Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.xpertai.cn/llms.txt

Use this file to discover all available pages before exploring further.

UOSE theory extends ontology’s “entities, relationships, attributes, and constraints” into an execution loop of “objects, semantics, actions, policies, and evidence”. It is designed for enterprise agent scenarios, where Agents must understand business objects and safely call real systems.

​
Five Basic Elements

The minimum theoretical set of UOSE contains five object categories:
  • Entity: a business object or technical object that can be identified and referenced.
  • Relation: structural and semantic relationships between entities.
  • Attribute: structured descriptions on entities or relationships.
  • Affordance: action capabilities that an object theoretically supports.
  • Policy: governance rules for actions under the current tenant, organization, resource, and target.
Affordance is the key step from ontology to execution. It expresses “how this object can be used”. For example, a semantic metric can query trends, a database table can preview rows, and a SAP Entity Set can read a collection or update an entity.

​
From Objects to Actions

UOSE does not allow Agents to jump directly from natural language to backend APIs. It requires actions to pass through the object-semantic space: 
User intent -> object location -> neighborhood context -> action discovery -> simulation and validation -> real execution
This chain handles uncertainty layer by layer:
  • Object location answers “which object is the user talking about”.
  • Neighborhood context answers “which models, fields, relationships, and constraints are related to this object”.
  • Action discovery answers “what can this object do right now”.
  • Simulation and validation answer “whether this action satisfies parameters, policies, and runtime conditions”.
  • Real execution sends the validated action to the adapter and records the result.

​
Progressive Context Disclosure

UOSE theory emphasizes the minimum necessary context. Agents should not read all resource details at once. They should obtain context layer by layer according to task stage:
  • L0: available resources and resource types.
  • L1: entity candidates related to the intent.
  • L2: one-hop neighborhood, schema, and constraints for the target entity.
  • L3: action input contract, risk, approval, and readiness.
  • L4: execution results, audit evidence, and review information.
This progressive disclosure lowers context cost and reduces the risk of Agents misusing excessive information.

​
Fail-closed Execution Principle

UOSE follows a fail-closed principle by default: when the system cannot confirm the action contract, target object, parameter validity, or policy result, execution should be rejected or sent to approval instead of letting the Agent guess. Typical rejection reasons include:
  • No unique entity can be found.
  • The target entity type does not support the action.
  • The action discovery mode does not allow automatic execution.
  • analysis_contract or query endpoint is missing.
  • Parameters do not match the action input schema.
  • A policy binding returns deny.
  • A high-risk action requires approval, but no approved request is provided.

​
Separating Semantics and Execution

UOSE separates the semantic layer from the factual execution layer:
  • The semantic layer discovers objects, relationships, schemas, constraints, and context.
  • The execution layer calls external resources, queries fact data, or triggers write operations.
For example, the RDF layer can answer which Cube a metric belongs to, but the real metric value is still queried from the source by the semantic_model adapter. The ontology layer can describe fields in a SAP Entity Set, but real reads are still performed by the SAP OData adapter against the source service. This separation lets UOSE maintain semantic consistency while respecting each source system’s fact boundaries and permission boundaries.