Amazon Web Services (AWS) has released "Strands Harness," an open-source tool that makes it easy to build custom AI agents based on major Large Language Models (LLMs) and LLM runtimes.
NewsAmazon Web ServicesStrands harness
AWS Releases "Strands Harness" as Open Source to Assist in AI Agent Construction
This article is a translation. Read the Japanese original
AI agents perform tasks autonomously by combining LLMs with a surrounding environment referred to as a "harness." This harness includes mechanisms for remembering instructed goals, access permissions to external environments such as files, networks, databases, and APIs, and guardrails to control behavior.
Strands Harness supports a variety of LLMs, including models on Amazon Bedrock, Anthropic's Claude, OpenAI's GPT, Google's Gemini, and Ollama. It can be constructed with just a few lines of code, and built agents come standard with capabilities such as shell execution, file reading/writing, and web searching.
Furthermore, it features prompt caching and context management functions. For example, it automatically handles tasks such as truncating tool execution results when they exceed a certain amount, compressing information via summarization when the context window exceeds 85%, and recovering when an overflow occurs.
This tool is not dependent on specific LLMs or deployment destinations and can be deployed in any environment equipped with a Linux container. Additionally, constructed agents can be exported as TypeScript or Python code. As a benchmark result, AWS demonstrates that agents using Strands Harness can achieve higher capabilities at a lower cost compared to Claude Code.
Sources
- AWS、AIエージェントを自作できるツール「Strandsハーネス」をオープンソースで公開 特定のLLMに依存せず入れ替え可能 （ITmedia AI+、2026-09-29）