Autonomous Model of Distributed Artificial Intelligence to Improve Sustainability and Safety in AI Environments
The MAIA project is an initiative of Blueroom Innovation and the i2CAT Research and Innovation Centre, developed within the framework of Projectes R&D Empresarial supported by ACCIÓ.
The overall objective of the MAIA project is to create a multi-agent Artificial Intelligence system (based on open-source language models) that is autonomous, collaborative, sustainable and secure. From i2CAT’s perspective, the aim is to lead deep technological research to design the architecture of these agents and, above all, to equip them with a layer of governance, cybersecurity and privacy. The goal is to enable AI to make complex decisions without compromising data.
The project provides digital sovereignty, trust and sustainability. Currently, using large AI models involves sending sensitive data to third-party cloud servers, raising privacy concerns and driving high energy consumption. i2CAT provides cutting-edge solutions to address this, integrating advanced cryptography such as Proxy Re-encryption, zero-knowledge proofs (ZKPs) and decentralised identities. This allows AI agents to share context and execute tasks in a fully traceable, auditable and private manner.
The solution will be implemented either locally (on-premises) or in Edge environments (devices with limited resources, such as mini-PCs or equipment within the industry itself). This way, the models run directly on the company’s or institution’s infrastructure, ensuring critical information never leaves the corporate environment. The system is designed modularly to support strategic sectors with high data management and security needs, such as Industry 4.0, logistics (optimisation of industrial processes and energy management), or public administration (supporting professionals and interacting with citizens). Also in specific sectors such as digital health or energy.
