— The problem
Today's AI stack is five databases wearing a trenchcoat.
A vector store for similarity. A search cluster for words. A graph for relationships. A triple store for meaning. A warehouse underneath it all.
Every one of them holds a copy of your data. Every copy drifts. Every seam between them is a place where your AI can be confidently wrong — and nobody can tell you why.
The ancients had this problem too. They solved it with one instrument.
Pillar I The engine
One engine. Six ways to ask it anything.
An armillary sphere carries many rings, but it measures one sky. Each ring is a different way of fixing the same point. Take a ring away and you don't get a smaller sky — you get a blind spot.
Ring 01 · Graph
“How is this connected to that?”
Relationships aren't something you compute at question time. They are the data. Follow them as far as the question goes — through one hop or twenty — and the cost follows the answer, not the size of the database.
Ring 02 · Vector
“What else is like this?”
Find things by what they mean, not by the words they happen to use. Because similarity lives in the same engine as everything else, a search by meaning can be narrowed by fact in the same breath — no round trip to a separate service, no second copy to keep in step.
Ring 03 · Ontology
“What kind of thing is this?”
Say once that a cardiologist is a doctor, and a doctor is a person. Then ask for people — and get cardiologists. The knowledge lives in the data itself, not in every query you write, and not in a prompt you hope the model reads carefully.
This is the onto in Onto AI.
Ring 04 · Spacetime
“How do these relate?”
Everything that connects, connects in one of four ways. Four families instead of ten thousand improvised labels — a shape small enough for a machine to actually reason about.
- Nearnessthis is like that
- Causethis leads to that
- Containmentthis holds that
- Expressionthis is a property of that
Ring 05 · Full-text
“Where were these exact words?”
Sometimes you need the phrase, not the vibe. Real search across the same data, in the same place — and it never quietly falls back to guessing when it can't find what you asked for.
Ring 06 · Federation
“What about the data we don't hold?”
Reach into the stores you don't own, while the question is still being asked, and fold what comes back into the same answer. Your graph doesn't have to stop at your own walls.
— The shift
Retrieval was only ever
the first letter.
Every AI product today is built on three letters: RAG. It was a good idea in a hurry, and it has carried the industry further than it should have. But the R is doing all the work, and the R is the weakest part.
-
RAG
Retrieval-Augmented Generation
Chop the documents into fragments. Find the fragments that look statistically similar to the question. Paste them into the prompt. Hope.
One ring, turned alone. Similarity is the only question it knows how to ask, so similarity is the only answer it can give.
Fragments that look similar → the model guesses the structure → an answer you can't check
-
OAG
Ontology-Augmented Generation
Stop handing the model look-alike paragraphs. Hand it meaning that is already structured — what a thing is, what it belongs to, what it causes, what contains it.
Two rings. Now it can reason instead of resemble. It walks the relationships rather than inferring them from prose, and it can show its working.
Structure that is true → the model reasons over it → an answer that cites its source
-
UAG
Unified-Augmented Generation
Every ring turning at once. The context isn't retrieved — it's assembled.
- From the data itself
- From the connections between it
- From what the ontology infers that nobody ever wrote down
- From how things relate in space and time
- From the exact words, when the exact words matter
- From the stores you don't even own
Everything the question touches → assembled, not retrieved → an answer with its reasoning attached
Here is the part that matters, and it's the same point this page opened with. You cannot assemble a unified context out of five separate systems. You can only stitch one together — five round trips, five copies that disagree, and a seam at every join where the answer comes apart.
UAG isn't a clever technique layered on top of the engine. It's simply what becomes possible once the engine is one thing.
An LLM that cites is an LLM you can trust.
Pillar II The workflow
A smart workflow for developing.
The engine was written from scratch in months, not years. The products on top of it took less time still. That can't be heroics — heroics don't repeat, and this has repeated six times. It's a way of working.
-
01
A fleet, directed
A small team commands a fleet of AI agents. People hold the architecture and the judgement — the decisions that are expensive to get wrong. The fleet carries the volume. One mind can drive far more hands than it used to.
-
02
The foundation does the repeating
Most of a new product exists before it starts: the data model, the permissions, the audit trail, the admin screens, the assistants. What's left to build is the part that is actually new — which is the only part worth a person's attention anyway.
-
03
Every product hardens the next
A fix in the foundation is a fix in everything standing on it. Six products don't cost six times one. They cost progressively less, because each one leaves the floor firmer than it found it.
Evidence Six products. One engine. Under a year.
The engine was written first. Everything below was built on top of it, in domains with nothing to do with one another — which is the whole point. A foundation that only supports the thing it was written for isn't a foundation.
- A professional certification platformPublic site, student and trainer portals, and a back-office of thirty-seven screens. In commercial production.
- A private-wealth trust systemTrust administration, a read-only partner portal, and three assistants that answer from the graph and cite it.
- A personal knowledge appFree text in, structured graph out. iPhone and Android.
- A classical sky atlasThe full knowledge graph on the device. iPhone.
- A photo-scan companion appiPhone and Android.
- A desktop translatorMenu bar, macOS.
Three unrelated domains — education, private wealth, consumer. One substrate underneath all of them. That isn't a demo. That's the evidence it generalises.
Pillar III The architecture
A unified architecture for delivery.
Three layers, and nothing exotic in any of them. The reason new products arrive quickly isn't heroics — it's that nothing underneath them has to be rebuilt to make the next one.
- Web
- Back-office
- AI agents
- iOS
- Android
Built on the two layers below. Not integrated with them — grown out of them.
- UAG
- Reasoning
- Citation
- Agent tooling
- Any model
Where context gets assembled and answers get their receipts. Model-agnostic by design — swap the model, keep the product.
- Graph
- Ontology
- Vector
- Full-text
- Spacetime
One unified engine, speaking an open international standard. Not a proprietary dialect — nothing here locks you in.
- StableOne engine, one copy of the truth. Nothing to sync, so nothing to drift.
- ScalableFrom a phone in a pocket to a cluster in a rack. Same engine, same questions.
- ReasonableLiterally: it can reason. Over structure that's true, not text that looks close.
- TraceableEvery answer can point at the fact it came from, and the fact can point back.
— What you build on it
Five surfaces. One substrate.
The same foundation reaches all the way from a browser tab to a phone that's in aeroplane mode. These aren't roadmap items — each one is carrying a product that already shipped.
-
01
Web
Public sites, portals, dashboards — talking to the engine directly, in whichever language the team already writes.
Shipped A professional certification platform, in commercial production.
-
02
Back-office
The unglamorous half that decides whether a system is real: permissions down to a single field, every change recorded and nothing quietly overwritten.
Shipped Thirty-seven admin screens on one platform; a trust administration system on another.
-
03
AI agents
Assistants that answer from the graph and cite it — and that refuse, deterministically, before a question they shouldn't answer ever reaches the model.
Shipped Three assistants in a private-wealth system, each scoped to what its reader is allowed to know.
-
04
iOS
The whole database runs inside the app. Not a cache, not a sync client — the engine itself, on the device. No server to call, no round trip to wait for, no cloud bill that grows with your users. It works on a plane.
Shipped A classical sky atlas carrying its entire knowledge graph on the phone.
-
05
Android
The same products, the same graph, the same questions — reaching the engine over the network.
Shipped Two consumer apps built from one shared codebase.
— Use cases
Three things people come to us for.
Different buyers, different problems — and underneath each one, the same three pillars doing the work.
-
Use case 01
AI agents
Assistants that answer from the graph and cite it — each scoped to what its reader is allowed to know, and refusing a question they shouldn't answer before it ever reaches the model. Grounded, governed, and able to show their working.
-
Use case 02
Second brains
Free text in, structured graph out. The things a person carries — people, places, moments, the connections between them — held as a graph they own, on their own phone, private by construction rather than by promise.
-
Use case 03
Unified data services
One service, not five behind one API. Vector, full-text, connection, geo and ontology reasoning all answering from the same data, with AI integration built in — at speed, and at scale.
The instrument is built.
The question is what you point it at.
Onto AI — a super engine for AI solutions, a smart workflow for developing, a unified architecture for delivery. A small team, directing a fleet of agents.
Contact z@funtg.com v@funtg.com