SMART adds a WhatsApp channel to Amazon Bedrock agents
An open source Python SDK and Terraform stack deploys the WhatsApp integration layer into your own AWS account, leaving agent logic to you.
SMART puts the WhatsApp channel inside your AWS account
The AWS Machine Learning Blog published a walkthrough on 17 November 2026 for SMART, the Social Messaging Agent Runtime Toolkit, an open source Python SDK and Terraform stack aimed at teams that want to build WhatsApp AI agents on Amazon Bedrock AgentCore. It deploys the WhatsApp channel layer into your own AWS account with one command, and the post's framing is that the agent is rarely the hard part while the integration is. You bring an agent that already runs on AgentCore, and SMART covers webhook ingestion, media handling, conversation memory, scheduling and Meta message templates.
The list of work that would otherwise be yours is long. Inbound WhatsApp messages arrive as deeply nested, type-dependent payloads that must be authenticated and normalized before an agent can read them. Voice notes, images and documents each need a separate processing path. Memory has to persist across sessions, reminders need durable schedules, external systems need an authenticated route for triggering outbound messages, and anything sent more than 24 hours after the user's last message must use a pre-approved Meta template. Per the announcement on the AWS Machine Learning Blog, wiring that together means configuring more than 10 AWS services.
What the toolkit takes over
- Webhook ingestion: signature verification, nested payloads and a dozen message types, with the agent receiving a clean prompt.
- Media: download, transcribe, store, clean up and outbound upload across the full lifecycle.
- Templates and the 24-hour window: declared in YAML and synced by the CLI.
- Memory: Amazon Bedrock AgentCore memory behind a single function call.
- Scheduling and triggers: EventBridge Scheduler wiring, authenticated ingress and routing built in.
- Infrastructure: one terraform apply.
- Local development: test through the CLI without redeploying.
The contract with your agent
The boundary is deliberately small. Your agent receives the inbound message and returns text, optionally with rich message attachments. Session wiring, memory binding and delivery scheduling sit outside it. The sample agent.py in the post wraps a Strands Agents agent in BedrockAgentCoreApp and calls create_session_manager(payload), which binds AgentCore memory to the sender's phone number and keeps history for the memory retention period, seven days by default.
A second helper, socialagent_notify_context(), catches messages a tool wants to send while the agent is still working, such as a button list or a progress update, and releases them in order once the turn finishes. The post notes that sending them immediately would corrupt conversation history and break the user's next message. A socialagent init command scaffolds the project structure.
AgentCore is framework-agnostic, so the post says Strands Agents gets memory wired automatically, while LangGraph, Pydantic-AI or another framework that controls the conversation loop can integrate through the same memory and session hooks.
What changes for a team shipping one
Webmotors, cited in the post, had the work scoped as a two-month engagement, compressed delivery to six weeks and moved an internal team into controlled production ahead of a full launch. The division of labour is the point: your framework, model, prompt and business logic stay yours, while the channel becomes code deployed in your account rather than a third-party service.
Limits and open questions
The post does not give pricing, licence terms or a repository link, and it does not say whether SMART is generally available or how the Terraform stack is upgraded after the first deploy. WhatsApp Business account setup and Meta app review are not covered. The architecture section is cut off in the source, so the service layer beyond AWS End User Messaging Social, a Lambda function and AgentCore runtime, memory and observability is unclear.
Two inferences. Because the stack lands in your own AWS account, patching and Terraform drift likely become your team's job, which suggests the one-command promise is mostly about the initial deploy. And with a default retention of seven days, conversation history beyond that window likely resets unless it is configured otherwise.
What it means for building apps with AI
Most of this story is about agent infrastructure rather than app UI, but the split is familiar. If you are still shaping the agent itself, Draftship builds and hosts an app from a plain-language description with a built-in database and user sign-in, and its documentation on adding AI to your app covers the in-app side. The WhatsApp channel described here is a separate AWS deployment, and Draftship is not affiliated with AWS or Meta.
Primary sources
- Build WhatsApp AI agents with Amazon Bedrock AgentCore · AWS Machine Learning Blog, Nov 17, 2026
This article was drafted with AI assistance from the primary sources listed below and checked automatically against them. An editor hasn't reviewed it yet.