AWS has released Strands Agents, an open source SDK that simplifies building and running AI agents using a model-driven approach. This framework allows developers to create agents in just a few lines of code, scaling from simple to complex use cases. Multiple AWS teams already use Strands for their AI agents in production, including Amazon Q Developer, AWS Glue, and VPC Reachability Analyzer.
Key Features of Strands Agents
Strands Agents simplifies agent development by leveraging state-of-the-art models to plan, chain thoughts, call tools, and reflect. The SDK connects two core pieces of an agent: the model and the tools. It plans the agent’s next steps and executes tools using advanced reasoning capabilities of models.
Core Concepts
- Model: Strands offers flexible model support, allowing developers to use models from various providers like Amazon Bedrock, Anthropic, and Meta.
- Tools: Developers can choose from thousands of published Model Context Protocol (MCP) servers or use Python functions as tools.
- Prompt: A natural language prompt defines the task for the agent.
The Agentic Loop
The agentic loop is central to Strands’ capabilities. It invokes the LLM with the prompt, agent context, and tool descriptions. The LLM can respond in natural language, plan steps, reflect on previous steps, or select tools. Strands executes selected tools and provides results back to the LLM.
Example Use Case: Naming AI Assistant
To demonstrate Strands’ capabilities, an example of building a naming AI assistant is provided. This agent uses a default model in Amazon Bedrock, an MCP server, and a pre-built Strands tool to suggest names for an open source project.
Deployment Options
Strands is designed for production use and supports various architectures:
- Monolithic Deployment: Both the agentic loop and tool execution run in the same environment.
- Microservices Architecture: Separating the agentic loop and tool execution into different environments.
- Return-of-Control Pattern: Clients run tools while the agent is hosted elsewhere.
Strands provides instrumentation for collecting agent trajectories and metrics using OpenTelemetry.
Community and Contributions
Strands Agents is an open source project under the Apache License 2.0. The community welcomes contributions, including support for additional model providers and tools.
To get started with Strands Agents, visit the project’s documentation on GitHub.
