The bounded world model behind every KUON agent. A declared, typed, parser-enforced runtime where every capability is a command and every state change is reversible.
The Agentic Ontology is the bounded world model and command surface that every KUON agent reads, writes, and operates against. It is not a diagram of objects. It is the runtime that turns intent into controlled, auditable, reversible action.
Where most agent platforms treat ontology as a data layer — a graph of types and relationships behind a query language — the Agentic Ontology treats ontology as an operational one. Commands. State. Audit. Rollback. Approval. The grammar of agency, applied uniformly to every object in the world.
A declared, typed set of objects every agent can see and act on. Memory is history. The world model is the present — parser-enforced, replayable, scoped per agent, refusing anything the agent cannot name.
A small set of verbs every command supports. Same verbs, every object. The language refuses to skip a step — and that refusal is the point.
list, describe, plan, dry-run, validate.apply, status, logs, diff, rollback.Propose, authorize, audit, restore. The engine turns a command from a string into a controlled mutation — with a verifiable trail and a one-verb rollback when something goes wrong.
The Agentic Ontology is not a vision deck. The toolchain is the live runtime: every product KUON ships is the ontology, made concrete against a real workflow.
config.apply for OpenClaw.The world model is finite and declared. Every object the agent can read or write is named here, with a typed contract and an action surface. Nothing implicit. Nothing hidden.
A RapidClaw voice session is a sequence of ontology commands. The voice host reads, resolves, renders, and closes. The agent cannot invent a number because the ontology does not give it a command for inventing numbers — only for reading, resolving, and rendering the source of record.
session.create opens the bound voice session scoped to the named project.evm.status reads CPI and SPI from the Empower source of record — bottom line first.evm.drivers resolves the lowest-level WBS elements driving the variance.chart.render writes the native Empower chart to the watch pane.session.close releases the bound context. Every step logged; every state change reversible.$ session.create --project BOOMERANG session_id: sess_2bc14f8a scope: BOOMERANG bound_at: 2026-06-28T21:56:56-07:00 $ evm.status --project BOOMERANG --source empower cpi: 0.861 spi: 0.909 verdict: cost-inefficient, behind schedule $ evm.drivers --project BOOMERANG --metric cost --depth leaf 1. Load Propellant cpi 0.689 2. Develop Code cpi 0.821 $ chart.render --chart spi_cpi --project BOOMERANG → wrote chart to watch pane (native Empower) source_of_record: empower://BOOMERANG/wbs $ session.close --session sess_2bc14f8a state: clean rollback: not needed
Four systems are already on the Agentic Ontology. Each one is a thin slice of the bounded world model — visible in a dashboard, governed by the same verbs, reversible by design.
The bottom line first, then the variance. RapidClaw is the voice co-pilot for EVM Advisor — every answer is a command sequence against the ontology, traceable to the Empower source of record.
Open RapidClaw → Operations Kurosawa production dashboardA live OpenClaw-connected dashboard for AI production workflows, assets, rework, and orchestration visibility. The ontology's bounded world model, rendered for human review.
Open dashboard → Launch discipline Subagent SheepdogLaunch verification, watchdog behavior, milestone enforcement, and startup checks for sub-agents. The ontology's decision layer, applied to delegated work.
View skill → Change discipline Config SentinelOperational guardrails for OpenClaw config edits: snapshot, validate, rollback. The ontology's verb sequence, applied to the most dangerous command — config.apply.