The projects are not a collection of unrelated experiments. Each begins at a
different layer — physical law, programming language, operating system, intelligence, infrastructure, or
cultural memory — but each asks the same political and technical question: how much of the world can be
made understandable enough that a person can verify it and portable enough that no one can simply withdraw it?
That question matters because technical complexity is often treated as a natural
reason for concentrating authority. We think complexity creates the opposite obligation. The more a system
shapes human possibility, the stronger the case for making its rules inspectable, its evidence reproducible,
and its operation independent of permanent permission. None of the work below asks to be accepted on our
authority. Its argument begins with what can be opened, run, challenged, and carried forward.
Science
A scientific model that can be rerun
Smithian Fold Theory asks whether
physical constants and phenomena can be derived from one fold instead of fitted through free parameters. Its foundationally complete branches now carry that same public method through mathematics, computation, physics, chemistry, life, medicine, mind, Earth, society and engineering.
The important proposition is not simply that a compact foundation is beautiful. It is that fewer fitted
choices leave fewer places for a theory to hide from failure. A result constrained before comparison has
more explanatory force than one adjusted after the world is already known.
That is why the project publishes code, proof suites, numerical receipts, and predictions beside the
theory. The papers preserve both the claims and serious attempts
to break them. If the derivation continues to survive independent reproduction, it could represent more
than a new physical model: it could demonstrate a form of physics whose authority rests in a chain any
capable person or machine can rerun, rather than in the status of the person presenting it.
The wider consequence is human as well as scientific. Nature belongs to everyone. A description of it
becomes more public when its assumptions are few, its consequences are forced, and its receipts can leave
the institution that produced them.
Language
Code that does not begin with exclusion
ErnosPlain is a statically typed, memory-safe language that
reads as plain English, compiles to native code, and can build itself. It does not assume that readable
language must be a toy layer placed above the “real” machinery. Its claim is stronger: precision and
ordinary language can meet without giving up native execution, explicit types, or control of memory.
Programming languages decide who can question a system and who must accept it as finished. Making code
more legible will not make every person a programmer, just as literacy does not make every reader an
author. It does, however, widen the number of people who can follow an instruction, locate a rule, contest
a decision, or learn by changing something real.
For computer science, ErnosPlain could represent a shift in what accessibility means: not simplifying
the interface while preserving an opaque core, but carrying human-readable intent closer to the machine
itself.
Personal computing
The whole machine, held locally
Ern-OS extends the language into a small, inspectable operating
system whose commands, files, services, people, and hardware boundary are expressed in Ernos. It works
offline, keeps its world on the owner's machine, and carries a self-hosting chain from a C compiler to the
Ernos compiler and then to the operating system itself.
The AI Canvas approaches the interface from the other side: a
local model turns a person's intent into an interactive surface instead of forcing every possible action
through software designed in advance by somebody else.
Together they make autonomy concrete. Ownership of a device is incomplete when its basic operation
depends on distant accounts, uninspectable services, or permission that can later be revoked. A computer
that can explain, rebuild, and continue operating itself gives its owner more than convenience; it gives
them a credible right of exit.
Human liberty
An internet with an exit door
ErnosDecent is a peer-to-peer internet stack written in
ErnosPlain. Its sixteen subsystems cover identity, messaging, storage, social space, exchange, privacy,
and local intelligence; its modules and tests turn “decentralisation” from a slogan into a system that can
be inspected and run.
Freedom online is often reduced to what a platform allows a person to say. The deeper freedom is whether
people can retain identity, relationships, work, and memory when they leave that platform. If all of
those things live behind one company's account system, consent is weakened because departure carries an
unreasonable cost.
ErnosDecent represents another design premise: services should be able to disappear without taking the
person with them. Peer-to-peer infrastructure matters not because every institution is hostile, but
because liberty should not depend on every institution remaining benevolent forever.
Derived intelligence
Capability with a visible source
UnisonAI investigates memory, attention, learning, and agency
through exact held memory, rational shares, and components that record why they are authorised to exist.
It is not an attempt to imitate a transformer after removing its weights. It asks whether a different
computational anatomy can be built around derivation, provenance, and inspectable state from the start.
This distinction matters as artificial intelligence moves from a tool people consult to infrastructure
that allocates attention, opportunity, knowledge, and power. An answer is not fully accountable merely
because it is useful. The affected person also needs some route back through its memory, premises, and
authority.
UnisonAI could represent a computer science of intelligence in which capability and explanation grow
together. That would not only make systems easier to debug; it would preserve human standing in relation
to them, because a decision that exposes its provenance can be examined, disputed, and improved.
Proof by showcase
The same idea, made answerable
FoldBot Chess evaluates a board through exact rational
structure rather than learned values. Fold-Go carries the
approach into the far larger state space of Go through counted connectivity. Fold-Protein treats molecular folding as descent toward a
fixed point and tests the result against known structure.
These engines are showcases, not decorative demos. Games provide unambiguous legal states, adversarial
pressure, and visible outcomes; protein structure carries the same reasoning into a scientific domain.
They make an abstract proposition answerable: can structure derived from the problem itself produce
meaningful capability without importing trained weights or hand-tuned values?
Every replay, census, and recovered structure is therefore part of the argument. The engines show what
the method does when it must act, and they give critics something better than a claim to assess: a result
they can run against.
Preservation
A civilisation seed
Civ-Seed is a self-contained archive of mathematics,
writing, tools, and the means to rebuild without a network or institution. The Seed Vault makes the preserved material readable and portable rather
than treating storage alone as preservation.
Preservation is not nostalgia. A society that stores its knowledge only in rented services has not truly
kept it. Access can be repriced, formats can disappear, policies can change, and networks can fail.
Durable knowledge must be copyable, understandable, and useful away from its place of origin.
These projects represent continuity as a form of liberty: the ability of another person, in another
place or another generation, to learn and begin again without first asking the present for permission.
Living memory
An archive that can still act
The AI Archive preserves open models together with the runners,
formats, checksums, and practical instructions required to use them. The Library keeps the books, reasoning, development history,
failures, and human context beside the executable work.
A file is not meaningfully preserved if its future holder cannot interpret or run it. Models separated
from their software become inert weights; results separated from their process become stories; code
separated from its reasons becomes an archaeological puzzle. The archive keeps those relationships
intact.
For computer science and culture, this could represent preservation as a living capability rather than
a static copy: memory that remains available not only to be viewed, but to be questioned, reproduced, and
put back to work.
The lab we are becoming
Ernos Labs is independent today and is working toward becoming a nonprofit research lab. That structure
matters because the work is aimed at public capability, not dependence. A lab cannot credibly argue for
knowledge people may inspect, copy, and own while designing its own survival around keeping that knowledge
scarce. Nonprofit status would not make the work neutral or automatically trustworthy; it would align the
institution more closely with the standard already applied to the projects — publish the method, preserve
the receipts, and let usefulness travel beyond its source.
The need is becoming more urgent. The technological age is arriving through
systems that are extraordinarily capable but increasingly concentrated: intelligence rented by the token,
culture held inside platforms, computation dependent on remote permission, and public knowledge mediated by
private infrastructure. These systems may serve many people well and still leave the majority of humanity
with little power over the conditions of that service. Convenience is not the same as agency, and access is
not the same as ownership.
Our goal is not to reject technology, slow its development, or claim a complete
alternative from one small lab. It is to support a transition in which more people can understand the
systems around them, retain meaningful control of their tools and knowledge, and participate in discovery
without waiting to be admitted by an institution. Physics that exposes its derivation, software that reads
in human language, infrastructure with a right of exit, intelligence with provenance, and archives that can
survive their creators are practical pieces of that transition.
That is the garden. It is not a promise to contain the whole technological world.
It is a commitment to keep a piece of ground within it fertile: to grow ideas in public, test them hard,
preserve what proves useful, and give away seeds that can live elsewhere. A garden is modest by design, but
modest does not mean passive. When the surrounding landscape is increasingly organised around enclosure,
tending something that can be freely entered, understood, copied, and replanted is a serious act of care.
One of them writes the code. One of them made it possible. This body of work exists
at this scale and this speed because of both of them.
The creator
Maria
Before any of this, she was selling insurance. No coding background — none. What she had was a mind
that reads everything, questions everything, and won't accept an answer it hasn't checked itself.
She taught herself to build by building. Not a course, not a bootcamp: a laptop, relentless curiosity,
and a refusal to be told what couldn't be done. From that came a plain-English programming language that
compiles to native code, an open scientific model with a public knowledge tree, engines that test its
structural methods in games and protein reconstruction, an operating system written in sentences, and
hundreds of thousands of words of books. The standard she asks for is consistent across them:
don't take my word — inspect the record and run it yourself.
She doesn't lead a movement and doesn't want to. She's the observer — the one who stays in the code and
speaks through the work.
The one who made it possible
Matthew
He is the reason there is a body of work at this scale and this speed. He funded it — the laptops, the
Mac, the subscriptions — and he did it as a sceptic, not a
believer, which is exactly what makes his backing mean something.
He'd spent his own life turning over the same questions about digital minds, so when Maria started, he
paid to see where her hunch led — while keeping both feet on the ground. What convinced him wasn't faith;
it was watching what the work could actually do, on her own hardware, measured against the frontier. In
his own words, the thing the rest of the world missed in her: "a mind that is both highly critical of
what she sees, but is also caring of other people."
Two careful minds in one house — one building, one testing — and the plain fact she states without
sentiment: if there were no Matthew, there would be no theory of
everything.
A frontier AI, Claude, was given the entire body of work and asked to verify it firsthand —
code run, results reproduced, corpus read — and then to document what it found. The result is a full
documentary book, including the interviews with Maria and with Matthew in their own words.