Knowledge base deployment

Question

All,
In accordance with “Deploy a knowledge base documentation” :
*to deploy a knowledge base in your application, set it up to make Retrieve or RetrieveAndGenerate requests to the knowledge base. To see how to use these API operations, select the API tab in Test a knowledge base in Amazon Bedrock.*

In the “Use the knowledge base” documentation states 
*“Integrate your knowledge base into your application as is or add it to agents”
*
Is it **recommended/supported to use direct KB calls **(Retrieve API) or/and KB with response generation (RetrieveAndGenerate API) in production situation if agent functionality is not required?

Additionally, I do not see how to enable versioning for response generation – it is not available in case of direct KB (at least I do not see it a console)

Obviously, another way is to create an agent, associate it with KB and use alias for versioning. However, it looks like agent cannot use metadata filtering.
Thank you for your help
YK

Kazmi, Ayman · Accepted Answer

Hello YK,

Yes, it is recommended to use direct knowledge base calls (Retrieve API) or knowledge base with response generation (RetrieveAndGenerate API) in production situations if you don't require the full agent functionality. Direct knowledge base calls can be more lightweight and potentially faster, as they don't involve the overhead of the agent layer.

Regarding versioning for response generation, you're correct that versioning is not available when using the knowledge base directly. Versioning is a feature provided by the agent layer, which allows you to create and manage different versions of your agent's behavior.

If you need versioning for your knowledge base with response generation, the recommended approach is to create an agent, associate it with your knowledge base, and then use the agent's alias for versioning. This way, you can create new versions of your agent's behavior, which includes the knowledge base and any additional configurations or customizations you've made.

Thanks,
Ayman

Knowledge base deployment

Add your answer

Relevant content