A Claude-app-style desktop window for the Claude Code running on your own Linux server, built for people who don't code. Setup needs only a server IP and its password.
Real appGrouped sessions, tool cards and per-turn downloadsReal appClaude asks, you click an optionReal appSubagent steps and background tasksReal appFirst-run wizard
Electron
ssh2
Claude Code
Windows
macOS
MIT
Try it in your browser
The chat below is a scripted replay in your browser with made-up requests and files. It does not connect to any server or to Claude. The real app's interface is in Chinese; get it on GitHub.
~/workspaceFiles
Pick a sample request below
Opus 5.5ToolsContext —
My serveryou@example.com · Max5 hours18%This week41%
Demo only: the real app saves the file to your computer.
How it works
Each message runs the official Claude Code CLI on the server in headless mode (claude -p with JSON in and out, plus the stdio permission channel so Claude can ask you multiple-choice questions), first with --session-id, then --resume. The JSON event stream becomes streaming Markdown, tool cards, subagent steps and background tasks.
Dropped files are uploaded over SFTP and their paths are passed to Claude. When a turn ends, the app runs find -newermt from the turn's start time and lists every new or changed file with a download button, including files written by scripts Claude ran.
The first-run wizard logs in once with your password or key, installs the app's own SSH key, creates a dedicated non-root account when you log in as root, installs Claude Code and gets a long-lived token through claude setup-token. The password is never stored.
Claude runs without per-command prompts inside that dedicated account, which is what keeps the rest of the server safe.
Servers that only accept the office network or a jump host can be reached through a proxy (socks5:// or http:// CONNECT); SSH, file transfer and the wizard all use it. New in 1.2.0.
File paths in replies become links: click to download, or the eye icon to preview. PDF, Word, Excel, PowerPoint and code open through preview plugins installed from a GitHub repository; each plugin runs sandboxed with no network access, and updates download only the files that changed. There is also a server file manager with folder download. New in 1.3.0.
Branch a new conversation from any reply, or rewind to before one of your messages and resend (the original is kept). A context meter, a time-machine style timeline, an MCP & Skill page with per-conversation switches, and updates from GitHub Releases. New in 1.4.0.
Import the Claude Code sessions already saved on your server, including ones from a terminal, and keep chatting in them. New in 1.4.2.
Type / in the input box for a command menu (switch conversation, change model, compact context, run a Skill), edit and resend any of your messages, and preview files in a side panel next to the conversation. New in 1.4.3.
Switch between several servers from the sidebar, each with its own conversations, and re-authorize Claude any time with /login. New in 1.4.4.
Each conversation works in its own project folder with a file panel beside it; type while Claude replies (messages queue and can be withdrawn); plan usage and a context breakdown are always at hand. New in 1.5.0.
Automatic conversation titles, servers shown only by the name you give them (addresses stay masked for screenshots), the Claude account each server uses, and macOS hosts including your own Mac. New in 1.5.1.
Windows PCs can be hosts too (OpenSSH server + Git for Windows); not yet tested on a real Windows machine. New in 1.5.2.
When Claude asks you to choose, the options are buttons you click; subagent steps show inside their card; background commands get a panel with live output and a stop button; HTML files preview as web pages, previews show download progress and small files stay cached; table columns can be dragged wider. New in 1.5.4.
The setup wizard reuses the Claude Code already installed (including Homebrew on a Mac) and lets you pick a folder you have used Claude in before; on a Mac host it explains why a separate login is needed; one Send/Stop button and Esc to stop; the timeline shows how long each message is. New in 1.5.5–1.5.7.
Use the computer you are on directly: no SSH, no Remote Login, and on a Mac the Claude login you already have in Terminal just works. New in 1.5.9.
Limitations
Installers are not code-signed yet, so Windows and macOS warn on first launch. Password login has not been tested against a real server (key login has). Windows hosts have not been tested on a real Windows machine. Not affiliated with Anthropic; Claude and Claude Code are trademarks of Anthropic, PBC.
