AWS launches AgentCore platform for building voice AI agents with Amazon Nova 2 Sonic
AWS has released AgentCore, a new platform for hosting and running voice-based AI agents, integrated with Amazon Nova 2 Sonic for real-time speech capabilities. The platform uses the open Model Context Protocol (MCP) to connect agents to backend systems and deploys each conversation in isolated microVMs.
AWS launches AgentCore platform for building voice AI agents with Amazon Nova 2 Sonic
AWS has released AgentCore, a new platform for hosting and running voice-based AI agents integrated with Amazon Nova 2 Sonic for real-time speech processing. The platform uses the Model Context Protocol (MCP) to connect agents to backend systems, according to AWS.
Platform architecture
AgentCore consists of two main components: AgentCore Runtime, which hosts conversation logic with each call running in an isolated microVM, and AgentCore Gateway, which exposes backend APIs as MCP tools that agents can discover and invoke. The system integrates with Amazon Nova 2 Sonic to handle bidirectional speech-to-speech interaction, including transcription, turn-taking, and interruptions in a single stream.
The platform supports both text and audio as input and output formats. AWS deploys the system on Amazon ECS with AWS Fargate, with telephony integration through Amazon Chime SDK Voice Connector providing SIP trunk and toll-free number capabilities.
Technical implementation
The reference architecture AWS published demonstrates a restaurant ordering system with three distinct layers: telephony (handling phone network audio and caller identification), agent (running conversations with Amazon Nova 2 Sonic), and backend (menu, carts, orders, locations). Audio streams from the phone network through a SIP gateway to the agent layer over signed WebSocket connections.
AgentCore Runtime invokes AWS Lambda functions to build system prompts stored in AWS Systems Manager Parameter Store, customizing them based on session identifiers and customer data from Amazon DynamoDB. The platform warms agent sessions while phones are still ringing to eliminate dead air for callers.
Deployment and infrastructure
The full stack deploys through AWS CDK, with container images built via AWS CodeBuild and stored in Amazon ECR. The backend layer uses Amazon API Gateway fronting REST endpoints secured by IAM, with AWS Lambda executing business logic and Amazon DynamoDB for data storage. Amazon Location Service handles geocoding and route calculation in the reference implementation.
According to AWS, the SIP gateway runs on Amazon ECS on AWS Fargate behind a Network Load Balancer, accepting SIP invites over TCP port 5060 and allocating Real-time Transport Protocol (RTP) service ports for media reception.
What this means
AgentCore represents AWS's entry into the voice AI agent infrastructure market, competing with platforms like Retell AI and Bland AI. The use of MCP as the standard protocol for connecting agents to tools allows backend systems to change without redeploying agents, and new channels like mobile apps or kiosks can connect to the same agent without backend rewrites. The microVM isolation model addresses security and multi-tenancy concerns that have slowed enterprise adoption of voice agents. Pricing for AgentCore has not been disclosed, though it will likely bundle compute costs with Amazon Nova 2 Sonic inference charges.
Related Articles
AWS Details MCP Bridge Architecture Connecting Cloud-Hosted AI Agents to Local Files and Tools
AWS published a technical breakdown of an MCP bridge that connects cloud-hosted AI agents on Amazon Bedrock AgentCore to MCP servers running on a user's local machine. The architecture, used internally in a finance assistant with over 41,000 conversations in its first year, tunnels MCP JSON-RPC messages through WebSocket and browser native messaging.
AWS Adds Temporal Policies to Bedrock AgentCore to Stop AI Agents From Exploiting Multi-Step Actions
Amazon Bedrock AgentCore now supports temporal policies—stateful authorization rules that evaluate an AI agent's entire session history, not just individual tool calls. The feature runs at the AgentCore Gateway, outside agent code, so it cannot be bypassed by prompt manipulation or agent bugs.
AWS Details Two Paths for Single-Region Claude Code Deployments on Amazon Bedrock
AWS published a technical guide detailing two methods for keeping Claude Code inference confined to a single AWS Region: Anthropic's newer Mantle endpoint and the classic Bedrock Invoke API with application inference profiles. The right path depends entirely on which Region compliance teams require.
AWS Ships Six Agent Skills to Automate Amazon Bedrock's Automated Reasoning Policy Lifecycle
AWS published a suite of six Agent Skills that automate the full lifecycle of Amazon Bedrock Automated Reasoning policies—from rule extraction to deployment—directly from coding agents like Claude Code, Cursor, Kiro, and Codex. The skills wrap Bedrock's formal-logic verification APIs in structured workflows built on Anthropic's open Agent Skills format.
Comments
Loading...