AgentDeck turns Claude Code into a team you direct. Write a ticket, pick a model — an agent branches, builds, and opens the PR while you watch the whole fleet on one live map.
Tool calls stream in as they happen. Type mid-run to redirect the agent. The context rail keeps the run's cost, branch, worktree state, and PR checks one glance away.
Team lead, projects, running threads — one graph, updating live. Every node carries its pid, token count, dirty state, and PR checks. Pan, zoom, and click any node to jump straight into its thread.
The relay pairs your daemon with your phone. Approvals arrive as push notifications you answer with a tap, threads keep streaming, and the board stays live — from the couch, the train, the queue for coffee. iOS first.
A real ledger, not estimates — every token and turn attributed by hour, model, and project. Live sessions show each agent's context window filling in real time.
A small background process owns your Claude Code sessions, tickets, and the usage ledger — SQLite, on your Mac or a VPS. Node 20+, one command.
The desktop app connects to your daemon — board, threads, map, usage. It even starts the daemon for you if it isn't running.
Write a ticket, pick a model, hit go. AgentDeck drives Claude Code with your account — Anthropic bills you directly, zero markup.
Daemon, desktop app, mobile relay — MIT-licensed, every line auditable. Self-host forever for free, or pay us to run it and never think about updates or backups again.
Read the source →Same product on every plan — paid tiers buy operations, not features.
every plan is bring-your-own-Claude — Anthropic bills your usage directly, AgentDeck never marks up a token
No forms, no funnel — mail lands in a real inbox.
Questions, feedback, early-access requests — anything goes.
For Hosted and Concierge customers — your daemon, our problem.
Found a bug or want a feature? File it in the open, like everything else.