Building a Context‑Aware AI Assistant with AgentCore and OpenClaw
AWS shows how to add durable memory to a personal assistant using AgentCore runtime, OpenClaw, and Claude models on Bedrock.
Original source published: October 6, 2026
Off‑the‑shelf assistants answer individual questions well but lack continuity. AWS demonstrates a personal assistant that accumulates context using OpenClaw on the AgentCore runtime, with AgentCore memory turning disposable chats into durable knowledge.
The solution is packaged in a single CloudFormation template and connects Telegram webhooks, API Gateway, Lambda functions, EventBridge, and Amazon S3. Two Claude models on Bedrock are routed by task—Claude Haiku 4.5 for text and Claude Sonnet 4.5 for vision. AgentCore memory records each turn as a short‑term event and extracts long‑term records classified as USER_PREFERENCE, SEMANTIC, or SUMMARIZATION, stored in per‑user namespaces. Retrieval caps at 50 results with a 3‑second timeout and ranks explicit preferences before inferred facts.
Running on consumption‑based pricing, light personal use is estimated at a few dollars per month. The architecture is domain‑agnostic, letting learners adapt the same pipeline for support bots, fitness coaches, or other assistants.