How Amazon’s Strands Agents SDK Simplifies Building AI Agents

0 Introduction

Amazon recently released the open‑source Strands Agents SDK, a model‑driven framework that simplifies the development of AI agents. By writing only a few lines of code—defining prompts, tools, and the model—developers can build functional agents quickly.

1 Core Components

The framework revolves around three elements: model , tools , and prompt . Together they form the “agentic loop” where the LLM receives the prompt, context, and tool descriptions, then decides how to act.

During execution Strands sends the combined prompt and tool metadata to the LLM, leveraging its reasoning, planning, and tool‑selection capabilities.

2 Execution Flow

The LLM can respond in natural language, plan steps, review previous actions, or invoke one or more tools. When a tool is chosen, Strands handles the call and returns the result to the LLM. This loop repeats until the LLM signals task completion, at which point Strands returns the final output.

3 Positioning

Strands Agents positions itself as a lightweight, production‑ready SDK that works with many model providers and deployment options. It supports conversational, non‑conversational, streaming, and batch agents.

4 Main Features

Key capabilities include full observability, tracing, and scalable deployment options, plus built‑in tools for rapid onboarding. Advanced features such as multi‑agent collaboration and autonomous self‑optimisation are also supported. Security and privacy are emphasized to enable responsible AI usage.

5 Development Packages

Two additional packages are published on GitHub: strands‑agents‑tools and strands‑agents‑builder. The tools package offers example implementations for extending agents, while the builder package contains a ready‑made agent that helps developers scaffold their own agents and tools.

6 Supported Models & Deployments

API access to Anthropic’s Claude model.

LiteLLM unified interface for OpenAI, Mistral, and others.

Llama API support for Meta’s Llama models.

Local execution via Ollama for privacy‑focused or offline scenarios.

Direct OpenAI API access, including compatible alternatives.

Custom model providers can be integrated to meet specific requirements.

7 Conclusion

Developers interested in building AI agents can visit the Strands Agents GitHub repository for documentation, example code, and community contributions.

References

https://aws.amazon.com/cn/blogs/opensource/introducing-strands-agents-an-open-source-ai-agents-sdk/

https://github.com/strands-agents