Strands harness provides a ready-to-use framework for building general-purpose AI agents with fewer tokens and support for multiple model providers.
AWS has released Strands harness, an open-source agent framework designed to provide the components needed to build and run general-purpose AI agents without creating the orchestration layer from scratch. Licensed under Apache 2.0, the framework can be deployed locally or in cloud environments and can be used with a range of AI models. AWS says it can be set up with a single line of Python or TypeScript.
The harness supports models from Amazon Bedrock, Anthropic, OpenAI and Google, along with local models through Ollama and LiteLLM. Developers can install it using Python or npm and deploy it in environments capable of running Linux containers, including services such as Amazon ECS, Google Cloud Run and Cloudflare Containers.
A major focus is reducing the amount of context consumed during agent operation. Strands harness truncates tool results exceeding about 1,500 tokens, begins context compaction when the context window reaches 85 per cent capacity and performs context recovery when an overflow occurs. Prompt caching is also enabled by default. AWS reports that across six benchmarks, the harness used 28 per cent fewer tokens than comparable harnesses while maintaining similar accuracy when running the same Claude or GPT models.
The framework is also intended to make agent development easier to move from experimentation to deployment. Through the Strands CLI, developers can describe an agent in natural language, prototype it and then export the resulting implementation to Python or TypeScript using the /export command. The underlying Strands ecosystem provides support for tools, model portability, memory, sessions, MCP, multi-agent patterns, streaming, guardrails and evaluation.
Strands harness is available as an open-source project, with its SDK published under the Apache 2.0 licence. By packaging context management, tool handling and other agent infrastructure into a reusable framework, AWS is targeting developers who want to build agents without recreating the underlying execution loop and supporting components for every application.
