Create an orchestration AI agent that hands the voice contact to Agentforce on the first turn, with Agentforce using its own voice. Save the agent configuration and create the agent. Connect streams PCM audio at 8 kHz, 16 kHz, or 24 kHz.
-
Create the AI agent.
awscurl --service wisdom --region <REGION> --profile <AWS_PROFILE> -X POST \ --header 'Content-Type: application/json' \ --data '{ "name": "agentforce-voice-orchestrator", "type": "ORCHESTRATION", "visibilityStatus": "PUBLISHED", "configuration": { "orchestrationAIAgentConfiguration": { "connectInstanceArn": "arn:aws:connect:<REGION>:<ACCOUNT_ID>:instance/<CONNECT_INSTANCE_ID>", "toolConfigurations": [ { "toolName": "Complete", "toolType": "RETURN_TO_CONTROL", "inputSchema": { "type": "object", "properties": {}, "additionalProperties": false } } ], "multiAgentConfigurations": [ { "handoffAgentConfiguration": { "agentTarget": { "applicationId": "<APPLICATION_ID>" }, "instruction": { "instruction": "Hand off the voice conversation to Agentforce. Agentforce uses its own voice for the customer conversation." }, "audioStreamingEnabled": true, "immediateHandoff": true } } ] } } }' \ https://wisdom.<REGION>.amazonaws.com/assistants/<ASSISTANT_ID>/aiagents -
Associate the security profile with the unversioned agent. After you publish a
version, associate the profile with the versioned ARN as well.
aws connect associate-security-profiles \ --instance-id <CONNECT_INSTANCE_ID> \ --entity-arn <AI_AGENT_ARN> \ --entity-type AI_AGENT \ --security-profiles Id=<SECURITY_PROFILE_ID> \ --region <REGION>