Questions & answers
What Veracy is, how it works, and what it does with AI.
Usual questions and short answers, in plain language.
The platform capabilities: in detail
What Veracy is
- What is Veracy?
- An autonomous data platform. Point it at a source database and it authors the data contracts, derives the warehouse, generates quality rules, aligns business vocabulary and serves governed answers to people and AI agents — with a verifiable receipt on every answer. You review and sign; the machine builds.
- Is Veracy a data catalog, a quality tool, a semantic layer or an ETL tool?
- It is all of those, generated from one signed contract and kept converged. That is the difference from point tools: a catalog, a quality suite and a semantic layer bought separately each start from a blank page and drift apart. In Veracy they are views of the same source of truth.
- Who is it for?
- Data leaders who want one platform instead of a stack and a glue team; data engineers who are tired of per-table pipeline code; analysts and business users who want the same answer every time; AI and platform teams that need governed, verifiable answers for agents; stewards and auditors who need proof to be an export, not a project.
- What stage is the product at?
- Pre-launch, with the waitlist open. A proof of value runs on a read replica of one of your systems in days, on your own data.
How it works
- What do I have to do?
- Connect a source (a read replica is enough), review the contracts Veracy drafts, and sign them. The signature is the only human step that is required; everything downstream is generated from it.
- What does "generated from the contract" mean?
- The signed contract is the input. From it Veracy derives the warehouse layers (Ingest → Trust → Serve), the quality rules, the catalog entries, the lineage and the metric definitions. When the contract changes, all of them are regenerated in step.
- What happens when a source changes?
- The change is compared to the signed contract. Compatible changes heal themselves and are versioned with history kept. A change of meaning is refused and escalated to a person — nothing downstream drifts silently.
- How long does onboarding take?
- Days from a read replica to a governed warehouse, because there is no pipeline code to write per table. The time goes into your team reviewing and signing contracts, not into building.
Trust and AI
- Does an AI compute my numbers?
- No. AI helps you phrase a question in your organisation's vocabulary; the answer is compiled as deterministic SQL over governed data. The language model is never in the calculation path, so the same question returns the same number every time.
- What is a receipt?
- A verifiable artefact attached to every answer: which metric and contract versions it was computed from, which policy applied, and evidence that can be checked offline by an auditor, a partner or another agent — without asking Veracy.
- How is access controlled?
- Access is declared once in the contract and enforced natively inside the engine, row by row, for every consumer — people, dashboards and AI agents. It fails closed: anything not explicitly allowed is denied and logged.
- Am I locked in?
- No. The warehouse is Apache Iceberg and Parquet; contracts are ODCS; vocabulary is SKOS/GS1/FIBO; provenance is W3C PROV; agents connect over MCP. The exit door is part of the product.
Glossary
- Data contract
- A machine-readable agreement about a dataset: schema, quality rules, service levels and ownership. In Veracy it is authored by the machine, signed by a person, and is the input everything else is generated from.
- ODCS
- The Open Data Contract Standard — the open, vendor-neutral format Veracy uses for contracts, so they are portable.
- Ingest → Trust → Serve
- The three warehouse layers Veracy derives from the contract: raw data as it arrived, a deduplicated and historised single source of truth, and the business-ready layer reports and agents consume.
- Data Vault 2.0
- The modelling method used for the Trust layer: it keeps full history and keeps every record's identity stable through time, so nothing is quietly rewritten.
- Semantic layer
- The single set of metric, dimension and hierarchy definitions every tool consumes, so "revenue" is defined once and computed the same way everywhere.
- Lineage
- The recorded path from a source column to every warehouse column, metric and report built from it. Veracy records it as a by-product of the build, in W3C PROV terms.
- Receipt
- The verifiable proof attached to every answer — metric and contract versions, policy applied, evidence — checkable offline by anyone.
- Fail-closed
- The access rule that anything not explicitly allowed is denied. Unknown principals get nothing by default.
- Agentic BI
- Analytics where AI agents are first-class consumers alongside people and dashboards — the same governed number, the same receipt, the same access enforcement.
- Quarantine
- Where rows that violate a contract rule are held, with the rule and version that rejected them, so they never reach a report.
The nine capabilities
Still have a question? Ask us, or see it on your own data.