Glossary
The language of governed AI analytics.
Definitions of the terms behind Nexatron: the semantic layer, MetricQL, MCP, evidence-backed answers, and the governance that applies to people and their agents alike.
- Semantic layer
- Also called Metrics layer, Semantic modelA semantic layer is a governed model that maps raw database tables and columns to the business concepts people reason in: metrics, dimensions, and the relationships between them. A question is then answered the same way every time, whoever asks it. In Nexatron the semantic layer is expressed in MetricQL and underpins certified metrics, so a query resolves to one agreed definition rather than ad-hoc SQL.Guide: what is a semantic layer?
- MetricQL
- MetricQL is the declarative language Nexatron uses to define metrics, dimensions, hierarchies, and joins as reusable building blocks in the semantic layer. Because every metric is defined once in MetricQL and compiled to warehouse SQL at query time, the same definition serves natural-language questions, certified-metric lookups, and agent queries, so the numbers match on every surface.
- Model Context Protocol (MCP)
- Also called MCP, MCP serverThe Model Context Protocol (MCP) is an open standard that lets AI agents and applications connect to external tools and data through a consistent, structured interface instead of bespoke integrations. Nexatron exposes an MCP server over Streamable HTTP (JSON-RPC 2.0) secured with OAuth 2.1, so an MCP-capable agent such as Claude, ChatGPT, or Cursor can ask governed questions, discover and run certified metrics, and, where the workspace allows it, run actions under the same governance as a human user.The Nexatron MCP server
- Evidence-backed answer
- An evidence-backed answer ships with what you need to check it, rather than a bare number you have to take on faith. A governed Nexatron answer returns the result rows, the exact query that produced them, the data lineage, and a confidence score, so a person or an agent can judge how much to rely on it before acting.
- Confidence score
- A confidence score is a numeric signal of how much weight to put on a generated answer, shown so the reader does not have to guess. Nexatron returns a 0-100 confidence score with each governed answer, shown as High, Verify or Low alongside the generated query and lineage, so low-certainty results are visible and can be reviewed before they drive a decision.How we measure answer quality
- Governed analytics
- Governed analytics is the practice of putting analytics behind enforced controls: who can see what, how each metric is defined, and a record of every query. Results are then consistent, permissioned and auditable. Nexatron applies four-layer row-level security, certified metrics, and a hash-chained audit log to every query, and applies the same governance whether the request comes from a person or an AI agent.
- Conversational analytics
- Also called Conversational BI, Chat with your dataConversational analytics, also called conversational BI, lets a person ask a question of business data in everyday language, typed or spoken, and get the answer back as a chart, a table or a sentence, then ask a follow-up that keeps the earlier context. It replaces waiting for a report to be built, not the need for agreed definitions. In Nexatron each question is answered through the semantic layer under the asker's own permissions, and the answer arrives with the query that produced it and a confidence score.Asking questions in Nexatron
- Enterprise context
- Also called Context layer, Business contextEnterprise context is the knowledge an AI system needs to answer a question correctly inside a particular company rather than in general: what its metrics mean, how records in different systems relate to each other, and who is allowed to see what. A model without it can produce fluent answers that use the wrong definition or ignore an access rule. In Nexatron that context is held in the semantic layer (certified metrics, joins and a business glossary) and in each person's own data permissions, and both apply to every question, whether a person or an AI agent asks it.
- Natural-language-to-SQL (NL→SQL)
- Also called Text-to-SQL, NL2SQLNatural-language-to-SQL (NL→SQL), also called text-to-SQL, is the translation of a plain-language question into an executable SQL query so non-technical users can query a database directly. In Nexatron the model's output is treated as untrusted and validated against the semantic layer before any query runs, and the generated SQL is returned with the answer so the translation is always inspectable.Asking questions in Nexatron
- Certified metric
- A certified metric is a business measure, such as revenue or active users, whose calculation has been reviewed and approved as the canonical definition, so different teams don't compute it different ways. Nexatron stores certified metrics in the semantic layer and lets people and MCP-capable agents discover and execute them, so the answer to "what's revenue?" is the agreed one every time.
- Row-level security
- Also called RLSRow-level security (RLS) restricts query results so each user sees only the rows they are permitted to. A regional manager sees only their region's records. Nexatron enforces four-layer row-level security at query execution, and applies it identically to human and AI-agent queries so an agent can never read rows its user could not.Security and governance in the Trust Center
- Cross-source federation
- Also called Data federation, Federated queryCross-source federation is the ability to run a single query that combines data living in different systems without first copying it all into one place. Nexatron can federate across its nine supported warehouse, database, and file connectors using an embedded DuckDB engine, so a question can join data from more than one source while governance still applies. Federation is on by default for new workspaces, and a workspace admin can turn it off.Cross-source analytics in Nexatron
- Self-healing semantic layer
- A self-healing semantic layer is one that detects when underlying schemas change (renamed columns, new tables, shifted types) and adapts its model so questions keep resolving instead of failing silently. Nexatron detects this drift and proposes updates for a person to review, which cuts the manual upkeep that normally keeps a metric model accurate.
- Lineage
- Data lineage is the documented trail of where a result originated and every transformation applied to produce it, so a number can be traced back to its sources. Nexatron returns lineage with every governed answer, alongside the exact query and a confidence score, so anyone reviewing a result can see which sources and logic produced it.
See these ideas at work
Connect a warehouse, ask in chat or over MCP, and get an answer with the exact query, lineage, and a confidence score behind it.