Researcher Zhengyi Luo released AgentsDock, an open-source IDE built for agentic AI research that runs multiple coding agents side by side in one workspace, according to the project's site and GitHub repository.
The tool integrates Claude Code, Codex and Cursor, and connects to multiple remote servers at once, so a researcher can switch between a lab workstation, a home server and a cloud GPU instance from the same interface, according to the project.
AgentsDock ships for macOS, Windows and Linux, plus iOS and Android clients currently in beta, letting researchers check on running experiments and training jobs, including plots, videos and other output, from a phone without staying tied to a laptop, the project says.
The tool, still in beta, is already used by researchers at Carnegie Mellon University, UC Berkeley and Nvidia, according to its documentation.
Mobile monitoring for long-running training jobs is a small feature solving a real problem, since GPU jobs run for hours or days and researchers need to check status without a laptop open the whole time. It is also a sign that the agent IDE is becoming its own product category, separate from the single-agent coding assistant most tools chase.