The OpenHands Software Agent SDK: Composable and Extensible (arxiv.org)

OpenHands released a redesigned Software Agent SDK aimed at making production-grade software engineering agents easier to build, extend, and deploy. The SDK provides a minimal interface that can implement a working agent in a few lines of code while remaining highly extensible—developers can add custom tools, memory managers, and richer agent behaviors. It packages secure, reliable execution features such as native sandboxed execution, lifecycle control, and model-agnostic multi-LLM routing, plus integrated REST/WebSocket services for seamless local-to-remote portability. Human interaction is supported out of the box through connectors for VS Code, VNC/browser workspaces, CLIs, and APIs. For the AI/ML community this matters because it lowers the engineering barrier to prototyping and scaling real-world agents: sandboxing and built-in security analysis reduce operational risk, lifecycle controls simplify orchestration, and multi-LLM routing lets teams experiment with and combine different models without locking into a single provider. The SDK is a component-level redesign of the already popular OpenHands framework (64k+ stars) and shows strong empirical results on SWE-Bench Verified and GAIA benchmarks, suggesting it’s not just flexible but performant enough for production use.