Strands发布开源AI Agent框架Strands Harness

Strands团队推出了Strands Harness，一款开源的通用AI Agent框架，支持本地运行或云端部署1。该框架在成本效益上具有显著优势，在六个基准测试中相比Claude Code和GPT模型的成本降低28%，同时保持相当或更优的准确度；使用Fable 5模型时成本更低77%1。

框架内置了shell、文件读写编辑、网页工具等功能模块1，并支持Amazon Bedrock、Anthropic、OpenAI、Google、Ollama和LiteLLM等多个模型提供商1。上下文管理方面，系统会对约1500个令牌的工具结果进行截断，当上下文窗口达到85%以上时触发摘要压缩机制1。此外，框架还内置提示词缓存和多步骤工作追踪等功能1。

Strands Harness采用Apache 2.0开源许可，用户可通过Python（pip install strands-harness）或TypeScript（npm install @strands-agents/harness）进行安装1。

The Strands team has released Strands Harness, an open-source universal AI Agent framework designed to run locally or be deployed to the cloud 1. The framework demonstrates significant cost advantages over existing solutions, achieving 28% lower costs compared to Claude Code and GPT models across six benchmark tests, with costs dropping 77% below Claude Code when using Fable 5 while maintaining comparable or superior accuracy 1.

Strands Harness incorporates built-in functionality including prompt caching, context management, and multi-step workflow tracking 1. The framework provides integrated tools for shell operations, file read-write-edit functions, and web utilities, and supports multiple model providers including Amazon Bedrock, Anthropic, OpenAI, Google, Ollama, and LiteLLM 1. Context management operates through a mechanism that truncates tool results at approximately 1,500 tokens and triggers summary compression when the context window exceeds 85% capacity 1.

The framework is available for both Python and TypeScript developers, with installation via pip install strands-harness for Python and npm install @strands-agents/harness for TypeScript 1. Strands Harness is distributed under the Apache 2.0 open-source license 1.