HumanOS

The Orchestration Engine for Human-AI Workflows

What is HumanOS?

HumanOS is the orchestration engine that coordinates humans and AI agents in complex workflows. It's the "brain" that routes tasks, enforces approvals, and tracks provenance.

Think of it as Kubernetes for Human-AI teams — you describe what needs to be done, HumanOS figures out who should do it, when humans need to approve, and logs everything.

Core Responsibilities

Route tasks based on capabilities

Enforce human-in-the-loop approvals

Create execution plans

Track full provenance

How It Works

┌──────────────────────────────────────────────────────────────┐
│                     HUMANOS WORKFLOW                           │
├──────────────────────────────────────────────────────────────┤
│                                                                │
│  1. TASK SUBMITTED                                             │
│  ┌────────────────────────────────────┐                       │
│  │ "Process 50 invoices and generate  │                       │
│  │  financial summary"                 │                       │
│  │                                     │                       │
│  │ Required: invoice_processing,      │                       │
│  │           financial_analysis        │                       │
│  └────────────────────────────────────┘                       │
│                    ↓                                           │
│  2. CAPABILITY MATCHING                                        │
│  ┌────────────────────────────────────┐                       │
│  │ HumanOS queries Capability Graph:  │                       │
│  │ "Which agents have BOTH            │                       │
│  │  capabilities?"                     │                       │
│  └────────────────────────────────────┘                       │
│                    ↓                                           │
│  3. EXECUTION PLAN CREATED                                     │
│  ┌────────────────────────────────────┐                       │
│  │ Step 1: Agent A processes invoices │                       │
│  │ Step 2: Agent B analyzes data      │                       │
│  │ Step 3: Human approves summary     │                       │
│  └────────────────────────────────────┘                       │
│                    ↓                                           │
│  4. HUMAN-IN-THE-LOOP (if required)                            │
│  ┌────────────────────────────────────┐                       │
│  │ Manager: "Approve" or "Reject"     │                       │
│  └────────────────────────────────────┘                       │
│                    ↓                                           │
│  5. EXECUTION & PROVENANCE                                     │
│  ┌────────────────────────────────────┐                       │
│  │ Agents execute, every step logged  │                       │
│  └────────────────────────────────────┘                       │
│                                                                │
└──────────────────────────────────────────────────────────────┘

Start a Workflow

SDK:

Key Features

Capability-Based Routing

HumanOS automatically finds agents with the required capabilities. No hardcoded routing logic.

typescript

// HumanOS finds agents with BOTH capabilities
requiredCapabilities: [
  'invoice_processing',
  'financial_analysis'
]

// Automatically routes to the best-fit agent

Human-in-the-Loop

For high-stakes decisions, HumanOS pauses execution and waits for human approval.

typescript

// Require human approval
humanInLoop: true

// Workflow status: "pending_approval"
// Human must call approveWorkflow() to continue

Execution Plans

HumanOS breaks complex tasks into steps and assigns them to agents.

typescript

// Execution plan is generated automatically
executionPlan: [
  {
    order: 1,
    description: "Extract data from 50 invoices",
    assignedTo: "passport_agent_invoice",
    requiredCapabilities: ["invoice_processing"],
    status: "pending"
  },
  {
    order: 2,
    description: "Analyze financial trends",
    assignedTo: "passport_agent_analyst",
    requiredCapabilities: ["financial_analysis"],
    status: "pending"
  },
  {
    order: 3,
    description: "Generate summary report",
    assignedTo: "passport_agent_analyst",
    requiredCapabilities: ["report_generation"],
    status: "pending"
  }
]

Monitor a Workflow

SDK:

Approve or Reject

Approve

SDK:

Reject

SDK:

Common Use Cases

Multi-Agent Workflows

Coordinate multiple agents (e.g., data extraction → validation → analysis → reporting)

Human Oversight

High-value transactions, legal documents, or sensitive operations require human approval

Dynamic Routing

Route work to agents based on availability, capabilities, and performance

Workflow Templates

Define reusable workflows (e.g., "Invoice Processing SOP") and execute them on demand

Escalation Paths

If an agent fails or a human rejects, escalate to a supervisor or different agent

Learn More

Patterns →

Orchestration examples