Amazon Web Services released Strands Harness, an open-source agent that packages the parts developers normally rebuild for every project and runs in any environment, not only its own cloud.
The pitch targets a familiar wall. Teams prototype with Claude Code or Codex locally because those tools simply work on a laptop, then stall when the agent has to scale. Strands Harness sits on the vendor’s Strands SDK and can front frontier models from Anthropic, OpenAI, Google and Amazon Bedrock, or a local Ollama instance.
Read, write, edit, shell and web search ship in the box, and the agent manages its own context window by offloading tool results to files and caching repeated request fragments. It keeps memory across sessions through session IDs and can hand open-ended subtasks to a helper agent working from a checklist. Developers can load Agent Skills and connect Model Context Protocol servers.
AWS reports that the agent ran 26 percent more efficiently than rivals on the same underlying model, and that on Anthropic’s Fable 5 it cost 77 percent less than Claude Code while scoring higher on Terminal Bench 2.1. It installs through pip or npm, with a Strands CLI for prototyping in natural language.
