MencIA is the Artificial Intelligence Strategy 2026-2030 of the Provincial Council of Lugo (Spain), approved unanimously by the Plenary on 26 May 2026. It builds on an AI Laboratory that is already in production, with 17 tools (8 applications and 9 assistants) running open-source and open-weight models on the Council’s own servers. Its aim is to extend this sovereign model, through a provincial platform, to the 66 municipalities in the province with fewer than 20,000 inhabitants.
Initiative overview
MencIA addresses the dilemma of “Digital Feudalism”, in which public administrations hand control of their data to proprietary “black box” technologies. The strategy proposes a different path, which it calls realistic technological sovereignty: open-source and open-weight AI models running on the Provincial Council’s own infrastructure. Small towns would then be able to use secure, publicly controlled AI, with privacy by design and energy efficiency, without depending on large technology vendors. The model rests on three rules:
- Data control and privacy: sensitive information never leaves the Council’s servers.
- Real adaptability: tools are designed around local needs, not the other way round.
- Economic sustainability: recurring licence costs give way to investment in assets and talent that stay under public control.
Sovereignty here is realistic, not dogmatic. Commercial models may be used as a testing ground for non-critical tasks, provided no sensitive data is ever involved.
The guiding principle is “AI as assistant”. AI augments the work of public employees and does not replace them. Human oversight cannot be waived for any administrative act with legal consequences, and the final decision and the responsibility always rest with a person.
The technical stack combines Ollama and vLLM for inference, Docker for containerisation, LangChain for orchestration, Open WebUI for general chat and Streamlit, React and others for purpose-built interfaces. Retrieval-augmented generation (RAG) works only on the Council’s own documents.
A key pillar is “Green by Design” efficiency, delivered through the “3 clicks, one task” paradigm. Each application solves a single, well-defined task and returns a short, structured output, so in practice it behaves like a small language model. This lowers the inference load and avoids the cumulative consumption of conversational chatbots.
The strategy is organised around four pillars: Governance and compliance, internal transformation, service to municipalities, and territorial ecosystem
It creates three new bodies:
- AI Governance Committee: it includes the Data Protection Officer, the legal and audit services, and one representative of each political group in the Plenary.
- Provincial AI Office: the Committee’s executive body.
- Provincial Register of AI Systems: publicly accessible from the Transparency Portal. Each entry records the system’s risk classification under the EU AI Act.
Other relevant details
The governance bodies will be complemented by a Responsible AI Use Policy, Code of Conduct, incident and impact-assessment protocol, and responsible AI procurement guide. This framework is intended to support compliance with the EU AI Act, GDPR and Spanish data protection law. MencIA is offered as a scalable model for territorial cohesion in Europe. Why provincial councils? AI adoption could open a digital divide in smaller municipalities. MencIA holds that provincial councils are well placed to lead because they are close to these municipalities and Law 7/1985 gives them responsibility for e-government services and centralised procurement in municipalities with fewer than 20,000 inhabitants. Lugo has 67 municipalities, only one of which exceeds this threshold. The Territorial Innovation Centre (CIT), funded with EUR 850,000 for 2025–2029, will transfer sovereign AI to municipalities and the local social and economic fabric. Results, outcomes and impacts: Phase 0 is complete. The AI Laboratory is operational with 17 tools (8 applications and 9 assistants); a digital maturity assessment has been carried out using the FEMP.IA methodology; and the Plenary has approved the strategy unanimously. Internal measurements show that daily email management in the Presidency Office fell from about 60 minutes to 5; extraction of insolvency notices from the Spanish Official State Gazette takes seconds rather than 15–20 minutes; and analysis of grant calls fell from about 30 minutes to 2. Legal assistants use a closed corpus and cite the applicable article rather than give opinions or interpret cases. The “AI Teacher” supports AI literacy and Article 4 of the EU AI Act. By 2028, targets are 30 active tools, 60% of staff trained in AI, a 20% average reduction in processing time for AI-assisted files and 15 municipalities on the platform. By 2030, the respective targets are 45 tools, 95%, 35% and all 67 municipalities, including if Lugo city joins. Next steps include establishing the governance bodies and register, running 3 to 5 low-risk pilots and launching the “Impulsa tu IA local” literacy plan in 2026–2027, municipal pilots in 2028–2029, and full integration in 2029–2030. Challenges and lessons learned: The main risks were falling into the “Valley of Pilots” and slipping into digital feudalism through dependence on proprietary solutions. The response was sovereign on-premises infrastructure with open models, so sensitive data remain within the public perimeter. Cultural resistance and fear of job replacement were addressed by framing AI as a co-pilot that augments staff rather than replacing them, supported by the “AI Teacher” programme. MencIA began in 2025 as a bottom-up initiative without a dedicated budget. Its first results preceded the formal framework and helped build the institutional and cross-party trust needed to scale. Success requires institutional leadership, ownership of computing resources and an “Efficiency by Design” mindset.
