Reference architecture
Draft · v0.2Semantic Context Blueprint
Seven layers of the reference architecture
The Context Runtime is specified as a platform-neutral architecture component. Not because we must build our own product, but because this is exactly where business semantics, metadata, policy and agent behaviour meet.
Technology roles
| Role | Standards & tools | Purpose |
|---|---|---|
| Business ontology | RDF, OWL, SKOS, SHACL | Canonical meaning and semantic validation |
| Knowledge graph | Apache Jena, RDF4J | Storage and querying of semantic relations |
| Metadata & governance | OpenMetadata, DataHub | Catalogue, ownership, lineage, quality |
| Open table format | Apache Iceberg | Cross-platform table and snapshot representation |
| Data contracts | ODCS, JSON Schema | Schema, quality and delivery agreements |
| Data lineage | OpenLineage, PROV-O | Traceability of data and knowledge |
| Policy enforcement | Open Policy Agent | Independent policy decisions |
| Agent context exchange | MCP | Making context and tools accessible |
| Agent interoperability | A2A | Collaboration between agents |
| Workflow orchestration | LangGraph, Temporal | Stateful agent workflows, controlled execution |
| Observability | OpenTelemetry | Distributed tracing and runtime instrumentation |
Platform integrations
Commercial platforms (e.g. Microsoft Fabric IQ, Foundry IQ, Work IQ) can be valuable implementations of the architecture.
They connect through a Context Adapter that normalises, authorises, validates and preserves provenance, so that a platform can be a powerful implementation of the architecture without defining the architecture itself.