Documentation

Learn how to use Squan to orchestrate AI agents and build software faster.

Getting Started

From zero to your first autonomous agent in minutes

Core Concepts

The building blocks of multi-agent development

Agents

AI agents that work autonomously in git worktrees. Each agent receives task instructions, writes code, runs tools, and signals when done.

Release Trains

Task groups that flow through the kanban board. Create a train, add instructions, dispatch an agent, and track progress.

Everything-as-Code

All project state lives in .squan/ — tasks as markdown files, charters, templates, docs. Committed to git, reviewable in PRs.

Kanban Board

Visual task management: Open → In Progress → PR Review → Landed → Cancelled. Auto-updates as agents work.

Git Worktrees

Each agent gets an isolated copy of the repo via git worktree. No conflicts between parallel agents. Each works on its own branch.

Root Agent (Mayor)

An optional orchestrator agent that coordinates other agents, reviews their work, and makes routing decisions.

Architecture

Squan.exe (Electron)
  └── Embedded Server (Node.js/Express)
       ├── REST API (:3001)
       ├── WebSocket (real-time events)
       ├── SQLite database (cache)
       ├── node-pty / tmux (agent terminals)
       └── Serves React client

Per-agent:
  └── Git worktree (isolated copy of repo)
       ├── task instructions
       └── AI Agent (autonomous)

Keyboard Shortcuts

ShortcutAction
⌘K / Ctrl+KCommand palette
⌘B / Ctrl+BToggle sidebar
⌘, / Ctrl+,Settings
⌘1-7 / Ctrl+1-7Switch views (Terminals, Kanban, Metrics, Events, Costs, Console)

Console Commands (sq>)

CommandDescription
statusFull overview — projects, agents, trains
agentsList all agents with status
spawn <project> "task"Spawn a one-off agent
create-train <project> "name"Create a release train
dispatch <train-id>Assign agent and start work
kill <name>Kill an agent
send <name> "message"Message a running agent
land <train-id>Mark a train as complete