Winner·United Nations AI for Good·Innovate for Impact Award
Innovate for Impact
An ITU initiative advancing AI solutions for global challenges. ReEnvision AI is honored to be recognized among the winners advancing inclusive, responsible AI deployment across 28 countries and 11 impact domains.
Sovereign Distributed Agent Operating System
ReEnvision AI introduces a Sovereign Distributed Agent Operating System (DAO-OS) that democratizes AI access for the Global South — bypassing centralized infrastructure to build, use, and share AI on a resilient distributed network of everyday devices.
The Case for Change · Why Now
As a world community, there are 3 serious problems that — right now — we CAN and MUST address urgently.
Sovereignty
Data, language, and intelligence are owned by a handful of companies — communities don't control their own AI.
If you don't own your AI, it can be switched off.
Environmental Harm
Centralized AI burns energy and evaporates water — concentrated exactly where resources are already scarce.
AI that burns energy & evaporates water today.
Accessibility
AI is locked behind cost, compute, and connectivity — shutting out billions of people and most of the Global South.
Locked behind cost, compute & connectivity today.
Our Answer: One distributed system — AgentOS on AgentGrid — directly solves all three: accessible, sovereign, and renewable by design.
The Map of Who Has AI
AI compute is not evenly distributed.
The compute is where the people are not. Where the population is largest, the infrastructure is smallest.
of the world's AI compute regions sit in just two countries — the US & China.
countries host more than half of all AI compute regions.
countries host any at all — in South America only Brazil, in Africa only South Africa.
AI compute regions by country (bubble size = regions)
people in Africa & South Asia — 9 compute regions between them.
people in N. America & Europe — 67 compute regions.
Four Walls Lock the Global South Out of AI
Centralized, energy-intensive AI — owned by a few — blocks the communities that need autonomous agents most.
Financial Exclusion
Exorbitant licensing and compute costs lock MSMEs, rural clinics, and schools out of the digital economy, widening global disparities.
Environmental Degradation ("Red AI")
Centralized AI requires massive energy and water, clashing with sustainability goals. Building sovereign data centers in resource-stressed regions is ecologically impossible.
Infrastructural Fragility
Cloud AI needs high-speed, continuous internet. In regions with unstable networks (fluctuating 5-20 Mbps), cloud-dependent systems fail catastrophically.
Linguistic Marginalization
High-resource models are overwhelmingly English-centric, ignoring the nuances of over 132 Global South languages. This scarcity of training data renders standard AI useless for local populations, enforcing cultural homogenization.
Our Solution
A platform to build, use, and share AI on a distributed network — five pillars that together make sovereign AI practical, sustainable, and inclusive.
Circular Green AI
Rather than building resource-intensive data centers, the platform aggregates idle processing power from existing consumer devices (up to seven years old) into a local mesh network. This eliminates the need for industrial cooling and high-voltage grids, reducing e-waste and promoting a circular hardware economy.
Agent OS
Community members can easily build, deploy, and share custom autonomous agents tailored to individuals, local education, healthcare, agriculture, and business needs.
Offline-First Resilience
Designed for high-latency environments, the platform uses LLM sharding and edge inference. Agents operate locally even during outages or when bandwidth drops below 5 Mbps, syncing only when connectivity returns.
Linguistic Equity
Native multi-lingual Model-as-a-Service (MaaS) shards support 227 languages, ensuring cultural sovereignty and high-fidelity service using open source models for a native/translated UI (AgenticOS).
Strategic Partnerships
Key partnerships with NVIDIA, AMD, and Apple provide hardware, software, and technical support for the platform. We also collaborate with MIT and Cornell University on Project NANDA (Networked Agents and Decentralized AI), an initiative originating from the MIT Media Lab.
Technical Architecture
Collaborative edge computing, adaptive model sharding, and swarm intelligence — engineered for the emerging Agentic Web.
Collaborative Edge Computing
Adaptively partitioning computation-intensive models into affordable shards distributed across heterogeneous edge devices, optimizing inference latency and throughput without sacrificing accuracy.
Swarm Intelligence
Diverse, specialized LLM experts adapt via swarm intelligence, enabling tuning-free model adaptation in low-data regimes with as few as 200 examples — ideal for critical local tasks like healthcare triage.
Sparsely Gated Mixture of Agents/Experts
Conditional compute models natively support 227 languages, breaking the language barrier for low-resource dialects while ensuring safe, high-quality translation and cultural representation.
Local-First Storage
Agents maintain operational continuity during blackouts or severe bandwidth throttling (5-20 Mbps), executing decisions locally and synchronizing with the broader decentralized network only when stable connectivity is restored.
Edge-Network-Cloud Continuum
Engineered for the emerging "Agentic Web," ensuring AI agents autonomously perceive, make decisions, and interact across the full continuum from edge to network to cloud.
Deployment · Grid Sizing
200 AI PCs can serve 10,000 users.
Decentralized MoE and speculative orchestration turn modest hardware into enterprise-grade AI infrastructure.
Privacy & Security
Security by Design.
No single server ever sees your complete request. Every connection encrypted, every handoff signed — privacy that is mathematically enforced, not policy dependent.
Each server sees one block — never your whole prompt.
The Circular AI Economy
ReEnvision rejects the unsustainable "take, make, dispose" model of standard AI hardware. Instead, the platform implements a Circular AI Economy based on the DCEA-4 framework — reintegrating consumer devices up to seven years old into the computational mesh and extending a device's life to save up to 190,000 liters of water per laptop.
Max device age reintegrated into the mesh
Water saved per laptop through reuse
Green AI lifecycle: selection → end-of-life
Framework guiding the Circular AI Economy
Smart procurement and decentralized algorithms monitor device health in real-time, rerouting shards to prevent catastrophic failure and maximize longevity. The transition follows a five-phase Green AI lifecycle — from selection and development to task optimization, maintenance, and end-of-life circularity. By repurposing potential e-waste as foundational infrastructure, ReEnvision proves high-performance AI is compatible with environmental sustainability.
The Choice
Red AI vs. Green AI.
The same intelligence can be built two ways. One burns new resources in a few giant buildings; the other reuses what communities already own.
Red AI
Centralized data centers
- ✗Cooling towers that evaporate water
- ✗Megawatt power draws on strained grids
- ✗New hardware — extraction, carbon, e-waste
- ✗Concentrated in water-stressed regions
- ✗Rented access that can be revoked
Green AI
Distributed on existing devices
- ✓Near-zero cooling water at the source
- ✓Idle capacity on devices already on
- ✓Reuses hardware up to 7 years old
- ✓Spread across communities, not concentrated
- ✓Owned, not rented — no kill switch
Green Datacenter · Energy Per Token
Two-thirds less power for every token.
of a traditional datacenter's energy per token
less power for every token generated
Bringing It Together
Three problems — one system solves all three.
Sovereign
Owned, not rented. It runs on the community's own devices — no kill switch, no borrowed permission.
Accessible
Everyone in. Any device, 227 languages, online or offline — AgentOS puts AI in every hand.
Renewable
Green by design. Two-thirds less power per token, on idle devices — no new data center needed.
One Platform: AgentOS on AgentGrid — sovereign, accessible, and renewable, today.
Global Impact & Sustainable Development Goals
The DAO-OS advances global development goals, aligning with 2026 ITU AI for Good tracks. Scaling beyond 2025 pilots, it delivers multi-sector impacts across five SDG areas.
Good Health and Well-Being
Medical triage agents run on older hardware in rural clinics. Multilingual support expands expertise to underserved areas, improving community well-being.
Quality Education
227-language tutoring agents democratize personalized learning. By overcoming device and language barriers, the system creates inclusive environments for remote students.
Industry, Innovation & Infrastructure
Resilient digital networks built from localized devices. This decentralized mesh enables advanced technology in resource-constrained areas without massive capital costs.
Reduced Inequalities
Removes licensing fees and English-centric software barriers, empowering local entrepreneurs and governments to join the digital economy equitably.
Climate Action
A circular hardware economy extends device lifecycles and rejects energy-intensive data centers, providing a sustainable path for digital transformation.
12-Month Priority Projects
ReEnvision AgentOS and Agent Grid v1.2 are ready for expanded community deployment. The next 12 months focus on repeatable, well-supported pilots that generate measurable outcomes and training playbooks.
Rural Clinic Agent Deployment
KPI: 2-3 clinics or community health partners; 3-5 workflows live: intake, triage support, translation, patient education, admin.
Partners: Health ministries, rural clinics, WHO-aligned NGOs, NVIDIA/AMD technical support.
Remote Learning Agent Hubs
KPI: 2 education partners; 200-500 learners reached; 5-8 tutoring or teacher-assist templates deployed.
Partners: Education ministries, UNESCO-aligned programs, universities, device makers, telecoms.
Local AI Infrastructure Mesh
KPI: 1-2 regional testbeds; 15-25% inference efficiency improvement on modest devices; offline/edge reliability testing.
Partners: NVIDIA, AMD, Dell, Lenovo, HP, telecoms, local ICT agencies.
Community Agent Access Program
KPI: 8-12 reusable agent templates; 5-10 community organizations onboarded; deployment guide completed.
Partners: UNDP digital transformation teams, local governments, NGOs, corporate CSR sponsors.
Circular Compute Deployment
KPI: Document 2-3 reuse cases for older hardware; measure device suitability and energy/cost savings estimates.
Partners: Corporate device donation programs, refurbishment NGOs, climate funds.
Offline Agent Installer
KPI: Beta release by Q3/Q4 2026; 3 pilot environments testing scheduled/offline agents.
Partners: Local IT partners, clinics, schools, municipal ICT teams.
Regional Deployment Priorities
Targeting regions where the use case is clear, local partners are likely, language diversity matters, and modest hardware or intermittent connectivity creates a strong need for sovereign local AI.
East Africa
Kenya, Rwanda, Uganda, Tanzania
Why It Makes Sense
Strong digital public infrastructure momentum, rural health and education needs, active NGO and innovation ecosystems.
Initial Use Cases
Clinic intake, community health agents, remote tutoring, agriculture/admin support.
12-Month KPI
1 health or education pilot; 10-15 local trainees; 3-5 agents deployed.
South Asia
India, Bangladesh, Nepal
Why It Makes Sense
Large rural populations, strong device refurbishment potential, high language diversity, health and education demand.
Initial Use Cases
Multilingual tutoring, clinic workflows, local government service navigation.
12-Month KPI
1 regional partner; 5-8 templates localized; device suitability report.
Southeast Asia & Pacific
Philippines, Indonesia, Pacific Islands
Why It Makes Sense
Geographic fragmentation and disaster risk make offline/local AI especially relevant.
Initial Use Cases
Offline education hubs, disaster preparedness agents, clinic support in island communities.
12-Month KPI
1 offline pilot; network outage test; training playbook completed.
Latin America
Colombia, Peru, Ecuador
Why It Makes Sense
Remote communities, multilingual/indigenous language needs, education and clinic access challenges.
Initial Use Cases
Rural learning support, health navigation, small-business agent templates.
12-Month KPI
1 community hub; 200+ users supported; case study produced.
West Africa
Ghana, Senegal, Nigeria
Why It Makes Sense
Fast-growing digital economy, youth employment need, multilingual context, strong entrepreneurship potential.
Initial Use Cases
Skills training, small business agents, education support, public service navigation.
12-Month KPI
1 training cohort; 10-20 fellows; 5 community organizations onboarded.
12-Month Activation Plan
A phased roadmap from preparation to scale readiness — each phase with concrete KPIs and funding requirements.
Prepare
Select 2 priority regions; finalize sponsor package; identify 3-5 pilot partners; build deployment checklist.
KPI: Partner shortlist, sponsor one-pager, KPI framework, pilot requirements complete.
Pilot
Deploy AgentOS 1.2 in 1-2 pilots; train local teams; launch first health/education templates.
KPI: 2 pilots live or in formal onboarding; 10+ local staff trained.
Optimize
Improve inference performance; expand templates; test offline agent installer.
KPI: 15-25% efficiency target; 8-12 templates; offline beta in testing.
Scale Readiness
Publish case studies; finalize train-the-trainer model; prepare year-two regional expansion.
KPI: 2-3 case studies; 1 regional installer cohort; year-two scale plan.
Use Case Requirements
Five core requirements define the technical contract of the DAO-OS — the standards the award-winning use case must meet.
REQ-01: The system must utilize adaptive model sharding to deploy computation-intensive LLMs across heterogeneous, localized edge devices without accuracy loss, circumventing traditional cloud reliance.
REQ-02: The architecture must support decentralized, sovereign AI deployment on modest on-premise hardware to ensure digital autonomy, data privacy, and economic sustainability.
REQ-03: The multi-agent system must incorporate swarm intelligence for collaborative model adaptation, allowing rapid optimization for specific community tasks in low-data regimes.
REQ-04: The inference engine must natively process 227 languages using Sparsely Gated Mixture of Agents/Experts to ensure high-fidelity cultural representation and linguistic equity.
REQ-05: The platform must implement autonomous agent communication protocols resilient to the edge-network-cloud continuum, enabling continuous offline execution during network outages.
The Innovate for Impact Process
From open call to global stage — how AI solutions are submitted, evaluated, and scaled for sustainable development.
Call for Use Cases
ITU invites governments, industry, academia, civil society, and innovators worldwide to submit AI use cases that address global challenges. Submissions span every region and sector, from healthcare to climate resilience.
AI for Good Scholars
A globally diverse cohort of AI for Good Scholars is selected to work closely with ITU. Scholars from developing countries receive funding to attend the AI for Good Global Summit in Geneva. They develop, analyze, and shepherd use cases toward presentation.
Technical Advisory Committee Review
A Technical Advisory Committee of global experts evaluates each submission against rigorous criteria: technical soundness, real-world impact potential, scalability, and alignment with sustainable development goals. Gender and regional balance are considered throughout.
Selection & Recognition
Outstanding use case authors are invited to present their innovations at the AI for Good Global Summit in Geneva and AI for Good Impact Africa in Johannesburg. Winners are celebrated for the power of innovation to address global challenges.
Publication & Scaling
Curated use cases are published in the annual Innovate for Impact Report and indexed in the AI Playbook, launched in 2026 to provide online access to 427 use cases. The resource helps governments, industry, and innovators access, share, and scale AI solutions for sustainable development.
2026 Use Case Domains
Eleven key domains anchor the Innovate for Impact Reports. These seven headline sectors highlight where AI is delivering measurable progress today.
Healthcare
AI diagnostics, remote triage, medical imaging analysis, and clinical decision support expanding access to quality care in underserved regions.
Education
Personalized learning pathways, AI tutors, language preservation, and adaptive curricula closing the educational divide across communities.
Climate Resilience
Early-warning systems, emissions monitoring, disaster prediction, and adaptive resource management strengthening community preparedness.
Agriculture
Precision farming, crop disease detection, yield forecasting, and supply-chain optimization improving food security and farmer livelihoods.
Finance
Credit scoring for the unbanked, fraud detection, micro-insurance, and inclusive digital payment systems broadening financial participation.
Digital Inclusion
Language translation, accessibility tooling, low-bandwidth AI, and community infrastructure connecting the next billion users.
Public Services
Citizen-service automation, policy modeling, transparent procurement, and evidence-based governance improving public sector delivery.
Where Winners Are Celebrated
Innovate for Impact workshops convene across three global hubs, bringing together governments, industry, academia, and innovators to exchange knowledge and showcase solutions.
Geneva, Switzerland
AI for Good Global Summit — 7-10 July 2026
Johannesburg, South Africa
AI for Good Impact Africa
Shanghai, China
Regional Innovate for Impact Workshop
Evaluation Criteria
Submissions are assessed by a Technical Advisory Committee of global experts against a rigorous, balanced rubric.
Technical soundness and scientific rigor of the AI approach
Demonstrated real-world impact and measurable outcomes
Scalability and potential for cross-region replication
Alignment with the UN Sustainable Development Goals
Inclusivity, equity, and responsible AI practices
Feasibility of deployment within the AI for Good framework
The AI for Good Scholars Program
Scholars represent talent from diverse corners of the world, brought together under the umbrella of the AI for Good Global Summit. They work closely with ITU to develop and analyze use cases and shepherd them toward featuring at the Summit in Geneva on 7-10 July 2026.
Scholars from developing countries receive funding to attend the Summit, and the program intentionally considers gender and regional balance to mentor international partners and broaden global participation in AI.
Building AI That Serves Everyone
Our recognition by the United Nations AI for Good program reflects a commitment to distributed, private, sovereign AI that closes the digital divide. Learn more about the company behind the award.