Product

One engine, five doors

Someone typing a question, an analyst approving a definition, an AI assistant calling out to your data, a question that spans three systems, and your own product embedding all of it. Every one runs through the same checks. There is no second, looser path that exists because it was quicker to build.

One semantic layerOne set of row-level rulesOne audit trail

Askanswers that show their workAsk questions of governed enterprise data in plain English or by voice, and get the query, the sources, and a confidence signal alongside every answer.
Semantic layerone definition, everywhereDefine metrics, dimensions and joins once in MetricQL. Certification stays a human decision, so a plausible definition never becomes enterprise truth.
Agents & MCPthe same rules, through a different doorNexatron exposes an MCP server so agents like Claude query enterprise data under the same row-level security, certified metrics and audit as a person.
Federationthe question that spans three systemsAnswer questions across a CRM, an ERP and a warehouse without a migration first, and refuse rather than guess when the join key was never certified.
Embedanswers inside your own productPut governed natural-language analytics into your own application with the @nexatron/chat SDK, with tenant isolation enforced at the query layer.

Every page here states what it won’t do

That is not modesty. A platform that answers everything has no way to tell you when it is guessing, and the answer you cannot check is the one that costs you.

How we measure answer quality