Intelligence That Belongs to You

The dominant AI industry has made a set of assumptions so pervasive they've become invisible. That intelligence must be centralized. That data must travel to the cloud to be useful. That a model trained on the entire internet, by a company headquartered in San Francisco, is the appropriate tool for a Diné woman in Crownpoint figuring out her next step in life.

We disagree with all of it.

HumaneFrame is built on a different set of assumptions — ones that begin not with what is technically convenient, but with what is humanly right. The result is a vision of augmented intelligence that is localized, relational, culturally grounded, and above all, owned by the communities it serves.

High Church and Low Church

There is a useful metaphor — articulated powerfully by Samuel Z. Alemayehu — from the history of Christianity that maps almost perfectly onto the current AI landscape.

The High Church builds cathedrals. Soaring, magnificent, centralized institutions that concentrate authority, require credentialed intermediaries, and demand that the faithful come to them. The architecture is awe-inspiring. It is also, by design, extractive — drawing people and resources toward the center.

The Low Church builds chapels. Locally owned, community-governed, close to the ground. The minister knows your name. The building belongs to the congregation. The theology is interpreted in light of lived experience, not handed down from a distant authority.

The current AI industry is the High Church. GPT-4, Gemini Ultra, Claude — these are cathedrals. They are extraordinary. They are also trained on humanity's data, monetized by a handful of corporations, and accessible only through API keys and subscription fees that flow revenue in one direction: away from the communities that need intelligence most.

HumaneFrame is building chapels.

Technically, this means what researchers at the University of Luxembourg discovered when they stopped sending sensitive institutional data to cloud APIs and started running fine-tuned open-weight models on-premise: local deployment is not a compromise. It is, in many cases, preferable — more private, more trusted, more aligned with the actual needs of the people it serves. The Luxembourg team called it "sovereign and fully private." We call it the Low Church, and we're building it at the chapter house level.

What "Edge" Actually Means

In technical parlance, "edge computing" means processing that happens close to where data is generated rather than in a centralized data center. In HumaneFrame terms, it means something more specific and more human: intelligence that lives in the community.

Today's hardware makes this genuinely possible. A device like NVIDIA's DGX Spark — roughly desktop-sized, powered by a Grace Blackwell Superchip — can run AI models up to 200 billion parameters with no internet dependency whatsoever. Sensitive data — health records, legal documents, cultural knowledge, a learner's personal history — never leaves the device. It sits on a desk in Window Rock. It processes in Chinle. The model that guides Alastair through her competency assessment doesn't need to call a server in Virginia to do it.

This is not science fiction. It is available now, and it changes everything about what's possible for sovereign communities.

But hardware alone is not the vision. The vision is what runs on that hardware — and this is where HumaneFrame's most original contribution lives.

The SNBH State Vector: When Philosophy Becomes Architecture

Most AI systems model the user as a set of preferences and a conversation history. This is, at best, a thin shadow of a human being.

HumaneFrame's co-founder Moroni Benally introduced us to a richer frame — one that has guided Diné life for generations. Sa'ah Naagháí Bik'eh Hózhóón (SNBH) is often translated as "walking in beauty" — an orientation toward harmony, balance, and right relationship with the world. It is not a metaphor. It is a precise description of what a flourishing human life consists of, and it refuses the western reduction of a person to their cognitive output.

Diné philosophy understands a person through four inseparable dimensions:

Western AI systems optimize for the first dimension and ignore the rest. A system that tells an exhausted, grieving, isolated person that their "learning pathway" recommends three more modules this week has not helped them. It has failed them while generating an activity metric.

HumaneFrame's agent, Saki, tracks all four dimensions as a continuous, living state vector. The goal of the system — formally, mathematically, architecturally — is Hózhó: holistic balance across all four. When cognitive load is overwhelming, Saki reduces friction. When the learner is coasting, Saki introduces productive challenge. When emotional load signals that this is not a moment for new content, the system knows to do something different. When relational connection is low — when Alastair is isolated — Saki acts to weave her back into community, not to deliver her next lesson.

This is the most original contribution in the field. No existing AI system uses an Indigenous epistemological framework as its formal, computable state space. The SNBH 4-quadrant vector is unprecedented in the AI literature, and it is what we are publishing openly under CC BY-SA — because a framework this important to human flourishing should not be owned by anyone, and certainly cannot in good conscience be patented.

Fine-Tuned, Not Generic

There is a second problem with cathedral AI beyond sovereignty: it is trained on everything, which means it is precisely calibrated for nothing.

A model trained on the entire internet reflects the biases, the majority assumptions, and the cultural defaults of whoever produced most of that internet. It knows approximately nothing about Navajo water rights law, Diné clan structure, the specific job competency requirements of the Navajo Housing Authority, or the labor market realities of a community with 50% unemployment and 2,000 vacant jobs sitting simultaneously unfilled.

HumaneFrame is building a Diné-aligned model — a fine-tuned, locally sovereign AI trained on Navajo Nation-specific data: tribal law, governance procedures, educational content, labor market intelligence, cultural knowledge. This is done using techniques like LoRA (Low-Rank Adaptation) and its distributed variant dLoRA, which allow a foundation model to be specialized without exposing training data to external systems. The tribal data stays sovereign. The model becomes, genuinely, Diné in its frame of reference.

The aggregate labor market and competency data that Saki learns from over time is not owned by HumaneFrame. It is owned by the Nation — in alignment with the CARE Principles for Indigenous Data Governance. This is not a policy choice — it is a design requirement, built into the institutional agreements from day one. Tribal data sovereignty is not window dressing. It is the architecture.

Augmented Intelligence, Not Artificial Replacement

We are deliberate about our language. HumaneFrame is a platform for augmented intelligence — the amplification of human capability, not its replacement. This distinction matters enormously, and it is not semantic.

The researchers Evans, Bratton, and Agüera y Arcas have argued that the next intelligence explosion will not be a single superintelligent AI — it will be the emergence of new forms of socially aggregated cognition, "centaur" configurations of humans and AI working together. What's striking in their research is that frontier reasoning models, when optimized for accuracy through reinforcement learning, spontaneously develop something like internal multi-agent debate — a "society of thought." Robust reasoning, it turns out, is inherently social, even within a single system.

This maps precisely onto the Diné understanding of what a person is: not an isolated cognitive unit, but a node in a web of relationships. Augmented intelligence in the HumaneFrame sense means giving Alastair access to tools that make her more herself — more capable of navigating the institutions that affect her life, more able to articulate and demonstrate what she actually knows, more connected to the community around her.

Phil Komarny captures the practical version of this with his concept of "scaling your talents." Thirty years of expertise in a domain, paired with AI tools, doesn't replace the expertise — it makes it dramatically more productive. For Navajo professionals with deep knowledge of water rights, land management, healthcare, governance, and education, AI doesn't compete with that knowledge. It amplifies it. The bottleneck was never the knowing. It was the building.

The Generative Friction Constraint: Why Saki Won't Just Tell You the Answer

One of the most counterintuitive design choices in HumaneFrame is what we call the Generative Friction Constraint (GFC). Saki is architecturally prohibited from simply answering questions that the learner is capable of working through themselves.

This is not cruelty. It is learning science. Decades of cognitive research — desirable difficulties, retrieval practice, interleaving — confirm that making learning feel harder in the moment produces dramatically better long-term retention and transfer. The Socratic method is not a historical curiosity. It is the most effective pedagogical technology ever invented.

But more than that, the GFC reflects a deeper philosophical commitment: a system that simply gives you answers is training dependency, not capability. It is the banking model of education that Paulo Freire spent his life dismantling — depositing knowledge into a passive vessel. HumaneFrame is not a search engine. Saki is a coach. She asks the question that makes you find the answer, and the record of that discovery belongs to you, permanently, as verifiable evidence of what you can actually do.

When Saki does answer directly — when the learner is genuinely stuck, when the emotional state signals overwhelm rather than coasting — that too is a calibrated choice, not a default behavior. The system always knows which mode it's in and why.

The Low Church Deployed: Chapter Houses as AI Nodes

The Navajo Nation's chapter house system is one of the most elegant pieces of civic infrastructure in the United States. There are 110 chapter houses across the reservation — local governance units where community members meet, make decisions, and support one another. They are trusted. They are already there. They are, in the HumaneFrame vision, the natural deployment site for localized AI.

A chapter house running a sovereign AI node doesn't need broadband connectivity to serve its community. It needs power, a local model fine-tuned on Navajo-specific data, and a human Weaver — the HumaneFrame term for a community navigator who bridges the AI system and the person in front of them. The Weaver Protocol is the human layer that the technology cannot replace: the person who knows Rosa Tso, who understands that she hasn't come in this week because her grandmother is ill, who can say "Saki can help you with the job application when you're ready, and here's what the chapter can offer right now."

This is the vision. Not a cathedral. A hundred and ten chapels, each one owned by the people who gather there, each one running intelligence that was shaped by and for this specific community, each one staffed by a human being whose job it is to make sure the technology serves the person and never the reverse.

What We're Publishing, and Why

HumaneFrame's technical specifications — the SNBH 4-quadrant state vector, the Generative Friction Constraint, the Weaver Protocol three-layer architecture, the formal behavioral model for Saki — are being published openly under CC BY-SA 4.0 license. Anyone can read them, implement them, build on them.

This is a deliberate choice, and it reflects the deepest value of the project: you cannot in good conscience apply Diné philosophy as a proprietary algorithm. The framework belongs to the tradition it comes from. Our competitive advantage is not the specification. It is the relationships, the institutional trust, the Nation data, and the years of on-the-ground work that cannot be published into existence by anyone who reads a white paper.

The High Church patents its liturgy. The Low Church shares it, because the point was never the liturgy. The point was the community.

References & Further Reading