> ## Documentation Index
> Fetch the complete documentation index at: https://docs.trymaitai.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Core Concepts

> The mental model: Company > Application > Intent, plus Sessions, Intent Groups, and quality tooling.

This page defines the key objects you’ll see in Maitai (and in the Portal).

## Resource hierarchy

* **Company**: the top-level tenant for your team.
* **Application**: a product/service you’re integrating (ex: `checkout`, `support_bot`).
* **Intent**: a specific action type within an Application (also called an **ApplicationAction**).

In practice, the hierarchy is:

**Company > Application > Intent**

## Cross-application grouping

* **Intent Group**: a company-wide grouping of Intents by shared `action_type` (so you can manage monitoring/testing across many apps at once).

See: [Intent Groups](/observe/resources/intent_groups)

## Requests and sessions

* **Request**: one chat completion (messages to model output).
* **Session**: a set of requests grouped by `session_id` (highly recommended). Sessions are the main way to debug multi-turn behavior in the Portal.

See:

* [Sessions](/observe/agent_sessions)
* [Applications](/observe/resources/applications)

## Agents

An **Agent** is a configurable AI runtime that can plan, execute capabilities, and synthesize results. Each agent has an **Execution Mode**, chosen when you build it:

* **Reasoning** (`reasoning`): a multi-step orchestration loop that plans actions, processes results, and iterates until ready to respond.
* **Route** (`route`): single-step capability selection, where each request goes straight to one capability.

Agents can delegate to **Sub-Agents**, call **API Actions**, run **LLM Actions** and **Workflows**, and maintain structured, session-scoped **Agent State**.

See: [Create Agent](/build/agents/create_agent) · [Execution Mode](/build/agents/execution_mode) · [Agent Call](/sdk/agent_call)

## Quality and testing objects

* **Sentinel**: an intent-group-level quality check (what to evaluate, severity, optional correction guidance).
* **Evaluation**: the result of running one or more Sentinels against a request.
* **Test Set**: a type-less collection of `input + expected output` items (from requests, workflow runs, agent tasks, or authored by hand) for regression testing.
* **Test Run**: executes a Test Set against a chosen consumer (a model, workflow, or agent) and scores it with evaluation criteria.
* **Evaluation Criteria**: reusable checks (deterministic, LLM-judge, or agent-trajectory) that score a Test Run.

See:

* [Sentinels](/observe/sentinels/overview)
* [Test Sets](/test/test_sets/creation)
* [Test Run Execution](/test/test_sets/execution)
* [Evaluation Criteria](/test/test_sets/criteria)

## Auto-creation on first request

When Maitai receives a request with a new `application` / `intent` pair for your company, it will create the corresponding Application/Intent resources so they show up in the Portal automatically.