AgentsDock launches an IDE for supervising AI agents across remote research machines

A new development tool called AgentsDock reached the front page of Hacker News on September 13, 2026, positioning itself as an IDE built specifically for agentic AI research. Rather than assuming a human is doing the typing, it is organised around handing work to coding agents — the project names Claude Code and Codex among them — and letting them carry out research tasks and model training runs on a researcher's behalf.

According to the project's website, AgentsDock follows a server-and-client model. A server component is installed on the machine where the work should happen, and a desktop client connects to it. The server can then be reached from any device, with the agents executing research and training jobs remotely.

How it works

The site's setup guide describes three steps: install the AgentsDock server by cloning a public GitHub repository and running its install script, launch the desktop app and paste in the server's URL, then start assigning work to agents. The sample commands are macOS-flavoured, and the page lists Tailscale and tmux as optional supporting tools, with the product's terminal feature built around tmux sessions.

Multiple agents, side by side

The core interface is a chat-based dock where several agents run side by side in one window. The pitch is that a researcher can keep conversations with Claude Code, Codex and other agents open simultaneously, rather than managing each one in its own terminal.

Results also arrive in those same conversations. According to the site, agents return plots, images and rendered rollouts inline, so the real output of a run can be reviewed without leaving the chat — useful for training and evaluation work where the interesting artifacts are visual rather than textual.

Remote machines at the centre

AgentsDock is built around remote compute. The site says a user can connect multiple servers — it gives the examples of a lab workstation, a Mac mini at home and a rented GPU box — and switch between them at any time, from any device they own.

Remote files can be opened, edited and syntax-highlighted inside the app, so no separate editor is needed for quick changes on a distant machine. Each chat is backed by a persistent tmux session that can be attached to from anywhere, which the project says gives you the same shell on every device.

The launch page does not detail pricing, licensing or system requirements, and the claims above come from the project's own marketing rather than independent testing. The server component is distributed through GitHub, according to the setup instructions.

Why it matters

Coding agents are becoming standard equipment for many developers, but the tooling around them still largely assumes a single user at a single terminal. An IDE designed around supervising several agents at once, on machines that may only be rented by the hour, is a different shape of product — closer to a control panel than a text editor.

The design also mirrors how machine-learning work actually runs: heavy jobs land on remote GPUs, sessions need to survive a laptop being closed, and results often arrive as images and plots rather than console output. Wrapping Claude Code, Codex and similar agents into that workflow, with persistent shells, multi-server switching and inline artifacts, addresses friction that researchers currently patch together with SSH, tmux and ad-hoc scripts.

As a fresh launch known mainly through one landing page and a Hacker News appearance, questions remain about maturity, adoption and how the project will develop. But as a signal of where developer tooling is heading — from tools for writing code toward tools for delegating it — AgentsDock is a launch worth watching.