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

This article explains how to deploy voice agents using Pipecat and Amazon Bedrock AgentCore Runtime, focusing on network transport approaches for real-time conversations.

• AgentCore Runtime provides secure, serverless infrastructure with auto-scaling and isolated microVMs for voice agents
• Low-latency streaming is critical; conversations require sub-one-second end-to-end responses for natural flow
• WebSockets: simplest approach for prototyping; good performance, easy implementation
• WebRTC with TURN: improved performance using UDP; requires VPC configuration and TURN servers
• Amazon KVS managed TURN: AWS-native alternative providing auto-rotating credentials without third-party dependencies
• Managed WebRTC providers: offer globally distributed infrastructure, analytics, and multi-participant support
• Telephony integration: enables traditional PSTN calls for contact centers and IVR replacement
• AgentCore Runtime requires ARM64 (Graviton) containers for deployment

The post provides practical guidance for selecting transport layers based on latency, reliability, and use case requirements, with code samples for each approach.