Harness Sdk

Strands Agents is an open source toolkit for building production AI agents in Python and TypeScript. The registry lists it under the name Harness SDK, which corresponds to the agent harness and SDK that Strands publishes at strandsagents.com. At its core is a standard Agent class, with a harness layer that returns a plain Strands Agent, so everything built for an ordinary agent carries over. The documentation covers the full path from prototype to deployment. Developers get guides for running agents on containers and serverless targets, persisting session and memory state to durable storage, and observing agents through traces, metrics and logs. Safety features include guardrails, PII redaction and guidance on trusted message history, which warns that history built from untrusted input can carry forged tool calls. A separate Evals SDK generates test experiments, plans topics so cases cover the range of tasks an agent handles, and scores agent runs with evaluators such as a trajectory evaluator, from code or from a command line tool. The project also announced Strands Box, an open source sandbox for AI agents with a policy layer called Dogwood, currently in preview, alongside research labs. Strands is open source and free to use. The only price found on the site is zero. Costs depend on the model providers and infrastructure you choose to run it on.