The intelligence underneath

We build with UNI.

UNI is our work in active inference. The thesis is plain: a living thing, a family, a team, or a whole system gets by the same way. It predicts what is about to happen, senses what actually happens, and updates both its understanding and its surroundings to keep the gap between expectation and the world small. Perception and action are two halves of one loop.

It is cited, falsifiable, and real. You can read it, you can run it, and you can try to break it.

I did the math. Scarcity is an accounting, and what is manufactured can be dismantled.

Michael Polzin, founder

What UNI is

A lens, not a product.

UNI is the loop, taken seriously, as a way to look at a child, a family, a team, a company, or a whole system, and notice where the loop is being crushed and where it could be set free. Most of what looks like a hard limit is really a boundary no one has examined yet. We find those boundaries, and we move them.

It is a modeling and design lens, not a clinical or diagnostic tool. Nothing here is medical advice. What it gives you is a clearer way to see, and a way to check whether the seeing holds.

See it move

Five interactive labs.

The loop made visible. Each lab is the engine on a workbench, running entirely in your browser. Drag a dial and watch the behavior change. No math required to start.

Precision Lab

An active-inference agent you can steer. Move the dials for how much it trusts its senses, its actions, and its own plans, and watch its behavior change in front of you.

Echo Lab

The same agent, sensing the maze by echolocation instead of immediate walls. A clean way to see how perception shapes action.

Loop Lab

A two-state model for treatment-sequencing arguments. It shows where standard exposure succeeds, stalls, and fails, with a map of the bifurcation underneath.

Heart Lab

The loop carried into how families and teams hold one another. Where the gap closes, and where it gets crushed.

Cell Lab

The smallest version of the same idea, watched at the level of the cell. The loop, all the way down.

Open the five labs

The benchmark world

The Stratified Palimpsest.

Underneath the labs there is a harder thing: a world we built on purpose to be difficult, and an agent that has to earn its way through it with no shortcuts. The agent sees only opaque numbers, never the world. It is paid no reward. It is checked, every step, by a monitor that re-derives the evidence from scratch and fails loudly if anything leaked. We named it the Stratified Palimpsest because the world is written in layers, and each layer is only legible to a body that has grown the organ to read it.

A world made hard on purpose

A layered world the agent never sees directly. It couples through one opaque interface only, twenty-nine numeric channels, permuted per run, so the same channel means a different thing in every world. Deeper layers stay invisible until the body grows the sense that can read them. Nothing is handed over for free.

No reward, ever

There is no score, no return, no fitness signal on any path the learner can read. The agent acts to minimise expected free energy, an information-gain term plus a preference term, both computed from its own beliefs. We checked, in code, that no per-action reward can be smuggled in: clone an action and its value is identical to the part in a trillion.

Proven not to cheat

A separate monitor re-derives the whole interface from the run's seed and checks every observation: only finite numbers crossed, no world words, every channel traces to an organ the body actually had. Tamper with one value and the check fails and names the frame. You can run it yourself.

Behavioural and organisational measures are necessary-not-sufficient substrates with ZERO evidential weight for awareness, consciousness, or life on their own. Passing a gate demonstrates the named behaviour, never experience.

See the benchmark world

The paper

Read it. Try to break it.

The mathematics behind the labs is written down and published in the open. It is a preprint: unrefereed, with expert review still pending. We present it that way on purpose. The claims are falsifiable, and each lab documents what would challenge its own interpretations.

Behavioral labels in the labs are computational descriptions, never diagnoses. This is not clinical evidence, and we do not present it as such.

POMDP formulation: Parr, Pezzulo and Friston (2022), Active Inference: The Free Energy Principle in Mind, Brain, and Behavior.

Read the preprint

In the field

Where this sits.

AJ Maren's Themesis 2026 resource map for learning active inference lists SolutionWright among five pathways into the field. We are glad to be in that company. It is a categorization, not an endorsement. The link goes straight to her words, and you can judge the fit yourself.

If you want to go from reading about it to building it, the UNI workshop is the hands-on path: twelve weeks, you ship a model, and you get the gates to check the claim for yourself.

Read Maren's resource map The UNI workshop

The living work

Go straight to the source.

The full UNI site is open and live. Wander in, steer the agent yourself, and see how the loop looks from where you stand.

Explore UNI