ONONO ENGINEProgressive AI Architecture
A visual journey

Step inside
our architecture.

ONONO Engine is a self-organizing architecture for recursive self-improvement (RSI), built on Meta-Ontology.

Click the glowing sphere to begin
Or scroll to see why we’re building it

A Digital Mind

Imagine AI that does not just respond, but understands.

Understanding is built over time. It compounds through context, memory, feedback, and action. ONONO is not another assistant layer. It is the foundation for a digital mind: adaptive, explainable, and always in human hands.

The ONONO Engine

A new intelligence layer needs a new architecture.

ONONO combines a multi-dimensional semantic network with an autopoietic runtime. Its modules maintain, extend, and optimize the network continuously. This is where a living mind ontology begins to emerge.

The Runtime Modules

Designed for performance, built for scale.

The runtime is composed of specialized modules that coordinate like a swarm intelligence. This distributed structure enables ONONO to grow in capability without collapsing into complexity.

Creating Understanding

Understanding is extracted from intent, context, and action.

Every module processes signals across who, why, when, intent, constraints, and outcomes. The result is not a flat answer, but a deeper model of meaning that can improve with every interaction.

Progressive AI

The next AI market will be built on progressive intelligence.

ONONO turns accumulated understanding into a strategic asset. As the system learns, a new form of augmented intelligence emerges: one that strengthens human judgment instead of replacing it.

Intelligence Augmentation

Human-grade intelligence. Always under human control.

For every decision that matters, ONONO keeps intelligence close to the people responsible for it. A new category is forming: digital minds that scale organizations without taking agency away from humans.

Why we are building Engine

Finite compute.Infinite complexity.

01Understand
02Choose
03Act
04Learn

One continuous loopEvery result informs
the next decision.

There is no end to what the world asks of us. There is a limit to the time and compute we can spend.

The challenge is to decide what matters, what is worth learning, and which action to take next within those limits.

02 / APPROACH

I do,
therefore I am.

Facio ergo sum.

ONONO’s approach starts with two inputs that must stay current.

Maintain values and goals.

Keep track of what matters, what the user and system are trying to achieve, and the principles and constraints that guide each choice.

Know the resources and costs.

Track available time and compute, estimate the cost of learning and acting, and account for uncertainty before committing resources.

Together, these guide autonomous planning: what to learn, which tasks to execute, and how much to invest. Plans must adapt as understanding, goals, and budgets change.

Technical depth From understanding to coordinated action

The semantic network represents goals, values, evidence, relationships, and uncertainty. Meta-ontology adds structures for examining how that knowledge is organized and interpreted.

The modular runtime provides a foundation for coordinating work around changes in this network. We are developing the planning and learning capabilities that turn these representations into sustained autonomous action within resource and permission limits.

“Facio ergo sum” draws on Chinese philosopher Zhao Tingyang’s account of existence through action.

03 / ADAPTATION

When conditions change,
the plan changes.

Learn, execute, or reconsider.
Then decide what to commit.

New evidence may call for more learning. A changed goal may reorder or cancel tasks. A tighter budget may require a smaller scope or a different method.

At each checkpoint, compare what matters now with results, remaining work, and available resources. Then revise the next action and its allocation before committing more.

Illustrative planning walkthroughHigh confidence

One budget.
A plan that keeps adapting.

Try the walkthroughSee what happens at the next step.

Budget cap100units
Spent0
Free reserve10
AllocationPlanned / spent
Evidence36 / 0
Synthesis34 / 0
Verification20 / 0
Uncommitted10 / 0
Checkpoint 1 of 4

Estimate from experience

Similar work supports a 90-unit estimate, with 10 units held for uncertainty.

Next plan

Gather evidence, then synthesize and verify. Keep the reserve uncommitted.

Within budget
Planning logic Estimate, commit, observe, revise

The walkthrough shows the planning behavior we are developing. Before each commitment, the plan should account for resources already spent, remaining work, and uncertainty.

Familiar work can start from prior cost estimates. Unfamiliar work needs a bounded pilot to test cost, quality, and complexity before a larger commitment. Results can change the method, task order, scope, or allocation.

If the remaining work cannot fit within the agreed constraints, pause new commitments and ask the user to revise the scope, strategy, or budget.

04 / DEVELOPMENT

Experience improves
the next plan.

Each result should improve the next judgment of importance, the next cost estimate, and the method used to plan.

Compare expected value, actual cost, and outcome. Use the difference to reassess priorities, calibrate estimates, and test better ways to learn and execute.

ActEvaluateImprove the next plan

Engine’s semantic network and modular runtime form the architectural foundation. Sustained autonomous planning, learning, and recursive self-improvement are still being developed and validated.

Technical depth Recursive self-improvement & evaluation

Recursive self-improvement (RSI) extends improvement to the methods by which the system learns and improves itself. The architectural scope includes knowledge structures, reasoning and planning strategies, resource allocation, and execution modules or code.

Proposed changes need to be evaluated against goals, value standards, resource costs, and observed results, then retained or rolled back.

The Gödel machine is a formal reference for self-modification backed by proofs of usefulness. Engine’s architectural direction does not imply those formal proof guarantees.

A little more context

Questions,
answered.

01

How does Engine relate to large language models?

Engine’s semantic network and runtime organize persistent understanding and ongoing work. Large language models supply capabilities such as analysis and generation within that architecture.

The architecture is independent of any one model, allowing different models and tools to serve different steps of a plan.

02

How do memory and agency fit together?

Experience needs context: what happened, why it mattered, what it cost, and what should change next. The semantic network provides a structure for these relationships, including uncertainty and conflicting evidence.

Agency is the developing ability to use that understanding to maintain goals, plan learning and work, and revise the plan within available resources and permissions. Users can correct priorities and redirect the work.

03

What is in place, and what is still developing?

The semantic network and modular runtime form the architectural foundation for representing understanding and coordinating work. Sustained autonomous planning, learning, and recursive self-improvement are still being developed and validated. The budget walkthrough illustrates the intended planning logic.

Self-improvement may extend from knowledge and cost estimates to planning strategies, learning methods, and execution modules or code. Any such change depends on available capabilities and permissions, and needs evaluation against goals, resource costs, and results.

Understanding is the foundation.
Purposeful action is the point.

Explore the research