product updateAmazon Web Services

AWS Releases AgentCore Platform for Building Voice AI Agents with Nova 2 Sonic Integration

TL;DR

Amazon has released Bedrock AgentCore, a platform for building and deploying AI agents with microVM isolation for secure session handling. The platform integrates with Amazon Nova 2 Sonic, a speech-to-speech foundation model, and supports the Model Context Protocol (MCP) for connecting agents to backend services.

2 min read
0

AWS Releases AgentCore Platform for Building Voice AI Agents with Nova 2 Sonic Integration

Amazon has released Bedrock AgentCore, a platform for building, deploying, and operating AI agents at scale. The platform features microVM isolation for user sessions and integrates with Amazon Nova 2 Sonic, a speech-to-speech foundation model available through Amazon Bedrock.

Platform Architecture

Bedrock AgentCore consists of two main components:

AgentCore Runtime hosts AI agents with microVM isolation, according to AWS. Each user session runs in an isolated virtual machine to prevent cross-session interference and data access. The runtime supports WebSocket protocol for real-time bidirectional audio streaming.

AgentCore Gateway provides standardized communication between agents and backend services using the Model Context Protocol (MCP), an open standard for connecting AI applications to external data sources and tools. The gateway exposes backend REST APIs as MCP tools that agents can discover and invoke.

Nova 2 Sonic Integration

The platform integrates with Amazon Nova 2 Sonic for real-time voice interactions. The model processes speech-to-speech without intermediate text conversion. When a user speaks an order, the agent processes voice input through Nova 2 Sonic, invokes backend tools asynchronously through AgentCore Gateway, and streams contextualized voice responses back over the WebSocket connection.

Reference Implementation

AWS published a reference implementation for an omnichannel ordering system on GitHub. The implementation uses AWS CDK for infrastructure deployment and includes:

  • Amazon Cognito for OAuth 2.0 authentication
  • API Gateway for REST endpoint exposure with IAM authorization
  • Lambda functions for business logic execution
  • DynamoDB for data storage with single-digit millisecond latency
  • AWS Location Services for pickup recommendations
  • AWS Amplify for frontend hosting

The authentication flow uses SigV4-signed WebSocket connections. Users authenticate with Cognito to receive JWT tokens, exchange ID tokens for temporary AWS credentials, and establish WebSocket connections to AgentCore Runtime. The runtime validates access tokens by calling the Cognito GetUser API and extracts verified customer context.

Requirements

The platform requires foundation model access for Amazon Nova 2 Sonic in Amazon Bedrock, Node.js 20.x or later, Python 3.13 or later, AWS CLI 2.x, AWS CDK CLI 2.x, and Boto3 1.38.0 or later for bedrock-agentcore-control service support.

What This Means

Bedrock AgentCore addresses operational challenges in deploying voice AI at scale by providing managed infrastructure with automatic scaling and session isolation. The microVM approach prevents the "noisy neighbor" problem common in multi-tenant systems where one customer's load affects others. The MCP integration standardizes how agents connect to backend services, reducing the integration code needed when modifying backends or adding tools. However, AWS has not disclosed pricing for AgentCore Runtime or Gateway services, making total cost of ownership unclear for production deployments.

Related Articles

product update

AWS Overhauls Bedrock AgentCore Runtime, Cuts Cold Starts to Flat 2 Seconds Regardless of Image Size

Amazon has released an updated Bedrock AgentCore runtime that holds cold start latency at roughly 2 seconds regardless of container image size, versus up to 30 seconds on the previous version. The update also changes memory billing to track real-time usage instead of peak allocation.

product update

GitHub Copilot Adds Whole-Codebase Indexing for Faster C++ Code Intelligence

GitHub has updated Copilot CLI's C++ code intelligence with whole codebase indexing, targeting performance in large repositories with millions of lines of code. The change addresses navigation and context challenges specific to deeply interconnected C++ source trees.

product update

Rabbit Launches OS3, a Cloud-Based AI Agent That Runs Without the R1 Device

Rabbit is rolling out OS3, a standalone AI agent that runs across Windows, Mac, and Linux devices without requiring its R1 hardware. The company has stopped manufacturing the R1 and is instead building a new 'cyberdeck' device to run OS3.

product update

OpenAI Lists GPT-6 Luna Pro: A High-Reasoning Mode for Its Budget GPT-6 Model, Not a New Checkpoint

GPT-6 Luna Pro, listed on OpenRouter with a Sep 22, 2026 release date, is not a distinct model but GPT-6 Luna run with reasoning.mode set to 'pro' for higher-quality outputs on complex tasks. It carries a 1.1M token context window and costs $0.10 per 1M input tokens and $0.50 per 1M output tokens under standard routing.

Comments

Loading...