Context-Efficient Sparse File Editing
Implement a windowed scroll-and-patch editing interface to let software agents view specific file segments and apply precise unified diffs, reducing token overhead.
Source paper
Xingyao Wang, Boxuan Li, Yufan Song, Frank F. Xu, Xiangru Tang, Mingchen Zhuge, Jiayi Pan, Yueqi Song, Bowen Li, Jaskirat Singh, Hoang H. Tran, Fuqiang Li, Ren Ma, Mingzhang Zheng, Bill Qian, Yanjun Shao, Niklas Muennighoff, Yizhe Zhang, Binyuan Hui, Junyang Lin, Robert Brennan, Hao Peng, Heng Ji, Graham Neubig
Generated by AI. The paper’s authors did not write these.
Implement a windowed scroll-and-patch editing interface to let software agents view specific file segments and apply precise unified diffs, reducing token overhead.
Integrate a vision-language model into the browser interaction loop to automatically generate visual descriptions from screenshots for catching rendering bugs.
Use execution results from Docker test suites as a reinforcement learning reward signal to train a critic model that blocks failing code actions.
OpenHands (formerly OpenDevin) is an open-source platform designed to facilitate the development and evaluation of AI agents capable of performing complex software engineering tasks. The platform provides a standardized event-stream architecture, a secure Docker-based runtime for code execution and web browsing, and a suite of benchmarks (such as SWE-Bench and WebArena) to assess agent performance against real-world scenarios.
The system utilizes an event-stream architecture where an agent interacts with a Docker-sandboxed environment containing a bash shell, an IPython server, and a web browser. The agent perceives the environment through a history of actions and observations (State) and executes tasks using standardized 'skills' (Python functions) or direct command-line execution, with support for multi-agent delegation.