Amazon has released Strands Agents, an open source SDK that simplifies AI agent development through a model-driven approach. The framework enables developers to build agents by defining prompts and tool lists with minimal code.
The project launches with backing from major technology and consulting firms, with Amazon noting that "several companies are joining us with support and contributions including Accenture, Anthropic, Langfuse, mem0.ai, Meta, PwC, Ragas.io, and Tavily." Strands scales from simple to complex agent use cases, and from local development to deployment in production, offering developers a streamlined path from prototype to production-ready AI agents.
The framework builds on three core components that define an agent: "a model, tools, and a prompt." These elements work together in what Amazon calls the "agentic loop," where "the agent uses these three components to complete a task, often autonomously." During operation, "Strands invokes the LLM with the prompt and agent context, along with a description of your agent's tools." The system capitalizes on current LLM capabilities, noting how "powerful LLMs have become and how well they can natively reason, plan, and select tools."
Source: Strands Agents
The execution flow follows a structured pattern where "the LLM can choose to respond in natural language for the agent's end user, plan out a series of steps, reflect on the agent's previous steps, and/or select one or more tools to use." Strands manages the technical complexity by handling tool execution: "When the LLM selects a tool, Strands takes care of executing the tool and providing the result back to the LLM." The process continues iteratively until "the LLM completes its task, Strands returns the agent's final result."
Strands Agents positions itself as "lightweight and production-ready, supporting many model providers and deployment targets." The SDK offers flexibility across deployment scenarios, supporting "conversational, non-conversational, streaming, and non-streaming" agent types for different workloads.
Key capabilities include "full observability, tracing, and deployment options for running agents at scale," along with built-in tools that allow developers to start quickly. The framework supports advanced implementations including "multi-agent and autonomous agents," enabling techniques such as "agent teams and agents that improve themselves over time."
Amazon emphasizes that Strands maintains "safety and security as a priority" to help organizations run agents responsibly while protecting data. The SDK's architecture provides a simple, customizable agent loop while remaining model, provider, and deployment agnostic to support various models from different providers.
The SDK includes two additional packages for development: strands-agents-tools and strands-agents-builder, both available on GitHub. The tools package provides example implementations that extend agent capabilities, while the builder package includes an agent specifically designed to assist developers in creating their own Strands agents and tools. These components support custom implementations and help developers extend the framework's functionality.
Strands Agents extends beyond Amazon Bedrock to support multiple model providers. Developers can access Anthropic's Claude models through direct API integration, while LiteLLM provides a unified interface for OpenAI, Mistral, and other providers. The framework supports Meta's Llama models through the Llama API and enables local model execution via Ollama for privacy or offline requirements. OpenAI models are accessible through direct API connections, including OpenAI-compatible alternatives. The SDK also allows developers to build custom providers for specialized implementation needs.
Developers interested in building AI agents with Strands can access the open source SDK on GitHub, where they can find documentation, examples, and contribute to the growing community around the project.