[Feature]: Logging the Bedrock guardrail response in DB #13690
Description
The Feature
Capture and log complete Guardrail api response in DB.
At the moment, this information is missing from the request metadata in litellm.
Example response from Bedrock Guardrail API:
{
"ResponseMetadata": {
"RequestId": "e548ae43-6556-4d94-9b7c-.......",
"HTTPStatusCode": 200,
"HTTPHeaders": {
"date": "Fri, 15 Aug 2025 14:00:32 GMT",
"content-type": "application/json",
"content-length": "1079",
"connection": "keep-alive",
"x-amzn-requestid": "e548ae43-6556-4d94-9b7c-......."
},
"RetryAttempts": 0
},
"usage": {
"topicPolicyUnits": 0,
"contentPolicyUnits": 1,
"wordPolicyUnits": 1,
"sensitiveInformationPolicyUnits": 1,
"sensitiveInformationPolicyFreeUnits": 0,
"contextualGroundingPolicyUnits": 0
},
"action": "GUARDRAIL_INTERVENED",
"actionReason": "Guardrail blocked.",
"outputs": [
{
"text": "Sorry, I cannot answer this question."
}
],
"assessments": [
{
"contentPolicy": {
"filters": [
{
"type": "INSULTS",
"confidence": "HIGH",
"filterStrength": "HIGH",
"action": "BLOCKED",
"detected": true
}
]
},
"invocationMetrics": {
"guardrailProcessingLatency": 270,
"usage": {
"topicPolicyUnits": 0,
"contentPolicyUnits": 1,
"wordPolicyUnits": 1,
"sensitiveInformationPolicyUnits": 1,
"sensitiveInformationPolicyFreeUnits": 0,
"contextualGroundingPolicyUnits": 0
},
"guardrailCoverage": {
"textCharacters": {
"guarded": 36,
"total": 36
}
}
}
}
],
"guardrailCoverage": {
"textCharacters": {
"guarded": 36,
"total": 36
}
}
}
Motivation, pitch
We require some of the fields emitted in the Bedrock ApplyGuardrail API response, for governance.
https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_ApplyGuardrail.html
For example, being able to log and monitor the type of Guardrail policy that was detected (Eg. insults) and the type of action that has been taken (eg. blocked).
LiteLLM is hiring a founding backend engineer, are you interested in joining us and shipping to all our users?
No
Twitter / LinkedIn details
No response