Glossary
Enterprise data and AI terms
Practical definitions for the concepts behind governed data access. Written for IT professionals, data teams, and enterprise architects working with Microsoft Dataverse, Power Platform, and AI systems.
Agent Guardrails
The mechanisms that constrain what an AI agent backed by a large language model is allowed to do. Guardrails operate at four layers: input, model, output, and tool calls. Only the tool-call layer holds deterministically when the prompt is broken.
Data Access Templates
Predefined, governed data operations that carry business logic, table relationships, and access rules into reusable units. Consumers select a template and supply parameters instead of writing or generating queries.
Data Governance
The set of policies, processes, and controls that ensure enterprise data is accurate, consistent, secure, and used appropriately. In practice, it determines who can access what data, under what rules, and with what audit trail.
Deterministic Execution
A data access approach where the same input always produces the same output. In contrast to AI-generated queries, deterministic execution uses predefined templates to guarantee consistent, predictable results.
Enterprise AI Agent
An AI agent designed to operate safely on enterprise systems, distinguished from generic AI agents by governed data access, deterministic execution, scoped credentials per agent, durable domain context, and per-call audit.
Governed Self-Service Data Access
A pattern where business users access enterprise data directly, without writing queries or waiting on IT, while access rules, business definitions, and audit trails are enforced by the platform serving the data.
Metric Definition
A documented, executable specification of a business metric, establishing how it is calculated, which records qualify, what the time boundaries are, and who owns it.
Microsoft Fabric
Microsoft's unified analytics platform: OneLake plus multiple compute engines, Data Factory pipelines, and Power BI in one SaaS environment. Fabric stores, prepares, and analyzes data; governing how downstream consumers reach it per person, agent, or app is a different layer.
Microsoft Power Platform
A suite of low-code tools from Microsoft, including Power Apps, Power BI, Power Automate, Power Pages, and Copilot Studio, built on top of Microsoft Dataverse for building apps, automating workflows, and deploying AI agents.
Model Context Protocol (MCP)
An open standard for connecting AI systems to external data sources, tools, and services. MCP provides a standardized interface so AI tools can access data without proprietary connectors.
Parameterized Data Operation
A pre-defined data operation that accepts typed parameters and runs the same tested query or business logic every time. Distinct from runtime-generated queries, which can produce different results for the same input.
Policy Enforcement (for LLM Tools)
A deterministic layer between an AI agent and the system of record that evaluates each tool call against rules in code and either executes or refuses. Policy enforcement runs outside the model and is not influenced by the prompt.
Prompt Injection
A class of attack where a user manipulates the input to a large language model so the model behaves outside its intended instructions. The injected content can arrive directly from the user or indirectly through data the model is asked to read.
Semantic Data Layer
An abstraction layer between enterprise data sources and the consumers that use them. It translates database complexity into business terms, applies governance rules, and provides consistent, trusted answers.
Text-to-SQL
An AI approach that converts natural language questions into SQL database queries. While powerful for ad-hoc exploration, text-to-SQL faces accuracy challenges on complex enterprise data.
See these concepts in action
Learn how dhino applies these principles to give enterprises a single governed data layer for people, systems, and AI.