Semantic Intelligence Runtime OEM · Enterprise meaning and assurance

Build enterprise meaning you can trust—without an LLM.

OntoGuard Semantic Intelligence Runtime OEM

Turn your existing data dictionaries, glossaries, policies, schemas and documents into one evidence-qualified enterprise ontology. Reduce manual reconciliation, make meaning reusable across systems and give data owners an evidence trail they can inspect.

Your data. Your environment.

No LLM. Fully offline.

The semantic core runs on your own CPU, without an internet connection, cloud AI service or GPU. Air-gapped deployment is possible with locally provisioned dependencies. Optional intelligence providers are your choice, never the source of governing truth.

No LLM requiredNo internet requiredNo GPU requiredNo cloud AI required

Why this matters: Keep sensitive data inside your environment. No model token fees are required for the core, and a pinned runtime, inputs and configuration support reproducible review without dependency on a changing cloud model.

Business problem → business value

From fragmented definitions to a governed business asset

What teams struggle with todayWhat OntoGuard delivers
Manual ontology workshops and mapping across incompatible systemsAutomated semantic construction from sources you already own; inspect the resulting concepts and evidence.
Glossaries, dictionaries and policies disagreeOne canonical ontology with different senses and unresolved conflicts visible.
Data owners cannot explain why a definition existsExact page, sheet or row provenance and semantic standing.
A field passes a schema check even though its business meaning changedNative semantic contracts, strict receipts and sealed-revision drift comparison.
A record says approved, executed or settled and a downstream workflow assumes it is trueSource assertion, interpreted meaning and operative fact kept distinct.

Keep your catalogs, data platforms, analytics and applications. OntoGuard supplies the evidence-qualified meaning they consume.

Interoperability built in

Every supported ingest and export format

Use the files and systems you already have. Publish the resulting semantic asset into the tools your enterprise already uses.

27Native ingest profilesDocuments, data, code, ontology and media
17Export formatsOne governed asset, multiple representations
157Documented API pathsVersioned OEM integration

All 27 qualified native ingest profiles

CategoryNative profilesCount
Business documentsDOCX, PDF, PPTX, Text4
Structured dataCSV, JSON, JSONL, XLSX, XML, YAML6
Code and integration artifactsDTSX, PySpark, Python, SQL4
Ontology and semantic standardsBounded OWL, JSON-LD, N-Quads, N-Triples, OBO, RDF/XML, SHACL Core, Turtle8
Images, audio and videoJPEG, MP4, PNG, WAV4
ArchivesZIP1

Native profiles are bounded tested processing capabilities, not necessarily distinct file extensions or unlimited interpretation of every object in a file.

All 17 output formats

CategoryOutput formatsCount
SpreadsheetsCSV, XLSX2
Structured dataJSON, JSONL2
Ontology and semantic standardsJSON-LD, N-Quads, N-Triples, OWL, RDF/XML, Turtle6
GraphsGraphML1
Queries and codeSPARQL, Python2
Documents and visualsPDF, PNG, SVG3
Complete semantic assetPackage1

The v1.89.332 release reports 17/17 export executions and zero core-field XLSX/JSONL mismatches on the qualified corpus.

Release-specific scope: v1.89.332; profile restrictions, optional dependencies and media interpretation boundaries are described in the interoperability and qualification detail.

Why OntoGuard

The combination that is difficult to replicate

01 · Differentiator

Meaning you can defend

Every visible constituted concept must have a real semantic identity, a definition, a domain, a category, standing and exact supporting source evidence. This is more than extracting names or generating plausible descriptions.

Business value: Reduce time spent tracing definitions and explaining decisions to data owners and auditors.

Beyond glossary extraction: The runtime interprets supported actors, events, objects and relationships in context, using bounded domain knowledge where applicable. A raw field name is not automatically a concept; different meanings remain distinguishable.

02 · Differentiator

One ontology, disagreements included

Competing meanings, uncertain classifications and unresolved conflicts stay visible. Business and technical concepts coexist without collapsing different senses or treating absence of evidence as false.

Business value: Avoid silently propagating inconsistent definitions into reports, models and agents.

03 · Differentiator

Business obligations, not just field names

The finance-first, domain-extensible model captures who owes what, to whom, under what conditions, with relevant amount, currency, time, lifecycle and lineage when supported. A source assertion is not legal discharge or authority.

Business value: Make contract, lending, payments and other operational information more usable without manufacturing facts.

04 · Differentiator

Three levels of truth

What a source says, what it means, and what is actually in force or authorized are separate. Unknown settlement or fulfillment stays unknown until appropriate evidence exists.

Business value: Prevent a status label from becoming an unsupported business conclusion.

05 · Differentiator

Contracts that govern meaning

Generate native semantic data contracts for eligible fields from a sealed revision. Strict validation, receipts, staleness and drift checks distinguish a changed definition from an unchanged physical data type.

Business value: Catch meaning changes before they flow silently into downstream reporting or integrations.

06 · Differentiator

One version, every representation

A governed semantic asset drives UI, API and export projections. Qualified input profiles and all 17 output formats, with release-reported cross-format parity checks, make the meaning reusable.

Business value: Reuse the asset across existing catalog, graph, BI and application tools rather than rebuilding maps for each consumer.

The defensibility is in the integrated assurance across the entire semantic asset, not a claim that any individual parser, ontology tool, graph or validation feature is unique.

Build versus assemble

What makes this more than a data catalog or AI generator

Existing approachWhat OntoGuard adds
Data catalogs and glossariesEvidence-qualified construction, distinct senses, explicit gaps and source-bound assertions across heterogeneous inputs.
LLM-assisted term generationDeterministic-first, completely offline construction; model assistance is optional and cannot establish truth on its own.
Semantic layers and knowledge graphsEvidence-bound concept identity, obligation/lifecycle semantics and sealed contract validation rather than graph connectivity alone.
Traditional schema validationMeaning and identity drift, version-specific validation receipts and impact in addition to physical type and constraints.
Manual ontology consultingA reusable software runtime and evidence-qualified assets to evaluate on your own corpus, rather than a new bespoke model for each engagement.

These are functional distinctions, not blanket assertions about every competing product. Some tools implement overlapping capabilities.

Evidence, not promises

See what the runtime actually produces

Actual controlled output

A conflicting definition stays visible

Two synthetic workbooks define Payment Settlement differently: final funds transfer versus acknowledgement before finality. The controlled ontology preserves the two senses for review instead of forcing a false consensus.

Inspect the actual 22-concept ontology · See obligations and lineage

Actual native contracts

Same INTEGER. Different business meaning.

“Principal Amount” changes from excluding interest and charges to including capitalized interest while still excluding other charges. Both assets retain INTEGER. Native comparison reports DEFINITION_DRIFT as a breaking change.

Read the business case · Inspect native receipts and revisions

Controlled synthetic-source output, separate from commercial production qualification; disclosure approval pending. No guarantee of perfect semantic accuracy or legal authority.

1,362Constituted conceptsv1.89.332 selected qualification corpus
3,434RelationshipsSame release-reported corpus
17/17Exports executedZero core XLSX/JSONL mismatches

The selected release campaign reports 99.3% established-definition precision (279/281) and 1,244 source plus 1,244 installed tests. Residual semantic errors remain; scoring was by the release engineer, not an independent external reviewer. See complete qualification scope.

Start with a paid, scoped evaluation

See the difference on your own material

01

Choose representative sources

Start with a glossary or dictionary, a policy and a schema. Agree source access, target use and what good looks like.

02

Build and inspect the asset

Run privately if required. Review evidence-qualified concepts, relationships, obligations, conflicts and contracts where eligible.

03

Measure the business case

Compare preparation and reconciliation effort, evidence quality, downstream reuse and export fidelity against your current workflow.

What you receive: a scoped governed ontology, reviewable gaps and conflicts, source links, supported graph and contract assets, and exports for your existing tools.
Technical product architecture, native evidence and additional buyer questions

OntoGuard Semantic Intelligence Runtime OEM

Business meaning is scattered across data dictionaries, schemas, policies, glossaries and documents. Manual reconciliation creates expensive rework and uncertainty. The runtime constructs evidence-qualified concepts and relationships and delivers reusable assets for existing data and AI platforms.

Dictionary + policy + schemaEvidence-qualified conceptsGraph + contracts + exports

Example: a lending dictionary defines Loan ID while a policy identifies an obligation and a schema establishes a physical amount type. The runtime can relate these observations without treating every raw field as a constituted concept. An established source assertion is not a guarantee of real-world truth.

What you receive

One ontology

Definition, Domain, Concept Category, standing and provenance; business and technical meaning coexist.

Construction & assurance

Obligations and lineage

Finance-first, domain-extensible interpretation of parties, conditions, lifecycle and consequences.

Obligation intelligence

Semantic data contracts

Versioned constraints bound to a sealed semantic revision; validation, drift and bounded impact.

Semantic contracts

Private and interoperable

CPU/local-first core, offline capability and qualified format profiles; reusable in OEM products.

Interoperability & deployment

Inspect the evidence path

The public workbench demonstrates how evidence can be navigated. The default sample is a minimized projection of a real installed v1.89.332 run from two synthetic workbooks (22 constituted concepts, 20 graph edges, two Needs Review rows). The separate illustrative learning mode explains additional behaviors not observed in this sample. See the release evidence and qualification boundaries.

Evaluate the runtime

Define the source or protected-action boundary, success criteria and integration responsibilities before a focused evaluation. No confidential data through the public site.

Source-backed lifecycle and obligation example

In a separate installed v1.89.332 run, two synthetic workbooks produced eight constituted concepts, three candidate graph edges and sixteen exact evidence locators. Native interpretation retained SUBMITTED (January 15), EXECUTED (January 16) and REVERSED (January 17) source-asserted states, with repayment conditions and USD context. Settlement, legal discharge and operative authority remain UNKNOWN. The example does not establish a linked real payment chain or production qualification.

Actual native contracts and sealed revision evidence

v1.89.332 generated a contract with two eligible properties. Recorded native validation, sealed revisions A/B/C, structural and semantic identity drift, stale receipt assessment, and downstream impact are available. A separate source-edited asset demonstrates definition drift with unchanged INTEGER type. Synthetic data; disclosure review pending; no action authority.

Inspect actual contract evidence · Download native-record projection

Does the runtime require an LLM?

No. Its deterministic-first core can run privately on CPU without an LLM, internet or GPU requirement. Optional qualified providers can augment evidence; they do not establish truth by themselves. The output is one constituted ontology with definitions, domains, categories and provenance, plus relationships, contracts and export assets.

The rich demonstration has 22 constituted rows (20 accepted/counting, two Needs Review), 20 candidate graph edges and 45 exact source locators across two workbooks. Explore the rich dataset. The separate native contract track validates integer versus string and compares real sealed revisions. Read the native case.

Evidence and buyer questions

Does SI need an LLM?

No. The Semantic Intelligence deterministic-first core runs privately on CPU without a required LLM, GPU or cloud AI. Optional qualified providers can contribute evidence; they do not turn a model response into established truth.

Inspect source or product evidence

Can SI run privately?

Yes. Semantic Intelligence supports private CPU processing without a required external model or internet connection. Optional providers are separately qualified and configured; private deployment still requires the buyer to manage its sources and environment.

Inspect source or product evidence