Query a document without creating a knowledge base - Amazon Bedrock
Services or capabilities described in AWS documentation might vary by Region. To see the differences applicable to the AWS European Sovereign Cloud Region, see the AWS European Sovereign Cloud User Guide.

Query a document without creating a knowledge base

To query a single document without creating or configuring a knowledge base, use the RetrieveAndGenerate API operation with an external source. This workflow is available through the Agents for Amazon Bedrock Runtime API and AWS SDKs, rather than the Amazon Bedrock console.

The operation retrieves information from the document, uses a foundation model or inference profile to generate a response, and returns citations to the source. You can provide the document from an Amazon S3 location or include its byte content in the request.

Note

You can include exactly one document in each request. If you include the document as byte content, the maximum file size is 10 MB.

You can't use a reranker model when querying an external source.

Before you send a request, configure the permissions described in Permissions to query external sources.

To query an external document

  1. Prepare the document in one of the following ways:

    • Upload the document to Amazon S3.

    • Base64-encode the document to include it in the request.

  2. Send a RetrieveAndGenerate request and configure the following fields:

    • Specify your query in input.text.

    • Set retrieveAndGenerateConfiguration.type to EXTERNAL_SOURCES.

    • In externalSourcesConfiguration, specify the model or inference profile ARN and the document source.

  3. Use the output and citations fields from the response. For subsequent requests in the same conversation, include the sessionId returned by the first response.

The following example queries a document stored in Amazon S3:

{
    "input": {
        "text": "your query"
    },
    "retrieveAndGenerateConfiguration": {
        "type": "EXTERNAL_SOURCES",
        "externalSourcesConfiguration": {
            "modelArn": "model-or-inference-profile-arn",
            "sources": [
                {
                    "sourceType": "S3",
                    "s3Location": {
                        "uri": "s3://bucket-name/document-key"
                    }
                }
            ]
        }
    }
}

To include the document in the request instead, set sourceType to BYTE_CONTENT and replace s3Location with a byteContent object. Specify the file name in identifier, the MIME type in contentType, and the base64-encoded document in data.

You can optionally use generationConfiguration to customize the prompt, inference parameters, performance configuration, or guardrail. For field details, see ExternalSourcesRetrieveAndGenerateConfiguration and RetrieveAndGenerate.