AWS releases Strands Decider 2B and Strands Box for AI agent development

New Open-Source Tools for AI Agents

AWS has launched two open-source projects aimed at facilitating AI agent development: Strands Decider 2B and Strands Box. These tools address specific challenges in creating agents that can utilize tools, execute code, and interact with external systems effectively and securely.

Strands Decider 2B: Focused Decision-Making

Strands Decider 2B is a two-billion-parameter model optimized for making decisions from predefined options rather than generating extensive text. It supports tasks such as model routing, tool selection, classification, and context management. This model allows developers to provide a set of choices and receive a decision with associated confidence scores.

The model is built on Qwen3.5-2B, with its language model head replaced by a pointer head for scoring candidate answers. It uses a rank-16 LoRA adapter for fine-tuning. AWS has made the model weights, training data, and scripts publicly available, enabling local execution and modification. Performance metrics include a median decision latency of approximately 115 milliseconds on an NVIDIA RTX 3090 and 153 milliseconds on an Apple M3 MacBook for small tasks. It achieved 72.3% accuracy on JevBench's 231 public tasks for its v19 checkpoint.

Strands Box: Secure Agent Execution

Strands Box, currently in developer preview, offers an isolated environment for executing AI agents. It combines operating-system-level isolation with Dogwood, an open-source policy language, to define and enforce agent actions. The sandbox supports shell commands, Python execution, MCP tools, filesystem operations, and network access.

This isolation mechanism routes agent interactions through a policy layer, allowing developers to permit or deny actions based on content and execution history. This approach enables controls like requiring agents to pass tests before code commits or restricting API call frequency, operating independently of prompt instructions to enforce execution boundaries. Strands Box is designed to be compatible with various models and agent frameworks, both locally and in cloud environments.