AgentsDock

AgentsDock is an open-source IDE built specifically for agentic AI research, letting users work with Claude Code, Codex, and Cursor side by side from a single desktop or mobile client. The project is licensed under Apache 2.0, with the client code published at github.com/ZhengyiLuo/AgentsDock and a companion self-hosted backend called AgentsServer. Desktop stable v1.0.0 shipped in September 2026, with mobile clients available for iPhone, iPad, and Android.

What It Is

AgentsDock is a cross-platform client-server IDE that separates the control surface (the app) from the compute (your own server). You install AgentsServer on any macOS or Linux machine you control — a lab workstation, a home Mac mini, or a rented GPU box — and then connect to it from the desktop app on macOS, Linux, or Windows, or from the mobile app on iPhone, iPad, or Android. The agent CLIs (Claude Code, Codex, Cursor) run on the server using your own credentials; AgentsDock provides the interface to start tasks, follow live activity, review results, and manage sessions from any device.

Self-Hosted Architecture

The client-server split is central to how AgentsDock works. AgentsServer listens on port 7850 and requires the relevant agent CLIs to be installed and authenticated on the server machine. The desktop and mobile apps connect over a local network or a private network like Tailscale. Because the server is self-hosted, chat history, code, files, and credentials stay on the machine you operate. Model requests still go to the provider (Anthropic, OpenAI, etc.) — self-hosting controls the workspace, not the models.

Key Capabilities

• Multi-agent side-by-side chats: run Claude Code, Codex, and Cursor sessions in parallel and switch between them in one interface.
• Persistent tmux terminal: each chat owns a persistent tmux session on the server; detaching the app leaves shells and long-running processes alive.
• Inline media review: agents return plots, images, and rendered video rollouts directly in the chat, so you can review training results without leaving the app.
• Scheduled and recurring jobs: queue interval jobs so long-running work runs unattended.
• Agent-to-agent messaging: address another agent with @ to hand off a task across chats or servers; AgentsDock tracks the exchange and surfaces the reply.
• Chat forking: branch any chat to try a new direction while keeping the original intact — useful for ablations or testing risky changes.
• Session import: resume Claude Code or Codex sessions started outside AgentsDock by pasting a session ID, or bulk-import all sessions from a machine.
• Slash command palette: a / menu surfaces file attachment, model switching, MCP server management, digest creation, and more.
• Multi-server management: connect multiple servers and switch between them from any device.

Platform Coverage

The desktop app is built with Electron, React, and TypeScript and ships as a universal macOS DMG (macOS 14+), Linux AppImage for x86_64 and ARM64, and a Windows x64 installer (Windows 10+). The mobile app uses React Native and Expo and is available for iPhone and iPad (iOS/iPadOS 17+) via the App Store and TestFlight, and for Android (arm64) as a direct APK download. The website lists the project as targeting AI researchers, with the homepage citing user testimonials from researchers at Carnegie Mellon University, UC Berkeley, and NVIDIA — though these are vendor-published claims on the product page.

Update: Desktop v1.0.0 Stable

The GitHub repository shows the v1.0.0 stable release was published on September 14, 2026, with the repository created on September 11, 2026. The release covers macOS universal, Linux x86_64 and ARM64, and Windows x64 builds. The Android client is separately tracked in the AgentsDock-Releases repository and was at beta.8 at the time of the latest push. The project is actively maintained, with the last push recorded on September 15, 2026.