Hello.
I’m working on implementing a native ML setup in OpenSearch, which involves a conversational agent linked to an external LLM (a custom one)
For our unique architectural needs, I need to pass along the OpenSearch user’s identity [ in the http header ] . I’m having trouble locating the correct variable (assuming there is one) to specify in the connector.
@LeHeron I think the most universal way would be calling _plugins/_security/authinfo endpoint with authenticated user.
What type of authentication do you use?
Hi @LeHeron, we can specify API key or AWS credentials for the authentication when creating the connector. You can refer to this doc for more information Connectors - OpenSearch Documentation. Can you share more details about your use case, including the authentication method and the current connector request payload you’re using?
Thank you for your responses and suggestions, @pablo and @nathhjo.
I believe there might be a slight misunderstanding regarding my need, and I apologize if my initial question lacked clarity.
My goal is not to authenticate the OpenSearch user to the external LLM. (authentication is already handled at a different architectural level and isn’t the issue here)
What I’m trying to do is pass the OpenSearch user’s identity (the “who” is making the request to the LLM) - undertone “who is using the Conversational Agent” - in the HTTP request sent to my custom LLM via the connector. In other words, I want the LLM to know which OpenSearch user originated the request, and I’m looking for the variable or method to retrieve this information within OpenSearch to include it in the connector’s request_payload.
As usage of Agent/Connector/Models/.. are already subject to permission management [cf. userroles, backendroles ], i guess this information already available somewhere. my point is to know if that one could be export
I hope this clarifies my request a bit better. Please let me know if you need more details.