Working with several coding agents at once otherwise means juggling a dozen terminal windows and losing track of which agent is waiting on what. Agent Deck pulls that together: every session in its own pane, grouped by project, all visible at the same time.
Each session runs in a real PTY through node-pty and xterm.js: not a reimplemented terminal, but the CLI exactly as it would run in a shell. Sessions get separate account profiles, so multiple logins can run in parallel without signing each other out. Besides Claude Code, Codex, Grok, Antigravity and OpenCode run as agents too.
On top of that come the things a plain terminal lacks: when an agent asks a question, it shows up inline in the terminal pane as a clickable dialog instead of scrolling past, step-by-step instructions land in their own view, projects can be connected through the GitHub device flow, and a graph tab visualizes the knowledge graph of linked memory-vault notes for the active project. Updates arrive automatically over separate alpha and stable channels.
Since version 3.4 there is a master mode: one session gets the master role and can list other sessions, look at their screens, give them tasks, start new ones and close those no longer needed. A human grants the role and takes it away again, everything in between is logged, and the supervision view has an emergency stop. On top of that come a quota display per profile with the rolling 5-hour window and the 7-day window, an ideas tab that files thoughts into the matching folder of the memory vault, and a guided tour on first launch.
The source is private; the installer and update metadata live in a public release repository. Currently at version 3.5.0, backed by 1,788 tests in 93 files.