Subscribe to to stay in the know
Get our newsletter and latest updates about our company, what we are working on, and other cloud-related news.
Onix Semantic Twin is a governed context layer that sits between your data and your AI. It gives every agent the data context it needs: what each KPI means, how your data connects, who may see it and where each fact came from. Built from the systems and documents you already run, approved by your owners and kept current, it is AI context your whole business can trust.
answer accuracy with semantic grounding, on certified questions
fewer tokens per answer through shared context
Context is the layer between enterprise data and AI. Cloud gives agents the data and models give them reasoning, but only context tells them what your data means, who may see it and where each fact came from.
A modern data platform. Data moved, validated and running on BigQuery or its peers.
Models, agent runtimes and MCP. Any agent can now reach any table.
The enterprise context layer
Shared me 192×18 dentity, policy and evidence that every agent reads at query time, kept current.
Many agents in production, each one cheaper to build than the last, with answers that hold up.
Onix Semantic Twin gives your AI the data context behind every number. It applies the right definition to every KPI, uses the joins your business already trusts and returns the number your teams agree on, with the source attached.
Illustrative model of what happens as agents go into production.
Agents in production
Effort saved
Grounded accuracy
Your owners approve every definition, and access rules and policy travel with the context to every agent.
A support agent answers a billing question without exposing a customer’s personal data, because the policy is part of the context it reads.
Every fact traces back to its source system, document and version through column-level lineage.
An analyst asks where “active member” comes from and gets the table, the query and the Confluence page that defines it.
Every answer carries its receipt: the definition, the source passage, the approver and the version it used.
An auditor challenges a revenue figure and sees exactly which approved definition and which data produced it.
From live pipeline health and SLA monitoring to root-cause diagnostics and autonomous remediation, your data engineers work reliability, lineage and anomalies from one platform.
EKE works out what each asset means, finds the same logical asset across copies, filters and renames, and connects your terms into a knowledge graph grounded in the industry-standard ontology you started from.
Approved rules apply to every user’s queries, and an owner can widen a rule from one domain to all of them. The approver, date and scope stay with the rule.
Approved definitions, descriptions and relationships go out through the surfaces your teams already use, and the context harness serves them to any agent runtime at query time. In a Google Cloud estate we write into Knowledge Catalog as entries, aspects, glossary terms and entry links.
Sentinel agents watch the sources. When a schema moves, a KPI is redefined or a policy changes, the twin re-discovers what changed and routes it back to the owner. Every approved change publishes as a new, immutable version, so grounded agents never read a definition that quietly went stale.
An inventory of your AI use cases, a count of how many times the same assets were contextualised, what that costs to maintain, and a hygiene report on duplicate and stale assets.
One data domain and one hero use case, such as Talk2Data, measured against a baseline agreed in week one.
Adjacent domains reuse the ontology, so each one stands up faster than the last.
Continuous refresh, drift resolution and a value scorecard your executives see every month.
Orchestration and schedulers
ETL and transformation
Databases
Databases
Process automation
Data catalogs
Orchestration and schedulers
BI and reporting
ETL and transformation
ETL and transformation
Databases
Databases
Data catalogs
Data catalogs
Orchestration and schedulers
ETL and transformation
ETL and transformation
Databases
IBM Informix Databases
Process automation
Data catalogs
Orchestration and schedulers
Databases
Databases
ETL and transformation
Orchestration and schedulers
Orchestration and schedulers
Data catalogs
Databases
Databases
Databases
ETL and transformation
BI and reporting
Orchestration and schedulers
Data catalogs
Process automation
Databases
Databases
ETL and transformation Orchestration and schedulers
ETL and transformation
Orchestration and schedulers
Data catalogs
Process automation
BI and reporting
The Semantic Twin isn’t built for one sector. EKE starts from the industry standard your business already speaks, then grows it from your own schemas and documents.
Core banking, loan origination, risk data marts, credit policy documents
Credit risk, AML investigation, relationship manager copilots
Talk2Data at a top-5 US health insurer: 90–95% answer accuracy at under $0.04 per request.
Policy admin, claims, reinsurance ledgers, underwriting guidelines
Claims triage, underwriting assist, regulatory reporting
Claims and eligibility, provider directories, clinical policies, benefit documents
Utilization review, member service, cost-of-care analytics
Talk2Data at a top-5 US health insurer: 90–95% answer accuracy at under $0.04 per request.
POS and e-commerce, loyalty and CRM, supply chain, promotion calendars
Merchandising analytics, loyalty agents, supply chain copilots
A national specialty retailer’s loyalty knowledge connected in one run: 29 documents became 2.2k entities and 231 cited passages.
Billing, CRM, network assurance, IVR and contact centre systems
Contact centre Talk2Data, churn prevention, network operations
One of the world’s largest telecom providers cut AI initiative setup from 6 months to 6 weeks.
Database schemas, OpenAPI specs, catalog exports, business glossaries
Greenfield bootstrap in a few weeks, then the same governed loop
A conversational analytics service grounded in the Semantic Twin, with an observability and traceability layer for audit.
answer accuracy
per request
Every answer carries its definition, lineage and policy.
Business users self-serve. Analysts analyze instead of reconciling four versions of one KPI.
New agents inherit the same governed context. Nobody re-teaches the business.
Fewer wrong answers, fewer retries, less human review and fewer tokens spent re-stitching context.
Onix Semantic Twin is a governed context layer for enterprise AI. It is a living model of your business, built from your data, code, reports and documents, that tells every agent what your data means, how it connects, who may use it and where each fact came from. Teams building AI agents often call this data context or AI context. The twin builds it once, has your owners approve it and keeps it current, so every agent and BI tool reads the same governed meaning.
A catalog documents assets for people. Onix Semantic Twin grounds agents at query time, across platforms, with policy attached. We integrate with your catalog and write into it rather than replacing it.
For your Google estate, largely yes, and it’s where we publish. We add the business meaning it can’t reach on its own: process and policy definitions, KPI logic your CFO will sign, one identity for an asset across copies, and systems with no connector at all, such as mainframe, bespoke apps and legacy BI.
Reverse ontology works from what exists today. Quality signals become part of the context, so agents know what to trust. Waiting to clean everything first is exactly what stalls these programs.
RAG retrieves documents. The Semantic Twin governs meaning (definitions, identity, lineage and policy) and serves it to any agent, including your RAG applications.
You do. The Enterprise Knowledge Engine proposes and your domain owners approve. Nothing grounds an agent without their sign-off, and every artefact we produce belongs to you.
Sentinel agents detect the change, re-discover what it affects and route it to the owner. The approved version updates, and every grounded agent reads the new definition from then on.
It’s the engine that creates the context layer. It connects to 40+ systems, reads the technical estate through Eagle (SQL, ETL, code, logs, lineage and BI) and reads documents such as wikis, PDFs and tickets. It builds the ontology from an industry seed, keeps a receipt for every fact, and proposes changes for your owners to approve.
Four things make it hard. Definitions drift as teams change their reports. The same asset lives in many copies under different names. Policy and quality sit in separate tools from the data. And much of the meaning lives in documents, not tables. EKE handles each one: it tracks definitions as they change, resolves one identity per asset, attaches policy and quality to the context, and reads documents alongside data.
Yes. The context is yours and it’s portable. Ontologies export in open formats such as RDF, OWL, Turtle and SHACL, and the twin publishes through open standards including MCP, A2A and OpenLineage.
No. The engine is the same in every industry. What changes is the seed ontology it starts from, such as FIBO for banking, ACORD for insurance, FHIR for healthcare, GS1 for retail or TM Forum for telecom. With no standard to start from, it bootstraps from your own schemas, API specs and glossaries.
No. Wingspan is powered by the Onix Semantic Twin, but the Semantic Twin is a product in its own right. Your own agents, copilots and BI tools can read from it through MCP, APIs, graph and SQL, with or without the rest of Wingspan.
Get our newsletter and latest updates about our company, what we are working on, and other cloud-related news.