AWS Strands Harness: Open-Source Agent Harness for Multi-Cloud AI Agent Prototyping

AWS released Strands Harness, an open-source framework for building AI agents that can run locally or across multiple cloud environments. The project is built on the Strands Harness SDK and is designed to solve the pain point of moving agent prototypes from local machines to scalable cloud deployments. For builders, this could mean faster iteration cycles and less vendor lock-in, though the headline performance claims come from AWS itself and need careful scrutiny.

What Strands Harness Offers

The harness comes with pre-built tools including read/write/edit, shell, and web search capabilities. Instead of requiring developers to design custom tools, it relies on the underlying model's existing tool-use abilities. The framework includes advanced context management that offloads tool results to separate files and caches reusable request parts, which should reduce token consumption and latency on long agent runs. It also maintains long-term memory across sessions using session IDs and includes a built-in helper agent for delegating open-ended subtasks via an automated checklist.

Multi-Environment Deployment

Strands Harness can run on AWS, Google Cloud, Microsoft Azure, Modal, Cloudflare, and other infrastructure. For local deployment, developers can point it to an Ollama-hosted model. This multi-environment support is the core differentiator: most agent frameworks tie you to a specific provider's SDK or runtime. Strands Harness lets you prototype locally with Ollama then deploy to any supported cloud without rewriting the agent.

Supported Models and Prototyping Workflow

The framework supports Anthropic Claude Code, OpenAI Codex, Amazon Bedrock, and Google's models. The Strands CLI enables anyone to prototype agents using natural-language commands. After selecting a model and adding prompts and tools, the /export command downloads the harness code as a Python or TypeScript file. This lowers the barrier for non-developers to experiment with agents while still producing deployable code.

Performance Claims and Caveats

AWS reported that Strands Harness was 26% more efficient than agents built on other frameworks using the same underlying model. On Anthropic's Fable 5 model, it claimed 77% lower costs than Claude Code on Terminal Bench 2.1. These are vendor-reported benchmarks, not independently verified. A Register report noted that the comparisons were limited to coding agents and that AWS "marked its own homework." Builders should test against their own workloads before assuming cost savings.

Getting Started

Install via pip install strands-agents-harness or npm install strands-agents-harness. The source code and documentation are available on GitHub. For now, Strands Harness is best used for rapid prototyping across environments. Production hardening and enterprise support details are not yet clear, and the project is in its early days.