Your agents never sleep.
Colony is an open-source multi-agent swarm orchestration tool for Claude Code. It's a simpler, more focused alternative to OpenClaw — with Ruflo-style intelligent routing, a TUI dashboard, always-on daemon mode, and WhatsApp/Telegram triggers.
We generated 104 SEO pages (500K+ words) in 30-minute batches using 10 parallel agents. Colony packages that approach so anyone can do it.
| Tool | Problem |
|---|---|
| OpenClaw | Broken with Claude Code subscriptions. Docker-only. Security nightmare. |
| Ruflo | Enterprise-grade, 60+ agents, complex. Overkill for most users. |
| Claude Code (native) | Has Agent tool, but no queue management, no dashboard, no always-on. |
| Colony | Simple CLI. Queue + swarm + dashboard + daemon + messaging. Works in 2 minutes. |
# Install
npm install -g @aumiqx/colony
# Initialize in your project
colony init
# Add tasks from a JSON file
colony queue add tasks.json
# Run with 10 parallel agents + live dashboard
colony queue run --parallel 10That's it. 10 agents will process your tasks in parallel with a live TUI dashboard.
Give it a queue of tasks, it spawns N parallel Claude Code agents and manages the whole pipeline.
# Process a JSON queue
colony queue add keywords.json --template seo-page --parallel 10
colony queue run
# Quick swarm for a single task
colony swarm "Refactor the auth module" --agents 5Real-time terminal dashboard showing agent status, queue progress, costs, and logs.
┌─ Colony Dashboard ──────────────────────────────────────┐
│ Queue: seo-batch Progress: 7/10 (70%) ETA: ~8 min │
│ ████████████████████░░░░░░░ │
│ │
│ #1 ● running gemma-4-review... 3m 12s │
│ #2 ● running cursor-3-review... 2m 48s │
│ #3 ✓ done gpt-5-pricing... 4m 22s │
│ #4 ○ pending mcp-guide... — │
│ │
│ Tokens: 1.2M Cost: ~$4.80 Speed: 1.7 pages/min │
│ [q]uit [p]ause [r]esume [s]top all │
└──────────────────────────────────────────────────────────┘
Run Colony on a Mac Mini 24/7. Caffeinate keeps it awake, cron runs your queues on schedule.
# Start daemon (caffeinate + scheduler + messaging)
colony daemon start
# Install as macOS LaunchAgent (survives reboot)
colony daemon install
# Stop
colony daemon stopSend a message from your phone, get 10 pages generated. Like OpenClaw, but it actually works.
You: "Generate 10 pages about AI coding tools"
Colony: "Starting swarm: 10 tasks, model=opus..."
Colony: "[====== ] 60% — 6/10 done, ETA ~4 min"
Colony: "✓ Queue complete: 10 done, 0 failed. Cost: ~$3.20"
Optional 3-tier intelligent model routing. Simple tasks use cheap models, complex tasks use premium — saving up to 85% on API costs.
{
"ruflo": {
"enabled": true,
"endpoint": "http://localhost:8080"
}
}Colony is designed to run 24/7. Here's how to set it up properly:
Always run Claude Code with caffeinate to prevent Mac sleep:
# Basic — prevent idle sleep
caffeinate -i colony daemon start
# Mac Mini (headless) — prevent system + idle sleep
caffeinate -i -s colony daemon start
# MacBook — prevent idle sleep only (preserve battery)
caffeinate -i colony daemon startInside Claude Code, use the built-in /loop command:
# Run queue every 4 hours
/loop 4h /colony queue run --parallel 10
# Run with cron expression
/loop --cron "0 */4 * * *" /colony queue run
# Install macOS LaunchAgent
colony daemon install
# Load it
launchctl load ~/Library/LaunchAgents/com.aumiqx.colony.plist
# Verify
launchctl list | grep colony# Stop all agents
colony stop
# Stop daemon
colony daemon stop
# Nuclear option — kill caffeinate
killall caffeinate
# Disable cron in Claude Code
export CLAUDE_CODE_DISABLE_CRON=1Create colony.config.json in your project root:
{
"swarm": {
"maxAgents": 10,
"model": "opus",
"retryFailed": true,
"maxRetries": 2
},
"templates": "./templates",
"postProcess": [
"npm run build",
"npm run lint"
],
"daemon": {
"caffeinate": true,
"cron": "0 */4 * * *",
"healthCheckInterval": 60
},
"messaging": {
"whatsapp": {
"enabled": true,
"provider": "baileys",
"allowedNumbers": ["+91XXXXXXXXXX"]
},
"telegram": {
"enabled": true,
"botToken": "YOUR_BOT_TOKEN",
"allowedUsers": [123456789]
}
},
"ruflo": {
"enabled": false,
"endpoint": "http://localhost:8080"
}
}[
{ "prompt": "Write a review of GPT-5.4", "outputPath": "content/gpt-5-4.md" },
{ "keyword": "best ai tools 2026", "vars": { "year": 2026 } },
{ "prompt": "Refactor auth module", "template": "refactor" }
]| Command | Description |
|---|---|
colony init |
Initialize Colony in current project |
colony queue add <file> |
Add tasks from JSON |
colony queue run [id] |
Process queue with parallel agents |
colony queue status |
Show queue progress |
colony queue list |
List all queues |
colony swarm "task" |
Quick parallel swarm |
colony dashboard |
Launch TUI dashboard |
colony daemon start |
Start always-on daemon |
colony daemon stop |
Stop daemon |
colony daemon install |
Install macOS LaunchAgent |
colony template list |
List available templates |
colony status |
Quick status |
colony history |
Past queue runs |
Colony also works as a Claude Code plugin with slash commands:
/colony init "Generate SEO pages for AI tools"
/colony swarm --agents 10 "Refactor auth module"
/colony queue run
/colony dashboard
/colony status
Colony builds on Claude Code's existing swarm infrastructure:
- AgentTool — Spawns parallel agents with model selection and background mode
- Coordinator mode — Orchestrates workers, synthesizes results
- Task system — Tracks agent lifecycle (LocalAgentTask, RemoteAgentTask)
- Ink TUI — React-based terminal UI framework
Colony adds: queue management, TUI dashboard, daemon mode, messaging triggers, Ruflo routing.
We built Colony after discovering that 10 parallel Claude Code agents generate 63 pages in 30 minutes — vs 0 pages in 6.5 hours with a sequential script. We used this approach to generate 104 SEO pages (500K+ words) for aumiqx.com. Colony packages that approach so anyone can do it.
PRs welcome. See ARCHITECTURE.md for the system design.
MIT - Aumiqx Technologies