Deploy voice agents with Pipecat and Amazon Bedrock AgentCore Runtime – Part 1

AWS and Pipecat have collaborated to create a deployment framework for building and scaling intelligent voice agents on Amazon Bedrock's new AgentCore Runtime. This serverless environment is specifically designed for dynamic AI agents, addressing critical challenges in voice AI deployment like low-latency streaming, security isolation, and unpredictable scaling. Each conversation session runs in isolated microVMs, provides built-in observability for tracing agent reasoning, and supports continuous sessions lasting up to 8 hours—ideal for long customer support interactions. The runtime auto-scales with traffic spikes and uses a pay-for-what-you-use model to prevent cost overruns from idle infrastructure.

Developers can package their Pipecat voice pipelines as ARM64 (Graviton) containers and deploy them directly to AgentCore Runtime with minimal setup. The framework supports three key network transport approaches: WebSockets for real-time API connections to speech models, WebRTC for browser and mobile applications, and telephony integration through SIP transfer for contact center calls. This architecture enables bidirectional audio streaming with sub-second response times by leveraging latency-optimized models like Amazon Nova Sonic. The solution specifically targets use cases where natural conversational flow is critical, including customer support, virtual assistants, and outbound campaigns where even small delays can break user experience.