Semantic Index

The Semantic Index maps natural software concepts to deterministic Software Graph entities.

It is not AI, vector search, embeddings, or a language model. The index is derived from graph facts only:

• stable IDs
• display names
• normalized names
• node kinds
• packages, modules, namespaces, and frameworks
• aliases from identifiers and framework conventions
• keywords from source paths, metadata, relationships, and parent chains
• relationship context and graph importance

ontoly build writes the default output bundle to ontoly-output/, including the Software Graph and derived indexes. Use ontoly build . --output .ontoly when you want the compact cache-style .ontoly/ directory instead.

CLI

ontoly search "sleep thresholds"
ontoly search authentication
ontoly find JWT
ontoly locate notifications
ontoly search "JWT secret" --category configuration --json

The output includes:

• matched concepts
• ranked candidate symbols
• confidence
• evidence
• recommended capability

API

import { createSemanticIndex, resolveIntent } from "@0xsarwagya/ontoly-core";
 
const index = createSemanticIndex(graph);
const result = resolveIntent(index, "Plan Definition Resource");

Every score is explainable. Candidates include score factors such as exact symbol match, exact alias match, token overlap, relationship context, graph importance, and architecture layer.

Validation

The permanent search validation corpus lives in validation/search.

pnpm --filter @0xsarwagya/ontoly-core build
node validation/search/evaluate-search.mjs --graph .ontoly/SoftwareGraph.json

The evaluator reports Top-1 accuracy, Top-5 accuracy, latency, and per-question results.