Proposal: Adopt ontology-query as an Ontology Access Layer tool

[zljie]

Proposal: Adopt ontology-query as an Ontology Access Layer tool for OSI

Summary

I propose that the OSI project adopt (or officially recommend) a lightweight reference tool called ontology-query as an Ontology Access Layer for OSI semantic models.

ontology-query loads an OSI YAML file, builds a small knowledge graph, materializes Datalog facts (deterministic reasoning), and exposes:

• a Python SDK for other agents/tools
• a CLI for interactive ontology interrogation

This is intended to support agent-to-agent architectures (Ontology Agent ↔ Data/Service Agent) by providing a rigorous, deterministic way to query and reason over the ontology (datasets/fields/relationships/metrics/behavior), without relying on NL→SQL.

Motivation

OSI standardizes semantic models, but many agent workflows need a consistent “ontology access” interface:

• Discoverability: list datasets/fields/metrics/actions
• Deterministic reasoning: relationship reachability / join-path discovery / impact analysis
• Attribution & planning support: determine which actions can change which fields (via behavior.actions[].effects)
• Enable an Ontology Agent to request data from a Data/Service Agent using a stable, machine-readable query plan

While tools can implement this independently, a small reference implementation would accelerate adoption and reduce fragmentation.

Proposed scope (initial)

Supported OSI concepts

• datasets + fields
• relationships (transitive reachability)
• metrics (names; expressions are not interpreted by this tool)
• behavior layer:
  • preferred: semantic_model.behavior
  • legacy: behavior embedded under custom_extensions JSON
  • actions (preferred) + action_types (alias)
  • effects (impact annotations)

Example queries

• “List datasets”
• “List fields of suppliers”
• “Which actions can change suppliers.status?”
• “Show effects of suppliers/block”
• “Is dataset A reachable to dataset B through relationships?”

Repository

Reference implementation: https://github.com/zljie/ontology-query

Why Datalog (rigor)

The core requirement is rigorous logical inference with reproducible results. Datalog provides:

• deterministic evaluation
• transitive closure / rule-based derivations
• a foundation for explainability (proof-style traces can be added incrementally)

Suggested next steps

1. If maintainers agree with the direction, we can:
   • add ontology-query under an OSI org repository (or list as an official companion tool)
   • align naming/packaging and CI expectations
2. Add a small “ontology access layer” section to OSI docs referencing the tool.

Questions

1. Would OSI maintainers accept an official companion tool for ontology access?
2. Preference: keep it as a separate repo under OSI org, or under tools/ in the main repo?
3. Any constraints on dependencies (e.g., allow pyDatalog) or target Python versions?