Implementation:BerriAI Litellm Aporia AI Guardrail

Attribute	Value
Sources	enterprise/enterprise_hooks/aporia_ai.py
Domains	Guardrails, Content Safety, Enterprise Hooks
Last Updated	2026-02-15 16:00 GMT

Overview

AporiaGuardrail is a guardrail integration that validates LLM request prompts and responses against Aporia AI's content safety policies, blocking requests or responses that violate configured guardrail rules.

Description

The AporiaGuardrail class extends CustomGuardrail and integrates with the Aporia AI validation API. It provides two hook points in the LLM request lifecycle:

• During-Call Moderation (async_moderation_hook) -- Validates the prompt messages before the LLM call. Transforms messages to Aporia-compatible roles (system, user, assistant, or "other" for unsupported roles like "tool").
• Post-Call Validation (async_post_call_success_hook) -- Validates the LLM response alongside the original prompt. Sets validation_target to "both" when both messages and response are present.

When Aporia returns an action of "block", the guardrail raises an HTTPException with status 400 and the full Aporia response.

The guardrail name is "aporia". It respects the should_run_guardrail event hook system and legacy should_proceed_based_on_metadata for backwards compatibility.

Requires either constructor parameters or environment variables: APORIO_API_KEY and APORIO_API_BASE.

Usage

Register AporiaGuardrail as a guardrail callback in the LiteLLM proxy to enforce Aporia AI content safety policies on prompts and responses.

Code Reference

Source Location

enterprise/enterprise_hooks/aporia_ai.py

Signature

class AporiaGuardrail(CustomGuardrail):
    def __init__(self, api_key: Optional[str] = None, api_base: Optional[str] = None, **kwargs): ...
    def transform_messages(self, messages: List[dict]) -> List[dict]: ...
    async def prepare_aporia_request(self, new_messages: List[dict], response_string: Optional[str] = None) -> dict: ...
    async def make_aporia_api_request(self, new_messages: List[dict], response_string: Optional[str] = None): ...
    async def async_post_call_success_hook(self, data: dict, user_api_key_dict: UserAPIKeyAuth, response): ...
    async def async_moderation_hook(self, data: dict, user_api_key_dict: UserAPIKeyAuth, call_type: CallTypesLiteral): ...

Import

from enterprise.enterprise_hooks.aporia_ai import AporiaGuardrail

I/O Contract

Inputs

Parameter	Type	Description
api_key	Optional[str]	Aporia API key. Falls back to APORIO_API_KEY env var.
api_base	Optional[str]	Aporia API base URL. Falls back to APORIO_API_BASE env var.
data	dict	Request data containing messages list.
response	LLM response object	The LLM response for post-call validation.

Outputs

Output	Type	Description
Pass-through	None	Request proceeds if Aporia action is "passthrough", "modify", or "rephrase".
Block	HTTPException(400)	Raised when Aporia action is "block", with aporia_ai_response in detail.

Usage Examples

# In proxy config YAML
litellm_settings:
  guardrails:
    - guardrail_name: "aporia"
      litellm_params:
        guardrail: aporia
        api_key: "os.environ/APORIO_API_KEY"
        api_base: "os.environ/APORIO_API_BASE"
        mode: "during_call"

# Programmatic instantiation
from enterprise.enterprise_hooks.aporia_ai import AporiaGuardrail

guardrail = AporiaGuardrail(
    api_key="your-aporia-api-key",
    api_base="https://gr-prd-trial.aporia.com/your-id",
)

Related Pages

• BerriAI_Litellm_LLM_Guard -- LLM Guard content moderation integration
• BerriAI_Litellm_Secret_Detection -- Secret detection guardrail
• BerriAI_Litellm_Enterprise_Custom_Guardrail -- Tag-based guardrail mode control