Using the semantic model

An executable description can be useful wherever another tool needs to understand structured data. The same Decl model can supply resolved values, relationships, and diagnostics to an application, a human reviewer, or an AI agent. These are uses of Decl’s current core, not additional language modes.

Decl for AI agents

Give agents a model of your data, not just its schema.

A shape description tells a consumer what fields may appear. A Decl model can also explain which entity a reference identifies, which values are derived, which relationships are valid, and how a failed constraint should be reported.

For example, an application can give an agent the Service Graph model, accept a proposed JSON document, and run the existing CLI:

decl evaluate topology.decl --input external=proposal.json --json

The application inspects the complete report’s ok and diagnostics, together with the command’s exit status. Named diagnostic IDs and paths let it associate a failed rule with the part of the proposal that needs review. A valid proposal can then be emitted as a completed document:

decl evaluate topology.decl --input external=proposal.json --output external

Run full evaluation or validation for acceptance. A partial Session result does not certify rules that have not been evaluated. A warning is also distinct from an error; the application decides which severities its workflow accepts.

This workflow uses current input binding and machine-readable reports. It does not claim that Decl already supplies an agent runtime, constrained decoding, automatic repairs, or an ontology export. Network access, tool calls, human approval, and applying a proposed change belong to the host application.

A foundation for conceptual models

Types can describe categories of data. References connect instances. Constraints express rules across them, while derived members calculate meaning from the supplied facts. The sequence type → relationship → constraint → instance can therefore be demonstrated with ordinary Decl modules.

That is a useful semantic foundation, but it is not a claim of OWL compatibility, open-world reasoning, inference over arbitrary facts, or globally persistent entity identity. Decl’s references identify canonical places in a fixed evaluation universe. Its relationship rules remain the authority.

Prove applications before expanding the core

Start with a concrete model, representative documents, and an expected report. Show where the model expresses a real rule and where application code is still needed. Add a graph or inference operation only when a repeatable example demonstrates the missing capability.

The Service Graph and hardware interconnect are starting points. The examples index explains their evidence and the questions to ask of a comparison with another language.