A fully assembled, open-source, general-purpose AI agent harness that runs locally or deploys to any cloud provider. It supports models from Amazon Bedrock, Anthropic, OpenAI, Google, and local Ollama, with built-in tools for web search, shell, file editing, and memory. It achieves 28% lower token costs than comparable harnesses while matching or exceeding accuracy on benchmarks.

Strands Decider 2B is a 2-billion-parameter open-source decision model released by AWS Strands Labs. Built on a Qwen3.5-2B torso with the language-model head replaced by a 1-million-parameter pointer head, it scores a supplied list of options and returns a calibrated choice in tens of milliseconds without generating text, making it suitable for local agent routing, tool selection and action gating.

Open-source sandbox for AI agents that combines OS-level isolation with Dogwood policies able to condition permissions on what the agent has already done, such as blocking outbound HTTP after reading customer data. Credentials stay outside the agent process. Currently in developer preview on macOS.

Amazon CloudWatch Omni is a new AI-powered observability experience that unifies monitoring of applications and AI agents in a standalone, off-console environment. It auto-discovers topology from OpenTelemetry telemetry, supports natural-language and SQL queries, includes evaluators for agent quality and correctness, and offers IDE extensions for VS Code and other tools plus a collaborative web interface with SSO.

AWS Well-Architected Agent is an AI-powered cloud optimization service that continuously analyzes AWS environments and delivers prioritized, goal-aligned recommendations across cost, security, performance, and resilience. It correlates metrics, topology, and IaC (Terraform, CDK, CloudFormation) and returns ready-to-deploy fixes such as SSM runbooks and updated templates. Public preview for Business+ Support customers.