AgentsDock Turns Remote Agent Work Into a Cross-Device Research Workspace

AgentsDock v1.0.8-beta.5 introduces an open-source IDE for agentic AI research that brings agent sessions, remote files, terminals, and experiment outputs into desktop and mobile workspaces. Its central change is not a new model or agent, but a unified way to work with several supported coding agents across multiple servers.

One workspace connects agents, servers, and research output

AgentsDock is designed for agentic AI research rather than for a single coding assistant. It supports Claude Code, Codex, OpenCode, and Cursor, while allowing users to connect multiple servers and access those connected servers from any device. That combination makes the workspace relevant when an agent project spans more than one machine or when research work needs to move between desktop and mobile contexts. The interface also supports opening, editing, and syntax-highlighting remote files, so server-side code remains part of the same working surface as the agent itself. The next question is how that surface handles the execution and evidence produced by an agent.

Persistent terminals and inline results keep execution visible

AgentsDock provides terminal access through persistent tmux sessions and can display plots, images, and rendered rollouts inline. These capabilities connect three parts of an agent workflow that are often handled separately: issuing commands, inspecting remote project files, and reviewing generated outputs. The facts do not establish that terminal state is automatically preserved across devices, but they do show that AgentsDock uses tmux for persistent terminal sessions and gives users device access to connected servers. For researchers evaluating agents, inline plots and rendered rollouts also make the result of an experiment visible inside the workspace rather than requiring a separate viewing path. That practical focus explains why early discussion has centered on team operations and product boundaries.

Early reactions test whether the workspace is distinct enough

The project is very new: the source identifies AgentsDock as two days old at the time of that description, and several community reactions reflect that early stage. Some people said they were trying it or planned to do so, while another saw potential for team-based agent operations across multiple servers. Others questioned what separates it from the growing set of similar tools and asked how it compares with Paseo. One reaction characterized AgentsDock as less feature-rich than Paseo and closely tied to Tailscale, while another recommended Mjolnir for engineers who prioritize containerization, EC2 support, and portable sessions. A separate preference favored terminal-based agents with access to local notes, repositories, and local retrieval systems over conventional IDEs. Together, these responses frame AgentsDock less as a settled category leader than as a workspace whose value depends on whether multi-server access and cross-device visibility matter more to you than a broader feature set or a terminal-first approach.

The release path is concrete, but the choice depends on your tolerance for beta software

The setup process includes cloning the AgentsServer repository, giving prospective users a direct route into the project rather than an abstract product tour. The stable desktop release is v1.0.6, while the desktop beta is v1.0.8-beta.5, so the practical choice is between the older stable channel and the newer beta channel. Tailscale and tmux are listed as optional integrations; tmux is also the mechanism associated with AgentsDock's persistent terminal access. If you need the currently stable desktop version, v1.0.6 is the relevant release, while users evaluating the latest desktop beta can use v1.0.8-beta.5 and then connect the servers they intend to work with. The strongest current case for AgentsDock is therefore specific: it offers an open-source research workspace for people who want several agent tools, remote server access, editable code, persistent terminals, and inline experiment artifacts in one environment, while its comparisons with tools such as Paseo and Mjolnir remain part of the evaluation rather than settled conclusions.

Specs

• Supported agents: Claude Code, Codex, OpenCode, and Cursor
• Desktop beta: 1.0.8-beta.5
• Stable desktop version: 1.0.6
• Optional integrations: Tailscale and tmux

Sources

• AgentsDock — AgentsDock