Version: 0.1 Draft
Date: October 2025
Initiator: André E. Karlsson
License: CC BY-SA 4.0
Status: Concept Specification (seeking collaborators and feedback)
ModelPub is an open protocol for federated, privacy-preserving artificial intelligence and distributed compute sharing.
It enables individuals and organizations to run their own AI nodes — learning from personal or local data — while securely sharing both knowledge and computational power across a decentralized network.
The goal is to make advanced AI accessible without central control or surveillance.
A network where data stays local, models collaborate, and compute is shared collectively.
Modern AI depends on massive centralized infrastructure. This creates barriers for smaller participants, restricts transparency, and locks innovation within corporate silos.
ModelPub challenges that model by proposing a federated ecosystem where participants pool knowledge and processing resources while maintaining full sovereignty over data and compute usage.
No single entity owns the model. No one must surrender their privacy or hardware to participate.
- Local First — Every node retains ownership of its data and compute capacity.
- Federated Learning — Knowledge is exchanged through encrypted model updates, not raw data.
- Federated Compute — Nodes can share CPU/GPU cycles under verifiable and ethical policies.
- Model Independence — Each node may use any model architecture as long as it speaks the ModelPub protocol.
- Open Standards — Fully transparent message schemas and protocols.
- Privacy by Design — Secure, consent-based participation.
- Ethical Federation — Respect for licenses, human consent, and sustainable resource sharing.
Each ModelPub node acts as both:
- A local AI instance — learning from user data, refining personal adapters.
- A compute contributor — donating or borrowing compute power from the federation.
Components:
Personal Vault— stores local data and model state (encrypted).Compute Pool— manages shared compute contributions and allocations.Federation Layer— communicates model and compute updates via the ModelPub protocol.Policy Engine— enforces consent, energy limits, and ethical restrictions.Adapter Manager— interfaces with local applications and devices.
Federation Design:
- Nodes may form trust domains or communities.
- Compute requests and model deltas are signed, rate-limited, and auditable.
- Privacy rules define what can be trained or computed remotely.
ModelPub is model-agnostic.
Each node may operate its own large language model or specialized AI system — open-source, closed-source, or custom-built — as long as it implements the ModelPub communication schema.
The protocol defines how nodes share knowledge and compute, not how they reason internally.
This ensures diversity, interoperability, and long-term resilience.
- A user on a low-power laptop trains a writing assistant using local notes.
- Their node requests short compute bursts from trusted peers with spare GPU capacity.
- Model deltas are aggregated across participants — improving everyone’s assistants collectively.
- All compute exchanges are signed, tracked, and reversible.
- Zero-trust architecture with signed messages and verifiable receipts.
- Encrypted compute envelopes — only the authorized model segment is shared.
- Local control policies for bandwidth, energy, and privacy budgets.
- Right to erasure — nodes can withdraw data and compute history at any time.
- Auditability — every exchange can be traced for transparency.
- Democratizes access to high-performance compute.
- Enables small actors to contribute meaningfully to AI ecosystems.
- Reduces dependency on corporate cloud providers.
- Builds collective intelligence through voluntary cooperation.
Beyond technical efficiency, federated AI can strengthen the social fabric of the digital world.
- Diversity of Voices — When anyone can host a node, knowledge emerges from many cultural, linguistic, and disciplinary perspectives — not only from global tech centers.
- Local Empowerment — Municipalities, libraries, universities, and small organizations can run their own nodes, training AI on contextually relevant data while contributing to global intelligence.
- Educational Access — Students and researchers can participate directly in real-world AI infrastructure without needing corporate APIs or expensive hardware.
- Resilience & Security — A distributed network is less vulnerable to single points of failure, censorship, or geopolitical disruption.
- Public Good — Shared compute and data turn AI into an open civic resource, rather than a private commodity.
Together, these qualities build a more pluralistic, adaptive, and secure foundation for the next generation of the internet.
While ModelPub defines the protocol for how nodes exchange knowledge and compute,
it does not prescribe how humans interact with their personal AI instances.
In the future, a user-facing layer could make this experience seamless —
allowing anyone to access their own ModelPub node from any device,
with responses powered by the shared compute network rather than local hardware alone.
This would make personal AI feel as fast and capable as centralized systems like ChatGPT,
but without requiring ownership of powerful machines or surrendering control of data.
A truly distributed, user-owned AI cloud.
Such an interface could also redefine digital participation itself:
- Digital Autonomy — Citizens can use AI as part of their personal infrastructure, not as customers of distant corporations.
- Educational Inclusion — Schools and universities could run low-cost ModelPub nodes, giving students direct access to AI systems they can study, adapt, and trust.
- Local Relevance — Communities could train and adapt AIs to reflect local languages, needs, and values, ensuring cultural and contextual diversity.
- Public Sector Adoption — Municipalities and public institutions could offer AI services transparently and securely, powered by shared compute from the federation.
- Sustainable Access — Shared compute reduces environmental and economic barriers, allowing more people to benefit from advanced AI capabilities.
In essence, ModelPub’s long-term vision is not just a network of machines —
but a network of communities, each shaping and benefiting from collective intelligence on their own terms.
All documentation and specifications are licensed under CC BY-SA 4.0.
ModelPub was initiated by an independent contributor driven by a belief in an open, free, and modern internet.
The project is not primarily a technological pursuit, but a vision for how artificial intelligence — as one of today’s defining forces — can evolve beyond corporate silos and serve the public good.
AI should be powerful, but also transparent, safe, and accessible to everyone.
By combining open collaboration, federated data, and shared compute power, ModelPub aims to make advanced AI a collective resource, not a private privilege.
I don’t come from the AI development side, but from the intersection where technology meets society,
data governance, and ethics. This idea grew out of a simple question:
How can we build AI that respects user autonomy and privacy?
I have no ambition to lead a project, but I want to contribute to early discussions —
about opportunities, risks, and possible paths forward.
This work is rooted in idealism — a belief that technology can reinforce autonomy, trust,
and shared progress — while staying grounded in practical experimentation and open dialogue with the community.
Contributions, corrections, and critical discussion are warmly welcome.
ModelPub is shared as an open concept — not a managed project.
It is meant as a foundation for discussion, experimentation, and collaboration among anyone interested in building a freer and more resilient digital ecosystem.
You are welcome to:
- explore or extend the protocol ideas,
- start your own implementation or fork,
- critique, question, and refine the vision,
- or simply use this document as inspiration for something new.
There is no central leadership or roadmap ownership.
The goal is simply to let the idea exist —
so that those who see value in it can carry it forward in their own way.
Join the Discussion:
Use tag #ModelPub on Fediverse (Mastodon, Lemmy, Kbin) or contribute via GitHub.