Frontier models learn from millions of texts. We train AI on users’ actions in millions of apps and systems, in the concepts people understand.
Today’s interfaces and APIs are built around device parameters. A washer receives three numbers, while the person knows what the sweater is, what state it is in and why heat is a risk. CDS keeps that understanding in the system, in seven shared ontological categories, and records every action as an event.
Every word of a request takes its place in the seven categories. Before, action and after form one event record.
Wash, rinse, dry and clean share objects, means and properties. The space checks coherence and bounds meaning.
Every action, in every space, is recorded as an event in the same categories.
Concepts linked by real events. It predicts, carries knowledge across spaces and composes what was never seen.
The device API becomes an expression of the concepts, not a replacement for them. Four components close the loop.
Turns signs in any language into conceptual structure. Language stays at the edge.
Matches the request to this device’s functional space and checks it before anything runs.
Calls the device or system API under an active functional space, deterministically.
Records what actually happened as an event in the same categories.
What was meant and what happened are kept together in one event record. The difference between them is what the organic ontology learns from.
Once actions are modelled as concepts and recorded as events, the same layer serves as ontology, middleware and interface, and becomes training data for AI that learns from what people do.
Seven fixed categories, with links grown from real events. Each company’s functional spaces form a private ontology; together they converge toward a universal one.
Conceptual structure sits between every input and every system. One coherent, event-based API across a company’s landscape, with canonical addresses for every event.
Voice, text, touch, gesture and telemetry compile into the same structure, so interfaces can change without changing what the system understands. P-Phone is built with blind users in mind.
Conceptual State Machines are deterministic. Forbidden states are shown unreachable before release, and every decision reproduces exactly for audit.
No party exposes its data model. Shared categories and event addresses make an insurer, a carmaker and a city mutually interpretable, without one vendor owning the schema.
A third path beyond symbolic AI and LLM scaling: reasoning on concepts grounded in what actions do, so knowledge transfers between spaces and composes the unseen.