Catenary

Run Claude Code, Codex and Antigravity side by side, in one IDE built for them. The same work session has two lenses, and switching between them never restarts a process:

• TerminalMap — a focused grid of terminals, splits and tabs, with a cockpit of session cards that shows what every agent is doing and which one needs you.
• Canvas — an infinite surface where terminals, editors, browsers and worktrees float, and wires connect one agent's work to the next.

Around them, a real editor: Monaco with a shared local LSP for TypeScript and Python that your agents can query too, aware of what you have not saved yet.

brew install --cask gionicolas/catenary/catenary

irm https://thecatenary.app/install.ps1 | iex

curl -sSL https://thecatenary.app/download/linux-appimage -o Catenary.AppImage
chmod +x Catenary.AppImage

Every file, with its SHA-256 checksum, is on the Releases page. Catenary updates itself after that.

• Two lenses, one session — toggle TerminalMap and Canvas with Ctrl/Cmd+Alt+S. Terminals, editors, browsers and dev servers stay alive across the switch.
• Session cockpit — one card per work session with your last prompt and the tool the agent is running. The bell rings only when an agent is blocked: an approval, a failure or an explicit request.
• Splits and Split Zoom — split the focused terminal right or down (⌘D / ⇧⌘D on macOS, Alt+Shift+D / Alt+Shift+H on Windows and Linux) and maximize one pane without touching the layout.
• Agent-aware terminals — Claude Code, Codex, Antigravity, Cursor, Grok, OpenCode and Pi report when they are working, waiting or done, so you see it without opening the terminal.
• Agents that talk to each other — agents in the same session reach each other with catenary ask, no copy-paste. On the Canvas, a wire makes the link visible and can relay a finished turn on its own.
• Maestro Mode — one agent recruits, briefs and connects a team: 2 helpers on Free, 10 on Pro by default (adjustable from 2 to 50, or unlimited). Recruiting or dismissing an agent stops at a card you approve.
• Native catenary mcp server — installed agents get catenary_* tools automatically: open files, create tasks and worktrees, split terminals, request your attention, delegate work and query the shared LSP.
• Worktrees for parallel work — a task gets its own git worktree and branch in one click, with a colour that follows it everywhere.
• Local and remote take the same path — point Catenary at a host over SSH or WSL and terminals, git, search and agents run there, while editors, browsers and the canvas stay local.
• Local-first — no account and no telemetry: Catenary itself sends nothing. Your agents talk only to their own providers.

• Desktop shell: Electron 41 + Node.js (strict context isolation)
• Frontend: React 18, TypeScript, Tailwind CSS
• Editor: Monaco (VS Code's editor) + typescript-language-server and pyright
• Terminal: node-pty (native PTY) + xterm.js with the WebGL renderer
• State: Zustand + atomic JSON persistence (.cate/workspace.json)

Found a bug or have an idea to make Catenary better? Please open an issue or email support@thecatenary.app. You can also send feedback from the app: Help → Send Feedback…