Aleph Alpha launches PhariaAI: The enterprise-grade operating system for generative AI combining future-proof sovereign design with LLM explainability and compliance

• PhariaAI is an open end-to-end solution that enables enterprise and public sector customers to integrate AI and transform their core processes. PhariaAI is tailored to these demanding applications, offering unique features such as hybrid execution, explainability and auditability, to allow building and operating AI systems that are ready for high scale production
• With the launch of PhariaAI, Aleph Alpha Research releases two 7B foundation models, Pharia-1-LLM-7B-control and Pharia-1-LLM-7B-control-aligned; including the complete training code, now publicly available and open-source for non-commercial research and educational purposes

Today, the German AI company Aleph Alpha launches PhariaAI, a full stack offer for enterprises and governments, to develop, deploy and distribute sovereign generative AI technology. Developed to meet the requirements of the most complex and critical information-based value chains, even in highly regulated environments, PhariaAI combines proprietary innovation with state-of-the-art open-source models and components.

This enterprise-ready, open AI stack is tailored to meet the unique demands of environments where general purpose chatbots do not bring the necessary capabilities and human-machine design. PhariaAI includes proprietary innovation to capture and persist human expert knowledge and unique approached for transparency and control enabling experts to take responsibility for high-impact outcomes.

Dedicated components for evaluation and design provide unique tools to accelerate the journey from AI development to deployment, ensuring that every step is securely aligned with each organizational goal. With advanced tools and robust governance features, PhariaAI provides unprecedented transparency, compliance, and control, allowing organizations to maintain complete oversight over their AI operations.

End-to-End AI Solution for Critical Business Solutions

PhariaAI’s Mission is to unify and streamline the development of AI applications in enterprises and public institutions. It empowers AI-developers to effortlessly scale AI projects from proof of concepts into full production systems, evaluating, adapting and orchestrating different proprietary and open-source LLMs.

Moreover, PhariaAI offers an extensive collection of developer tools out of the box to accelerate the development of AI applications. This collection includes for instance: components for finetuning and evaluation, debugging capabilities, unique innovation of Aleph Alpha for explainability and control and auditability of LLMs.

Another cornerstone of PhariaAI is the support for hybrid execution of LLMs. Hybrid execution allows for simultaneous inference of LLMs both in an on-premise and cloud environment. This creates flexibility to distribute capacity to the cloud during peak loads and to ensure that sensitive information is only processed in a secure environment. For common use cases such as Q&A, Summarization, Translation, Transcription and RAG with custom data sources, PhariaAI offers ready-made templates for quick and easy implementation. All these functionalities are available as SAAS solutions or on-premise in the customers’ own datacenter. PhariaAI always offers full flexibility with no vendor lock-in and the freedom to choose between commercial and open-source models.

“PhariaAI is the result of our strategy around innovation and sovereignty and a game-changer for enterprises and governments seeking to leverage AI without compromising control or compliance,” said Jonas Andrulis, founder and CEO of Aleph Alpha. “Our technology is designed to accelerate AI innovation while ensuring that every aspect of development and deployment is transparent, auditable, and fully compliant with organizational standards.“

Introducing Pharia-1-LLM: Capable and compliant for critical applications

As part of the PhariaAI product suite, Aleph Alpha introduces two 7B foundation models, Pharia-1-LLM-7B-control and Pharia-1-LLM-7B-control-aligned, that represent a significant advancement from its former Luminous model family and the latest addition to the LLM portfolio. Pharia-1-LLM-7B-control can be flexibly adapted to user preferences, making it suitable for critical applications without the risk of shutdown behavior, while Pharia-1-LLM-7B-control-aligned comes with additional safety guardrails. The foundation models, trained from scratch on seven languages – particularly optimized for English, German, French and Spanish – are tailored towards concise responses and stand on par with the latest open-source models in the 7B to 8B parameter range.

Moreover, both model variants have been trained in full compliance with applicable EU and national regulations, including copyright and data privacy laws. Pharia-1-LLM-7B is therefore well-suited for applications in large organizations such as enterprises or public institutions.

Advancing Global AI Research: Publication of the training code base

Aleph Alpha Research is one of the world’s first AI abs to openly publish their code base used for model training. “With the release of the Pharia-1-LLM models, Aleph Alpha Research delivers on its’ commitment to open research,” said Dr. Yasser Jadidi, Chief Research Officer of Aleph Alpha Research. “Both the model weights as well as the entire code base used to train the models have been made public for non-commercial open research and education. The code base release is unprecedented in the industry, and we are proud to stand out as the first industrial research institution in the world to have made their training code base openly available. By doing so, we hope to provide a valuable resource to the scientific community that fosters further open research and development.”