AWS open-sources SMART: deploy a WhatsApp AI agent on Bedrock AgentCore in one command

AWS ML Blog · rss · 2026-11-18

AWS's ML blog introduces SMART (Social Messaging Agent Runtime Toolkit), an open-source Python SDK plus Terraform stack that connects agents running on Amazon Bedrock AgentCore to WhatsApp, deploying the full infrastructure with one command.

The post argues the hard part of a WhatsApp agent is the channel layer, not the agent: inbound messages arrive as deeply nested typed payloads, voice/images/documents each need pipelines, and you must handle cross-session memory, durable schedules, and Meta's pre-approved templates for the 24-hour messaging window—over 10 AWS services if built from scratch. SMART handles webhook verification, full media lifecycle, YAML-declared templates, AgentCore memory binding, EventBridge scheduling, and local CLI testing.

AgentCore is framework-agnostic: Strands Agents gets memory wired automatically, while LangGraph and Pydantic-AI can integrate too. Brazil's Webmotors compressed a two-month engagement to six weeks and reached controlled production. The post includes a complete agent.py example and architecture details (SNS ingress, Lambda channel logic, Transcribe voice notes, default Bedrock Guardrails).

Original post →

More from coding & agent

coding & agent channel →