AWS Bedrock Agents¶
AWS Bedrock Agents allow you to make both Bedrock Agents and Bedrock AgentCore Agents available as a standard Dataiku Agent (represented by a pink diamond in the Dataiku Flow). Once configured, it can be leveraged anywhere a Dataiku Agent is supported, including:
The Dataiku Flow (notably Prompt Recipes)
Chat UIs such as Agent Hub
The LLM Mesh API
/agents/agent-review
/agent/agent-evaluation
Key Features¶
Real-time streaming of agent responses to provide an interactive user experience is supported
DSS automatically extracts and displays citations from Bedrock Knowledge Banks and information from Action Group (tool) invocations.
Authentication to AWS Bedrock is fully managed through the use of an existing S3 connection
Prerequisites¶
AWS Bedrock Agents support is provided by the “AWS Bedrock Agents” plugin, which you need to install
AWS Requirements¶
Bedrock Agent
An existing agent must be created and configured in the AWS Console.
You must have the Bedrock Agent ID and Alias ID (not to be confused with Agent Name and Alias Name. The ID are numeric)
The identity configured in the linked S3 connection must have the
bedrock:InvokeAgentpermission.
Bedrock AgentCore
An existing agent must be created and configured in the AWS Console.
You must have the Bedrock AgentCore runtime ARN, and optionally, qualifier
The identity configured in the linked S3 connection must have the
bedrock:InvokeAgentpermission.
Usage (Bedrock Agent)¶
Navigate to your Dataiku project.
Go to Agents & GenAI Models > + New Agent.
Select Bedrock Agent.
Select the AWS region and S3 connection to use for authentication
Enter the specific identifiers for your Bedrock agent.
You can then use the built-in chat interface within the Agent configuration page to test your agent.
Usage (Bedrock AgentCore)¶
Navigate to your Dataiku project.
Go to Agents & GenAI Models > + New Agent.
Select Bedrock Agent.
Select the AWS region and S3 connection to use for authentication
Enter the runtime ARN for your AgentCore agent
You can then use the built-in chat interface within the Agent configuration page to test your agent.
Cautions¶
In order to properly support multi-turn conversation, when using the LLM Mesh API, you need to pass a stable
conversationIdin the request context. Note that when using the built-in chat, or Agent Hub, this is autoamtically handledBedrock Agents using Action Groups that require user confirmation (returning control to the user) are not supported.
Only text-only interactions are supported. Image inputs are not supported