Skip to content

object-graph-semantic-layer.md

Latest commit

 

History

History
70 lines (51 loc) · 3.57 KB

object-graph-semantic-layer.md

File metadata and controls

70 lines (51 loc) · 3.57 KB
Raw

Object Graph Semantic Layer

中文：对象图语义层

UModel is the object graph semantic layer inside a vendor-neutral semantic runtime for enterprise AI, data governance, and operational intelligence. It does not replace data platforms, telemetry collectors, metric stores, tracing systems, Kubernetes, Prometheus, OpenTelemetry, CMDB systems, or AI tools. It gives them a shared object vocabulary, relationship semantics, and graph-shaped context surface that people, services, and AI agents can query.

Problem

Enterprise systems already produce and store a lot of data:

  • Business systems describe customers, orders, tickets, assets, and processes.
  • Data platforms describe tables, fields, metrics, ownership, and lineage.
  • Observability systems describe metrics, logs, traces, events, profiles, and runbooks.
  • CMDB, cloud APIs, and Kubernetes describe resources and workload state.
  • AI applications need trustworthy context before they analyze, predict, or act.

The missing layer: semantic alignment. Raw data exists, but core enterprise questions stay fragmented:

  • What business or operational object does this data describe?
  • Which objects are related?
  • Which fields, metrics, storage, query, and topology definitions explain that relationship?
  • What safe context can an AI agent read before acting?

UModel's Role

UModel models enterprise context as a workspace-scoped object graph:

  • EntitySet defines a class of business or operational objects, such as services, instances, operations, databases, assets, and external dependencies.
  • DataSet types define structured datasets and telemetry datasets, such as metrics, logs, traces, events, profiles, and runbooks.
  • Storage types describe where data lives.
  • Link types connect entities, datasets, and storage.
  • Entity and relation records provide runtime graph data.
  • Query Service exposes .umodel, .entity, and .topo as one read surface.

UModel Contribution

Layer Existing systems UModel contribution
Enterprise data Data warehouses, data catalogs, business APIs Gives datasets, fields, metrics, ownership, and lineage shared semantic anchors.
Telemetry and operations OpenTelemetry, agents, logs, metrics, traces Maps operational signals to modeled objects and relations.
Runtime resources Kubernetes, cloud APIs, CMDB Provides stable entity and relation semantics.
Query and exploration SLS, Prometheus, trace stores, graph stores Offers one Query Service for model, entity, and topology reads.
Agent context MCP clients and AI agents Exposes safe resources, query templates, and read-only tools by default.

Design Principles

  • Workspace first: every operation is scoped to a workspace.
  • Spec first: schemas, OpenAPI, MCP schemas, and public model types are treated as contracts.
  • Query first: reads go through Query Service instead of scattered domain endpoints.
  • Provider neutral: storage is behind GraphStore providers.
  • Agent safe: resources are metadata-oriented, and write tools require explicit enablement.

Public Surfaces

  • REST API: api/openapi/openapi.yaml
  • CLI: umctl
  • MCP server: umodel-mcp
  • Web UI: web/
  • SDKs: sdk/go, sdk/python, and generated/java

Related References