Desktop deck for agentic development
Run your fleet of coding agents from one deck
Launch, observe, and control Claude Code, OpenCode, and Codex side by side — each agent in its own pane, on its own branch
The whole fleet in one window
Panes tile themselves as agents join — the same square-driven grid the app uses. No tabs to dig through, no windows to hunt: every agent stays in view
- Up to 16 agents in one deck, tiled automatically
- Workspaces keep projects side by side — grids stay mounted on switch
- One glance tells you who is running and who is waiting on you
Every agent on its own branch
Spawn an agent into its own git worktree, created on the spot. Agents work in parallel without stepping on each other — and the branch chip in the pane header shows where each one lives
- A worktree and a branch per agent, made at spawn
- Mix isolated agents with agents in the main repo, per pane
- Branch chips keep the mapping visible at all times
kd/<workspace>/<n> — its own worktree · no chip — the main repo
Quit. Relaunch. Nothing is lost
The deck comes back exactly as you left it — workspaces, panes, transcripts. A restart stops being a reason to keep windows alive
- Workspaces and panes restored on launch
- Transcripts return whole — no replay, no retyping
- The layout lands exactly where it was
The agents you already run
KeepDeck detects the coding agents installed on your machine — anything missing stays out of the way
Principle
Hold stability under speed
Speed without stability isn’t velocity — it’s risk. The deck exists to hold both
Keep Building