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

# Core Concepts

> Key concepts in the MeshAI Agent Control Plane

## Tenant

A tenant is your organization in MeshAI. All data is isolated per tenant. Agents, policies, audit events, and billing are scoped to your tenant.

## Agent

An AI agent registered in MeshAI. Each agent has:

* **Name** and **description**
* **Framework** (CrewAI, LangChain, AutoGen, custom)
* **Model provider** and **model name**
* **Environment** (production, staging, dev)
* **Status** (healthy, degraded, down, unknown)

## API Key

Bearer token (`msh_...`) used to authenticate with the MeshAI API. Two types:

* **Admin keys**: full access (manage agents, policies, billing)
* **Proxy keys**: forward-only access (route LLM requests through the proxy)

## Policy

A governance rule enforced in real-time through the proxy. Eight types:

* `model_allowlist`: restrict which models agents can use
* `block_provider`: block specific LLM providers
* `require_approval`: require human approval (HITL)
* `budget_limit`: enforce monthly spend limits
* `rate_limit`: additional request throttling
* `require_human_review`: flag for periodic review
* `prompt_injection` - detect and block prompt injection attempts
* `pii_filter` - detect and redact PII in requests or responses

## Risk Classification

EU AI Act risk level assigned to each agent:

* **Minimal**: low risk, basic monitoring
* **Limited**: transparency obligations
* **High**: full compliance required (audit trail, HITL, FRIA)
* **Unacceptable**: should not be deployed without safeguards

## Audit Event

Immutable log entry recording a governance action (agent registered, anomaly acknowledged, policy created, approval decided). Required by EU AI Act Article 12.

## Readiness Score

Compliance score (0-120) across 7 components measuring your organization's EU AI Act readiness.