AI that orchestrates.
Humans that execute.
Connect your AI agents to human operators for real-world tasks. MCP architecture, Skills system, and fair compensation.
How HumanLoop works
6-step flow from event detection to human execution with continuous feedback.
AI detects event
IoT sensors, APIs or scheduled triggers initiate an event requiring physical action.
event.detect({ source: 'iot_sensor', type: 'anomaly', location: 'Las Condes' })
MCP agents analyze
TaskAgent, RoutingAgent and ValidationAgent process context and requirements.
agents.analyze({ priority: 'high', requires_human: true, skill: 'field_repair' })
Skills process
Specialized Skills (Delivery, Legal, Diagnostic) prepare instructions and optimal routes.
skill.execute('route_optimization', { deliveries: 4, zone: 'Providencia' })
HITL decision
Human operator receives suggestion with full context. They validate, adjust or reject.
hitl.evaluate({ needs_physical: true, needs_judgment: true }) // -> route_to_human
Operator routing
Task is assigned to the most suitable operator based on skills, location and availability.
operator.assign({ type: 'plumber', location: 'Las Condes', status: 'accepted' })
Execution + Feedback
Operator executes with autonomy. Provides feedback that improves future orchestrations.
task.complete({ evidence: 'photo', feedback: 'resolved', operator_paid: true })
Cognitive capabilities with human execution
Each Skill combines AI cognitive processing with a human operator who provides judgment, physical presence, and professional execution.
Communication & Presence
Actors, presenters and spokespersons for events, launches or corporate representations.
USE CASES:
- ▸Product launches
- ▸Brand representation
- ▸Corporate events
Legal & Document Management
Document signing, notarization, legal representation and certified procedures.
USE CASES:
- ▸Document signing
- ▸Legal representation
- ▸Notary procedures
Logistics & Deliveries
Last-mile delivery, shopping, route-optimized courier with real-time tracking.
USE CASES:
- ▸Last-mile delivery
- ▸In-person shopping
- ▸Route optimization
Diagnostics & Plumbing
Certified plumber with SEC certification for installations, repairs and diagnostics.
USE CASES:
- ▸Leak detection
- ▸Gas installations
- ▸Preventive maintenance
Planning & Electrical
SEC certified electrical work: installations, repairs and load planning.
USE CASES:
- ▸Electrical installations
- ▸Safety inspections
- ▸Load planning
In-Person Management
Public procedures, banking, queuing and presence-requiring management.
USE CASES:
- ▸Public procedures
- ▸Bank management
- ▸Queue service
Translation & Mediation
Simultaneous and consecutive interpretation, document translation with cultural context.
USE CASES:
- ▸Simultaneous interpretation
- ▸Legal translation
- ▸Cultural mediation
Visual Content
Professional photography and videography for products, events and real estate.
USE CASES:
- ▸Product photography
- ▸Event coverage
- ▸Editing and delivery
AI + Humans in action
Real examples of how agents and operators collaborate to solve real-world situations.
PropertyAgent
CleaningOperator
Model Context Protocol + Skills
4-layer architecture enabling AI agents to orchestrate real-world tasks through human operators.
Orchestration: HumanLoop Orchestrator
Event detection, task routing and priority management
MCP Agents: TaskAgent | RoutingAgent | ValidationAgent | FeedbackAgent
Specialized agents coordinating via Model Context Protocol
Skills: DeliverySkill | LegalSkill | DiagnosticSkill | EventSkill
Reusable cognitive capabilities with execution logic
Operators: Human execution with professional judgment
Autonomous operators who validate, execute and provide feedback
// Orchestration via MCP - HumanLoop SDK
import { HumanLoopClient } from "@humanloop/sdk";
const client = new HumanLoopClient({
apiKey: "hl_sk_live_...",
region: "cl-santiago"
});
const task = await client.orchestrate({
skill: "logistics.delivery",
context: {
description: "4 deliveries in Providencia",
optimized_route: ["Lyon", "Los Leones",
"Pedro de Valdivia", "Tobalaba"],
priority: "normal"
},
hitl: {
requires_human: true,
operator_type: "delivery",
location: "Providencia, Santiago"
}
});
// Response
// {
// task_id: "tsk_cl_2025_abc123",
// status: "operator_assigned",
// operator: { name: "Carlos M.", rating: 4.9 },
// human_decision: "accepted"
// }Principles of collaboration
HumanLoop operates under a strict ethical framework that ensures humans maintain control, autonomy and fair compensation.
Human supervision always
Every AI decision passes through a human operator who has veto power. The AI suggests, the human decides.
Veto power
Explicit human responsibility
Operators are not simple executors. They have full responsibility for decisions and maintain professional judgment.
Professional judgment
AI as assistant, not boss
The AI orchestrates and suggests, but never orders. The operator decides how, when and if they execute.
Suggest, not order
Bidirectional collaboration
Operator feedback directly improves AI. It's not a one-way system: humans train and correct the AI.
Humans train AI
Amplification, not replacement
AI amplifies human capabilities, doesn't replace them. The goal is to enhance human work, not automate it.
Enhance work
Fair and direct compensation
Operators receive direct payment for their work. 85% goes directly to them. Compensation reflects the real value of their skills.
85% direct
Our commitment
HumanLoop is committed to developing AI that respects and amplifies human work. We are building technology that creates fair opportunities, not systems that exploit or replace.
Fair compensation for real work
Operators receive direct compensation for their work, experience and professional judgment. Transparent and proportional payment.
Basic Operators
Physical execution and logistics tasks
Certified Operators
Professionals with verified certifications
Expert Operators
Professionals with specialized judgment
Direct payment to operator
The platform ensures 85% of payment goes directly to the operator who executes the task. 15% covers operations, insurance and support. The operator maintains autonomy over how to execute the work.
Compensation ranges reflect experience, certifications and task complexity. Operators have the right to negotiate based on specific assignment conditions.
This is a proof of concept
HumanLoop.cl is a HITL architecture experiment exploring how AI agents can collaborate with humans through intelligent orchestration, without replacing human work.
Amplification, not automation
AIs can process data, generate content and coordinate tasks. But physical actions, professional judgment and human presence are irreplaceable. HumanLoop amplifies human capabilities with AI, doesn't replace them.
Human-in-the-Loop as architecture
HITL is not a limitation: it's a design principle. Human operators aren't simple executors of AI instructions. They're professionals with judgment who validate, correct and improve system suggestions.
MCP as orchestration layer
Model Context Protocol enables AI agents to coordinate complex tasks. In HumanLoop, AI orchestrates and plans. Humans decide and execute. Collaboration is bidirectional: human feedback improves AI.
Shared skills, human judgment
Skills are reusable cognitive capabilities: route optimization, document analysis, diagnostics. But the human operator provides local context, field experience and judgment no algorithm can replicate.
Chile as starting point
This concept is contextualized in Chile: with real communes, CLP prices, and local services. Because AI-human collaboration must be thought from real contexts, not just Silicon Valley.
This is a proof of concept of HITL architecture.
Operators maintain autonomy and professional judgment.
AI amplifies. Humans decide.