: Your AI, Your Rules

The market is unequivocally shifting. The era of blindly entrusting critical AI workloads and sensitive data to public cloud platforms is drawing to a close for many strategic organizations. Fsas Technologies understands this imperative, and with Private GPT v1.5, we are navigating this changing landscape by empowering you with a cutting-edge, secure, and entirely private AI solution.

It’s time to own your AI. It’s time to safeguard your data. It’s time to ensure compliance without compromise.

Don’t wait to reclaim control. Contact Fsas Technologies today (ai.team@fujitsu.com) to learn more about Private GPT v1.5 and how it can secure and transform your enterprise AI strategy! – or register for an AI Test Drive.

Let us show you how true data ownership combined with advanced AI can drive your business forward, on your terms.