MencIA is a sovereign AI strategy that aims to offer a Platform as a Service (PaaS) to 66 municipalities in the province of Lugo with fewer than 20,000 inhabitants. It will enable local councils to adopt secure, local, open-source AI tools, avoiding dependence on large corporations. The initiative will bridge the digital divide, ensuring that small towns benefit from AI through publicly controlled technology, privacy by design, and efficiency.
Initiative overview
MencIA addresses the dilemma of “Digital Feudalism,” where public administrations cede control of data to proprietary “black box” technologies. Instead, this strategy proposes a new path: realistic technological sovereignty based on open-source AI models running on the Lugo Provincial Council's own servers.
The approach chosen by MencIA allows it to comply with its three key rules:
- Privacy and security, as data never leaves the perimeter of our servers.
- Adaptability, as by using AI models installed on our servers, we can design and adapt the tools we create to our needs.
- Economic sustainability, as the model allows us to save on recurring licensing costs and invest in talent.
The guiding principle of the model is “AI as Co-Pilot.” We defend the idea of an augmented administration in which AI never replaces public employees, but rather enhances their capabilities. Human oversight is non-negotiable for any administrative act with legal consequences. The architecture uses tools such as Ollama, vLLM, Docker, LangChain, and OpenWebUI to configure a dual system of specialized AI assistants and applications that solve specific problems with a minimal learning curve for use. MencIA natively integrates the principles of privacy, cybersecurity, and governance, risk, and compliance (GRC) into its design. The native integration of all these elements facilitates compliance with the European AI Act and the GDPR, as well as all our national regulations on AI and data protection. A key pillar of the model is “Green in AI” sustainability. By using Small Language Models (SLMs) and a “3-click” interaction design, the strategy minimizes token generation and energy consumption, demonstrating that high-performance AI can be environmentally responsible. MencIA serves as a scalable model for territorialcohesion in the European context and aligns with other Spanish and European AI implementation initiatives such as the Spanish Government's ALIA project, the OpenEuroLLM project, and the European Union's Apply AI strategy.
Other relevant details
Given the digital divide that the implementation of AI may entail in smaller municipalities, we defend the idea that Spanish provincial councils are the ideal actors to lead this transformation process at the provincial level for two reasons:
1. They are the ones who know the territory best and are closest to small municipalities, where the distance from regional and central governments is often too great.
2. Spanish legislation itself (Article 36.1.g) of Law 7/1985, regulating the bases of local government, assigns them the provision of e-government services and centralized procurement in municipalities with fewer than 20,000 inhabitants, a competence that fits perfectly with the AI implementation process. In the province of Lugo, there are 67 municipalities, and only the capital exceeds that number: for the other 66, the Provincial Council is their only guarantee of not being left behind.
Results, outcomes and impacts:
Results of the strategy in the Provincial Council: 90% savings with the intelligent email management tool (from 60 to 5 minutes) and with the BOE edict search tool (from 40 to 1 minute). Our legal assistants (public procurement, human resources), who use RAG techniques, guarantee legal certainty with accurate regulatory citations. Measurement: time per task and rigor in compliance. Future: scaling via PaaS to 66 municipalities (<20k inhabitants). The “Green in AI” design minimizes the carbon footprint through SLMs and “3-click” flows. Our “AI Teacher” tool closes the digital skills gap and allows us to advance in compliance with Article 4 of the AI Act. Impact: transition of employees from operational tasks to public value management.
Challenges and lessons learned:
The main risk was falling into the “Valley of Pilots” (isolated experiments that do not scale) or succumbing to “Digital Feudalism” due to dependence on proprietary solutions. We responded by building a sovereign on-premise infrastructure with open-source models, ensuring that sensitive data never leaves the public perimeter. Cultural resistance and fear of job replacement were mitigated by defining AI as a “Co-pilot” that augments, rather than replaces, staff, supported by the “AI Teacher” program. A key lesson is that “Realistic Sovereignty” is the only path to full compliance with the GDPR and AI Regulation with critical data. Success requires three conditions: 1) Strategic institutional leadership, 2) Ownership of computing resources (GPUs), and 3) An “Efficiency by Design” (Green AI) mindset to ensure environmental and economic sustainability.
