| Term | Description |
|---|---|
| Organization | A top-level tenant representing a single business entity. All data, projects, and executions are isolated per organization. |
| Project | A scoped unit within an organization that runs a specific AI Template, with defined data pipelines, AI Template, Clear business goal and assigned teams. All AI execution is contained within a project boundary. |
| Environment | An execution context (Core or Apex – sandboxed infra) that separates risk, configuration, and responsibility for AI workflows. |
| Sandboxing | An isolated environment used for AI systems. This adds an extra layer of security and ensures data isolation thorugh the Apex Environment. This is not for testing purpose like traditional means. |
| Core Environment | The primary operational environment for stable and approved AI workflows serving business-critical use cases. |
| Apex Environment | The highest-trust environment used for fully validated, production-grade AI execution and external system integrations. |
| Data Pipeline | A defined flow that ingests, processes, transforms, and prepares data for AI reasoning and execution. |
| AI Template | A scoped AI template designed to solve a specific business problem (e.g., customer segmentation). An AI Template typically consists of multiple AI agents working together in a coordinated workflow. |
| AI Agent | A specific AI entity responsible for performing a single, well-defined task, such as generating reports or executing controlled changes in external integrations. |
| RAG (Retrieval-Augmented Generation) | A governed context-retrieval system that uses Retrieval-Augmented Generation to supply approved, project-scoped business data to AI Images during execution. |
| Standard RAG System | The platform implementation of Retrieval-Augmented Generation that supplies governed, project-scoped business data as execution context to AI Images. |
| Task Execution | A single run of an AI workflow within a project, tracked with state, progress, inputs, outputs, and metrics. |
| Execution Control | The ability to start, pause, retry, stop, and monitor executions to ensure operational safety and reliability. |
| Integration | A controlled connection to external systems such as CRM, ERP, global commerce, platforms, or marketing tools, executed only through approved workflows. |
| Isolation | Enforcement of strict separation across organizations, projects, and environments to prevent data leakage and cross-tenant risk. |
| Observability | Visibility into AI behavior through logs, metrics, status, and performance indicators for debugging and compliance. |
| Roles | Predefined permission sets that determine what actions a collaborator can perform within an organization, project, or environment. |
| RBAC | A security mechanism that enforces scoped access across AI environments by mapping roles to permitted actions and resources. |
| Teams | A group of users assigned to a project who interact with, operate, and consume AI workflows and outputs. |
Overview
Key Concepts
This page defines the core terminology used across Avigrah. These concepts describe how the platform structures environments, executes AI, and enforces governance.