Architecture

The Ontos
Intelligence Stack

Three layers that turn disconnected enterprise data into compounding intelligence. Built for technical buyers who need to understand what's under the hood.

Overview

Three-layer architecture

01

Layer 1 — Data Layer

Connected Sources

Document parsing, entity extraction, provenance tracking. Every data source your business touches, unified under a single schema.

02

Layer 2 — Ontology Layer

The .onto Runtime

Compiled domain models, typed entity graphs, inference rules, confidence calibration, gap detection. This is the core intelligence engine.

03

Layer 3 — Agent Layer

AI Agents

Reasoning over the unified model. Natural language queries, proactive alerts, report generation, workflow automation — all grounded in structure.

Layer 1 — Data Layer

Connect. Parse. Extract. Track.

The data layer handles ingestion from any source, extracts entities against your .onto schema, and tracks provenance from the moment data enters the system.

Documents

PDF, DOCX, XLSX, CSV, PPT — parsed with layout-aware extraction that preserves tables, headers, and document structure.

Databases

SQL databases, data warehouses, and analytics platforms. Live connections with change detection and incremental extraction.

APIs & SaaS

CRM, ERP, HRIS, project management. REST and GraphQL connectors with OAuth and API key authentication.

Email & Comms

Email archives, Slack, Teams. Entity extraction from unstructured communications with thread context preservation.

Entity Extraction

Every document and data source is parsed against the .onto schema. People, companies, obligations, dates, amounts — typed and linked from ingestion.

Provenance Tracking

Every extracted fact traces back to its source document, page number, and extraction confidence. From day one. Non-negotiable.

Layer 2 — Ontology Layer

The reasoning engine

This is the core IP. The ontology runtime compiles .onto domain models into executable intelligence — resolving entities, firing inference rules, scoring confidence, and detecting gaps.

// The compilation pipeline
domain.onto  →  Compiler  →  Domain IR  →  Runtime

// At compile time:
// ✓ Type checking (all entities, relations, rules validated)
// ✓ Constraint resolution (inheritance, required fields)
// ✓ Rule optimization (pattern matching compiled to IR)
// ✓ Source binding (data sources mapped to extraction targets)

// At runtime:
// ✓ Entity resolution across all connected sources
// ✓ Inference rule evaluation on every data change
// ✓ Confidence scoring with source reliability weighting
// ✓ Gap detection against the domain model
// ✓ Sub-millisecond rule execution

Compilation Pipeline

The .onto file compiles to a Domain Intermediate Representation (IR). Type checking, constraint validation, inheritance resolution — all at compile time.

Entity Resolution

The same entity appears across dozens of sources with different names, formats, and attributes. The ontology runtime resolves them into a single canonical entity with merged attributes.

Inference Rules

Pattern-matching rules that fire when data meets conditions. Competitive threats, compliance gaps, risk thresholds — detected automatically when new data arrives.

Confidence Scoring

Every value carries a confidence score based on source reliability, extraction quality, and corroboration across sources. Agents know what they know — and what they don't.

Gap Detection

The model knows what it should know. When required entities are missing, when relationships are incomplete, when data is stale — the gap detector identifies blind spots and proposes fixes.

Contradiction Resolution

When two sources disagree, the runtime doesn't pick one — it flags the contradiction with both values, their sources, and confidence scores for human resolution.

Layer 3 — Agent Layer

Intelligence that acts

AI agents that don't guess — they reason over typed entity graphs with confidence scores and full provenance. Every answer is grounded in the structured model.

Natural Language Query

Ask questions in plain English. Get structured answers with citations, confidence scores, and the reasoning path the system took to reach the answer.

Proactive Alerting

Inference rules fire in real-time. When a new document contradicts an existing obligation, when a risk score crosses a threshold — you know immediately.

Report Generation

IC memos, due diligence summaries, compliance reports — generated from structured intelligence with full provenance. Every claim is sourced.

Workflow Automation

Actions triggered by intelligence: route documents for review, create approval workflows, flag items for human decision, update downstream systems.

Compounding Intelligence

Each layer feeds the next.

Layer 1 feeds Layer 2. More data sources mean more entities, more relationships, more complete models.

Layer 2 makes Layer 3 intelligent. Richer ontology models mean smarter agents with better reasoning and fewer blind spots.

Layer 3 improves Layer 2. Agent outputs identify gaps, correct errors, and propose schema extensions that make the model smarter.

The more data flows through, the smarter the system gets. This is compounding intelligence.

Standards

Built on decades of research

Not invented in a garage. Built on internationally recognized ontology standards — made practical for enterprise AI.

BFO

Basic Formal Ontology

ISO/IEC 21838-2. The most widely adopted upper-level ontology in science and industry. Our foundation.

IOF

Industrial Ontologies Foundry

Industry-standard ontologies for manufacturing, supply chain, and industrial operations. Built-in, not bolted on.

FIBO

Financial Industry Business Ontology

The EDM Council's standard for financial services. Regulatory compliance, risk management, and instrument modeling.

See the stack
in action.

We'll walk you through each layer with your own data.