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

# Anthale Glossary for Policies and Guardrails

> Look up the Anthale terms used across the documentation, including organizations, policies, guardrails, enforcement requests, actions, and trust boundaries.

This glossary defines Anthale terms as they are used across tutorials, guides, explanations, and reference pages.

## Core Anthale terms

| Term                  | Meaning                                                                                                                                                                                                                                                    |
| --------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `Organization`        | The account boundary for Anthale resources such as policies and API keys.                                                                                                                                                                                  |
| `API key`             | The server-side credential your application uses to call Anthale.                                                                                                                                                                                          |
| `Policy`              | The configuration contract that decides which guardrails run and how their results are enforced.                                                                                                                                                           |
| `Guardrail`           | A risk-specific evaluator such as prompt injection protection, data leakage prevention, content moderation, topic control, or link control.                                                                                                                |
| `Enforcement request` | The runtime API call that evaluates content against a policy.                                                                                                                                                                                              |
| `Action`              | The runtime result Anthale returns: `allow`, `detect`, `redact`, or `block`. When Anthale returns `redact`, downstream systems should use the sanitized content. When Anthale returns `block`, the request or response should stop at that trust boundary. |
| `Trust boundary`      | A point where Anthale should usually run, such as before model input, before retrieved context enters a prompt, or before model output reaches a user or downstream system.                                                                                |

## Runtime path terms

| Term                | Meaning                                                                                             |
| ------------------- | --------------------------------------------------------------------------------------------------- |
| `input path`        | The path where content is evaluated before it reaches a model or other downstream step.             |
| `output path`       | The path where content is evaluated before it reaches a user, tool, or downstream system.           |
| `retrieved context` | External content retrieved for a prompt, such as search results, documents, or connector output.    |
| `tool output`       | Content returned by a tool or integration that may be fed back into a model or shown to a user.     |
| `Metadata`          | Application context attached to an enforcement request for tracing, correlation, and investigation. |

## Evaluation terms

| Term         | Meaning                                                                                                                                                                             |
| ------------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `Evaluation` | The result Anthale returns for a guardrail during policy execution, including evidence, metadata, and zero or more threads.                                                         |
| `Thread`     | A finding attached to an evaluation. A single evaluation can contain zero threads when nothing is found, or multiple threads when Anthale finds several issues in the same content. |

## Next steps

Read [What Is Anthale](/docs/) for the product model, [How Anthale Works](/docs/how-anthale-works) for the runtime flow, or [Policies](/docs/learn/policies) for the configuration contract.