Fleet Orchestration

One machine is a workstation. Multiple machines working together? That is a fleet.

When your AI workload outgrows a single computer, you scale horizontally. Fleet orchestration coordinates multiple machines -- dispatching tasks, sharing state, monitoring health, and ensuring no single failure breaks the system.

What you'll learn

Task dispatch: routing work to the right machine
Heartbeat monitoring: knowing which machines are alive
Machine registries: tracking capabilities and availability
Scaling patterns: when and how to add machines

Concept

Why a Fleet?

A single machine has limits. Your M3 Mac handles AI inference, runs your agents, serves your website, and manages your brain. But when you need to generate video while running a data pipeline while serving web requests -- one machine is not enough.

A fleet distributes work across multiple machines. Each machine has a role, a set of capabilities, and a workload. The orchestrator assigns tasks to the right machine based on what it can do and how busy it is. If one machine goes down, the others keep running.

This is not cloud computing. These are YOUR machines -- a desktop, a laptop, a mini PC, maybe a cloud VPS. All under your control, all part of your sovereign infrastructure.

Architecture

The Machine Registry

Every machine in the fleet needs a registration entry that describes what it can do:

// Machine registry stored in the brain
{
  "machines": [
    {
      "id": "m3-forge",
      "hostname": "m3-forge.local",
      "role": "primary",
      "capabilities": ["inference", "code", "planning", "brain"],
      "specs": {"ram": "64GB", "chip": "M3 Max", "gpu": true},
      "status": "online",
      "last_heartbeat": "2026-04-29T10:30:00Z",
      "current_load": 0.45
    },
    {
      "id": "m4-mirror",
      "hostname": "m4-mirror.local",
      "role": "parallel",
      "capabilities": ["inference", "social", "deploy", "testing"],
      "specs": {"ram": "48GB", "chip": "M4 Pro", "gpu": true},
      "status": "online",
      "last_heartbeat": "2026-04-29T10:29:55Z",
      "current_load": 0.20
    },
    {
      "id": "gcp-watcher",
      "hostname": "34.11.241.254",
      "role": "cron",
      "capabilities": ["monitoring", "cron", "alerts"],
      "specs": {"ram": "2GB", "chip": "x86", "gpu": false},
      "status": "online",
      "last_heartbeat": "2026-04-29T10:30:01Z",
      "current_load": 0.05
    }
  ]
}

Implementation

Task Dispatch

Task dispatch routes work to the right machine. The dispatcher considers three factors: capability (can this machine do the task?), availability (is it online and not overloaded?), and priority (which machine should handle urgent work?).

// Task dispatch logic
function dispatch(task) {
  const registry = JSON.parse(brain.read('system.machine_registry'));

  // Filter to machines that can handle this task
  const capable = registry.machines.filter(m =>
    m.status === 'online' &&
    m.capabilities.includes(task.requires) &&
    m.current_load < 0.8  // Not overloaded
  );

  if (capable.length === 0) {
    // No machine available -- queue for later
    return { action: 'queue', reason: 'No capable machine available' };
  }

  // Route to the least-loaded capable machine
  const target = capable.sort((a, b) => a.current_load - b.current_load)[0];

  return {
    action: 'dispatch',
    target: target.id,
    hostname: target.hostname,
    task: task
  };
}

// Example dispatch
dispatch({
  title: 'Generate social media posts',
  requires: 'social',
  priority: 'normal',
  payload: { topics: ['AI automation', 'sovereign stack'] }
});
// Result: dispatched to m4-mirror (has social capability, load: 0.20)

🔒

This lesson is for Pro members

Unlock all 355+ lessons across 36 courses with Academy Pro. Founding members get 90% off — forever.

Go Pro — $4.90/mo ← Back to course

Already a member? Sign in to access your lessons.