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

# Guardrails

> Spending limits, model and provider restrictions, prompt-injection and sensitive-info policies for a group of keys.

A guardrail is a named set of policies applied to API keys in a [workspace](https://nemonist.com/docs/concepts/workspaces.md). The default guardrail applies to every key in the workspace; others apply to the keys you assign. A key keeps every guardrail that applies to it.

Create them on the workspace's **Guardrails** page.

| Policy | What it does |
| - | - |
| Budget | A spending limit that applies to each assigned key separately. |
| Models and providers | Allow or block models and providers. Empty allow lists allow everything. |
| Zero data retention | Only route to provider endpoints that do not store your data. Requests no such endpoint can serve are refused. |
| Prompt injection | Requests that match prompt-injection patterns in user inputs are refused with `403`. No added latency. |
| Sensitive info | Find email addresses, phone numbers, Social Security numbers, credit card numbers, IP addresses, API keys and secrets, or your own patterns, in user inputs. **Redact** replaces matches before the model sees them; **block** refuses the request. |

Each guardrail has a preview: see what the model would receive, or whether a request is blocked.

On the Enterprise plan, an owner can **lock** a policy so admins cannot loosen or remove it.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.