Skip to content

Latest commit

 

History

History
78 lines (73 loc) · 2.71 KB

bedrock.md

File metadata and controls

78 lines (73 loc) · 2.71 KB

Bedrock

AWS's bedrock service · portable interface driver.Bedrock · AWS index

Operations (65)

Operation Description
ApplyGuardrail
Converse
CountTokens
CreateAutomatedReasoningPolicy
CreateEvaluationJob
CreateFoundationModelAgreement
CreateGuardrail
CreateGuardrailVersion
CreateInferenceProfile
CreateMarketplaceModelEndpoint
CreateModelCopyJob
CreateModelCustomizationJob
CreateModelImportJob
CreatePromptRouter
CreateProvisionedModelThroughput
DeleteAutomatedReasoningPolicy
DeleteCustomModel
DeleteFoundationModelAgreement
DeleteGuardrail
DeleteInferenceProfile
DeleteMarketplaceModelEndpoint
DeleteModelInvocationLoggingConfiguration
DeletePromptRouter
DeleteProvisionedModelThroughput
DeregisterMarketplaceModelEndpoint
GetAsyncInvoke
GetAutomatedReasoningPolicy
GetCustomModel
GetEvaluationJob
GetFoundationModel
GetFoundationModelAvailability
GetGuardrail
GetInferenceProfile
GetMarketplaceModelEndpoint
GetModelCopyJob
GetModelCustomizationJob
GetModelImportJob
GetModelInvocationLoggingConfiguration
GetPromptRouter
GetProvisionedModelThroughput
InvokeModel
ListAsyncInvokes
ListAutomatedReasoningPolicies
ListCustomModels
ListEvaluationJobs
ListFoundationModelAgreementOffers
ListFoundationModels
ListGuardrails
ListInferenceProfiles
ListMarketplaceModelEndpoints
ListModelCopyJobs
ListModelCustomizationJobs
ListModelImportJobs
ListPromptRouters
ListProvisionedModelThroughputs
ListTagsForResource
PutModelInvocationLoggingConfiguration
RegisterMarketplaceModelEndpoint
StartAsyncInvoke
StopEvaluationJob
TagResource
UntagResource
UpdateAutomatedReasoningPolicy
UpdateGuardrail
UpdateMarketplaceModelEndpoint

Not in scope

  • ApplyGuardrail, and the guardrails the agent runtime applies, evaluate only a guardrail's custom denied words (case-insensitive whole words or phrases): a match returns GUARDRAIL_INTERVENED with the blocked input or output message. Managed word lists, content filters, denied topics, PII and regex filters, contextual grounding and automated reasoning are stored but not evaluated, and no assessments are returned.