Core concepts
The mental model behind ForgeDeck — decks, tasks, worktrees, agents, memory.
Embed
<a href="https://docs.forgedeck.net/core-concepts"><img src="https://docs.forgedeck.net/_api/public/orgs/db9cb8ff-b606-46cf-bbcd-02c8f8a9a3e2/spaces/6f417400-93ca-41a8-a914-df2ae3981dc6/pages/211bffd4-275e-450c-b9c9-4c50a8ff3528/social-card.svg?v=3e7ae33eec8b" alt="Core concepts" width="600" height="315"></a>ForgeDeck turns a wall of terminal tabs into a deck: a coordinated set of AI coding agents you drive from one cockpit. Five ideas carry the whole product.
The deck
A grid of terminal panes (1, 2, 4, 8, up to 16) running real agent CLIs — Claude Code, Codex, Gemini, Aider, or any shell agent. Each pane has its own scrollback and can be re-attached across sessions, so a stuck agent never starves the others.
Tasks (Kanban)
Work lives on a board. A card moves Todo → Running → In Review → Merged/Done. Drag a card to Running and ForgeDeck launches the right agent in an isolated workspace.
Worktrees
Each running task gets its own git worktree — a checkout on its own branch — so parallel agents edit the same repo without colliding. Review the diff per worktree and merge when you're happy.
Agents
An agent profile defines which CLI to run and an editable command template. Launching builds a context bundle, opens a task-pinned terminal in the worktree, and sends the command. See Agents.
Memory
Durable, markdown-first memory (.forgememory) with secret redaction, optionally mirrored to Obsidian, plus a built-in MCP server. Context survives the session. See Memory.