Amazon Web Services (AWS) has released "Strands harness" as open source, a tool that allows users to easily build high-performance AI agents based on major Large Language Models (LLMs) and LLM runtimes such as GPT, Claude, Gemini, and Ollama.
Product LaunchesAmazon Web ServicesStrands harness
AWS Releases "Strands Harness" Open Source for Building Custom AI Agents
This article is a translation. Read the Japanese original
AI agents are constructed by combining a "harness" (environment) that includes mechanisms for long-term memory of context—such as instructed goals and environmental settings—access to external tools like files, networks, databases, and APIs, and guardrails to prevent operations from deviating from established policies.
Strands harness supports a wide range of models, not only models on Amazon Bedrock but also Anthropic's Claude, OpenAI's GPT, Google's Gemini, and even Ollama. With just a few lines of code, users can build agents equipped with capabilities such as shell execution, file reading/writing, and web searching.
Additionally, it features prompt caching and context management capabilities. For example, when the context window (the amount of information a model can handle at once) exceeds 85%, it performs compression via summarization; if a context overflow occurs, context restoration is handled within the loop.
Once built, agents can have their LLMs swapped out later and can be deployed to any environment equipped with Linux containers. This enables the construction of highly customizable agents that are not dependent on a specific LLM or a specific cloud environment.
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、任意のコンテナ環境にデプロイ （Publickey、2026-09-28）