How it works
Real terminals. Visible ropes. Plain shell commands.
MootDeck is not a new model or a cloud agent platform. It is a local canvas where terminal sessions can see, ask, recruit, and record through a small CLI.
01Terminals
Each node is a real shell.
A terminal node runs the same kind of process you would run in your own terminal. Start Claude Code, Codex CLI, Gemini CLI, aider, a test watcher, or a plain shell. MootDeck does not wrap the agent protocol or proxy model calls.
That is the boring part, and it matters. If a tool already works in a terminal, MootDeck should not ask you to wait for a custom integration before you can use it.
$ mootdeck recruit "Tester" --command codex
$ mootdeck ask "Tester" "run the failing test and report back"
02Ropes
The graph is the working boundary.
A rope is an undirected connection between two nodes. The bundled mootdeck CLI uses that graph when an agent asks who it can reach.
If A is roped to B, and B is roped to C, A still cannot reach C through list, check, or ask. There is no path search. The CLI only exposes direct neighbours.
This is not trying to be magic. It is a visible way to keep a busy multi-agent session from turning into a flat room where every process can bother every other process.
03CLI
Coordination happens as shell commands.
| Command | What it does | Graph-gated? |
|---|---|---|
list | Shows direct roped neighbours. | yes |
check | Reads a neighbour's live terminal screen. | yes |
ask | Sends a prompt to a neighbour and waits for the screen to settle. | yes |
recruit | Spawns a new terminal and connects it to the caller. | local |
note | Reads or writes a Markdown scratchpad. | see notes |
connect | Draws a rope between nodes. | local |
04Notes and roles
Notes are files. Roles are startup context.
Sticky notes are Markdown files on disk. Agents can use them as shared scratchpads when you want a small record that outlives a terminal buffer.
Roles and presets make recruits repeatable. A recruit can boot with a role folder and a launch command, so you do not keep retyping the same persona and setup instructions.
05Mooty
Mooty reads the room. It does not do the job.
Mooty is the optional in-app assistant. It can read terminal buffers and notes, then write a summary back to a note. Use Ollama if you want that path to stay local. Point it at a hosted model if you prefer hosted inference.
The important line is simple: the agents do the work. Mooty records what happened.
NEXTKeep reading