Important
Community project — not affiliated with AWS or the strands-agents team.
Think Docker Compose, but for AI agents
Strands is a powerful agent SDK. Building multi-agent systems with it means writing the same wiring code over and over — models, MCP servers, hooks, orchestration topology. Strands-compose kills that boilerplate: describe your entire agent system in YAML, call load(), and get back plain, fully-wired strands objects. No wrappers. No magic. The real deal.
models:
default:
provider: bedrock
model_id: us.anthropic.claude-sonnet-4-6-v1:0
agents:
researcher:
model: default
system_prompt: "You research topics."
writer:
model: default
system_prompt: "You write reports."
coordinator:
model: default
system_prompt: "Coordinate the team."
orchestrations:
team:
mode: delegate
entry_name: coordinator
connections:
- agent: researcher
description: "Research a topic."
- agent: writer
description: "Write the report."
entry: teamsdk-python · PyPI · Examples
The heart of the organization. Everything else builds on top of it.
Your config.yaml goes straight to production — no refactoring, no new abstractions. Dseploy your entire agent system to AWS Bedrock AgentCore with fully managed infrastructure, auto-scaling, and zero ops.
A ready-to-use chat frontend that connects to a strands-compose server. Point it at your server URL, and get a streaming chat panel with a live Agent Flow graph — watch every reasoning step, delegation, and agent completion render in real time as your system runs.
This project is built in the open — everyone is welcome.
Whether you're fixing a bug, adding a new provider, improving docs, or just kicking the tyres — contributions of any size are appreciated. Come join the effort.
- ⭐ Star sdk-python to show support and help others discover the project
- 🐛 Open an issue if something doesn't work or you have a feature idea
- 🔀 Submit a PR — from a typo fix to a full new feature, all contributions are welcome
- 👀 Watch this org to follow new repos and releases as they land


