Implementation:BerriAI Litellm Enterprise Custom Guardrail

Attribute	Value
Sources	enterprise/litellm_enterprise/integrations/custom_guardrail.py
Domains	Guardrails, Tag-Based Routing, Enterprise
Last Updated	2026-02-15 16:00 GMT

Overview

EnterpriseCustomGuardrailHelper provides enterprise-only logic to determine whether a guardrail should run based on request tags and guardrail Mode configuration.

Description

This class contains a single static method, _should_run_if_mode_by_tag, which implements tag-based guardrail execution control. This is an enterprise-only feature that enables fine-grained control over which requests trigger which guardrails based on request tags.

How It Works:

1. The method checks if the event_hook is an instance of Mode (as opposed to a simple GuardrailEventHooks enum). If not, it returns None (no tag-based decision).
2. It extracts request tags from litellm_params and proxy_server_request using StandardLoggingPayloadSetup._get_request_tags.
3. It checks if any request tag matches the guardrail's Mode.tags list.
4. If no tag matches, it checks if any request tag matches the Mode.default list as a fallback.
5. Returns True if a match is found, False otherwise.

A premium user license is required; an exception is raised if the user is not premium.

Usage

This helper is called internally by the CustomGuardrail.should_run_guardrail method when a guardrail's event hook is configured with Mode (tag-based) instead of simple event hooks.

Code Reference

Source Location

enterprise/litellm_enterprise/integrations/custom_guardrail.py

Signature

class EnterpriseCustomGuardrailHelper:
    @staticmethod
    def _should_run_if_mode_by_tag(
        data: dict,
        event_hook: Optional[Union[GuardrailEventHooks, List[GuardrailEventHooks], Mode]],
    ) -> Optional[bool]: ...

Import

from litellm_enterprise.integrations.custom_guardrail import EnterpriseCustomGuardrailHelper

I/O Contract

Inputs

Parameter	Type	Description
data	dict	Request data containing proxy_server_request and litellm params.
event_hook	Optional[Union[GuardrailEventHooks, List[GuardrailEventHooks], Mode]]	The guardrail's event hook configuration. Must be Mode for tag-based evaluation.

Outputs

Output	Type	Description
Tag match result	Optional[bool]	True if a request tag matches; False if no match; None if event_hook is not a Mode instance.

Usage Examples

# Guardrail config with tag-based mode
guardrails:
  - guardrail_name: "my-guardrail"
    litellm_params:
      guardrail: aporia
      mode:
        tags: ["production", "high-risk"]
        default: ["standard"]

# Request with tags that trigger the guardrail
response = litellm.completion(
    model="gpt-4",
    messages=[{"role": "user", "content": "Hello"}],
    metadata={"tags": ["production"]},
)

Related Pages

• Implementation:BerriAI_Litellm_Aporia_AI_Guardrail -- Aporia AI guardrail that uses this helper
• Implementation:BerriAI_Litellm_LLM_Guard -- LLM Guard integration
• Implementation:BerriAI_Litellm_Secret_Detection -- Secret detection guardrail