I've been exploring AWS Strands Agents recently, it's their open-source SDK for building AI agents with proper tool use, reasoning loops, and support for LLMs from OpenAI, Anthropic, Bedrock, LiteLLM Ollama, etc.
At first glance, I thought it’d be AWS-only and super vendor-locked. But turns out it’s fairly modular and works with local models too.
The core idea is simple: you define an agent by combining
- an LLM,
- a prompt or task,
- and a list of tools it can use.
The agent follows a loop: read the goal → plan → pick tools → execute → update → repeat. Think of it like a built-in agentic framework that handles planning and tool use internally.
To try it out, I built a small working agent from scratch:
- Used DeepSeek v3 as the model
- Added a simple tool that fetches weather data
- Set up the flow where the agent takes a task like “Should I go for a run today?” → checks the weather → gives a response
The SDK handled tool routing and output formatting way better than I expected. No LangChain or CrewAI needed.
If anyone wants to try it out or see how it works in action, I documented the whole thing in a short video here:
Also shared the code on GitHub for anyone who wants to fork or tweak it: Repo link
Would love to know what you're building with it!
Top comments (2)
AWS using the name Strands 😂, nice.
Tangenting out from the conversation, there's a Game Death Stranding where the main character is Sam Strand.
Maybe the AWS team got inspired by this. 🤔
Thanks for the tutorial! I will throughly watch the video when I will have free time!