Initial Crowdstrike AIDR Guardrail Plugin #1504
Conversation
|
|
||
| const url = `${parameters.credentials.baseUrl}/v1/guard_chat_completions`; | ||
| const target = eventType === 'beforeRequestHook' ? 'request' : 'response'; | ||
| const json = context[target].json; |
There was a problem hiding this comment.
Since the whole request body is being passed without any transformation, this would only work for chat completions API but it might fail for OpenAI responses API and Anthropic messages API format if the Crowdstrike guardrails API does not support them.
Are responses and messages format supported? I do not see them in the documentation
There was a problem hiding this comment.
Marking this as a known limitation and merging this for now.
There was a problem hiding this comment.
@VisargD Thanks, I'll look into it on our end. Right now the endpoint supports any kind of JSON, but it does implicitly expect the format to be the OpenAI completions AI format and we do have special logic for the messages array which might be lost.
2 participants
Description: (required)
This is effectively an update for the Pangea Guardrails plugin. First, it is a rebrand, since Pangea has been acquired by crowdstrike, and has a new product name
It's updated for much newer APIs. The current APIs support blocking and redaction based on configured upstream rules. The new endpoint is designed to work with structued LLM inputs, e.g. you should pass it the entire
messagesarray, but also supports havingtoolsand other properties a well.I think we should consider removing the Pangea guardrails as well.
Tests Run/Test cases added: (required)
Type of Change: