Amazon Web Services Inc. says it’s trying to help developers solve the problem of scaling artificial intelligence agents to cloud environments with the launch of Strands Harness, an open-source agent that can help them build and deploy AI applications in any environment.
In a blog post today, AWS explained that many developers have already built prototypes of AI agents locally using tools such as Anthropic PBC’s Claude Code and OpenAI Group PBC’s Codex, because those environments “just worked” on their own machines. But when it comes to transitioning from a local setup to a scalable cloud environment, they experience problems.
This is where Strands Harness comes in. It provides what AWS said is a complete, ready-to-use agent that can run locally or in any cloud, meaning not just AWS but also Google Cloud, Microsoft Azure, Modal, Cloudflare and others.
Strands Harness is built atop the Strands Harness software development kit, which is an open-source framework developed by AWS for managing and running multi-agent patterns. It’s an extremely flexible foundation, capable of running atop of the latest frontier models from Anthropic, OpenAI, Amazon Bedrock and Google. If developers prefer to keep their agents hosted locally, they can also point Strands Harness to an Ollama model hosted on their own machine.
AWS said Strands Harness comes with numerous essential components out of the box, including read, write and edit, shell and web search capabilities. Instead of getting developers to design bespoke tools for each agentic task they want their agents to be able to perform, Strands Harness relies on tools that the underlying model already knows how to use.
The model also features advanced context management capabilities, intelligently managing its own context window by offloading tool results to separate files and caching reused parts of requests. That enables it to cut down on processing times and makes it more efficient in terms of token consumption.
Strands Harness also maintains a long-term memory across runs, which enables it to resume previous conversations through session IDs. There’s even a built-in helper agent it can delegate open-ended subtasks too using an automated checklist. Developers can also upload various “Agent Skills” and integrate external tools such as Model Context Protocol servers.
According to AWS, Strands Harness performed strongly on a range of industry benchmarks. It found that the agent was 26% more efficient than agents built on other frameworks when using the same underlying model.
It also had greater token efficiency compared to Claude Code and Codex, and performed better than those agents on several other benchmarks. For instance, in one test using Anthropic’s Fable 5 model, Strands Harness cost 77% less than Claude Code on the same tasks, while achieving a higher overall score on the Terminal Bench 2.1 benchmark.
AWS said the primary use case for Strands Harness is rapid prototyping of new AI applications. One of AWS’ teams used Strands Harness to develop the Strands CLI, which is a command-line interface that enables developers and those without coding skills to prototype AI agents using natural language commands. All the user has to do is select the underlying model, add the prompts and tools they think the agent needs, and then use an “/export” command to download the underlying harness code as a Python or TypeScript file.
Strands Harness is available to download from GitHub. It can also be installed directly via standard package managers using pip install strands-agents-harness or npm install strands-agents-harness.
