Skip to main contentsb0 separates your agent logic from infrastructure concerns. You focus on building the agent, while sb0 handles everything else.
You implement:
- Agent handler - A Python class with your custom agent logic using the Claude Agent SDK
sb0 handles:
- CLI tooling for scaffolding, testing, building, and deploying
- Docker containerization for production deployment
- Platform integration (authentication, hosting, API)
- Dependency management with uv
The Development Flow
- Initialize your agent project with
sb0 init
- Develop your agent handler in Python
- Test locally with
sb0 run --prompt "test input"
- Build Docker image with
sb0 build
- Deploy to platform with
sb0 push
- Query your hosted agent with
sb0 query "question"
This separation means you can focus on your agent’s behavior and capabilities, while sb0 ensures it runs reliably in production.
Local vs Production
- Local development: Use
sb0 run --prompt to test your agent quickly without Docker
- Production deployment: Use
sb0 build to create a Docker image, then sb0 push to deploy
You can iterate rapidly during development, and only build/deploy when ready for production.