Skip to main content

​
Understanding HITL.sh

HITL.sh is a platform that adds human oversight to AI workflows. It acts as a bridge between your AI systems and human reviewers, ensuring quality, safety, and compliance in automated processes.

​
Key Concepts

​
Human-in-the-Loop (HITL)

Human-in-the-loop refers to a workflow where AI systems perform tasks but require human oversight for critical decisions. This approach combines the efficiency of automation with the judgment and expertise of humans. Benefits:
  • Quality control: Human reviewers catch errors and improve outputs
  • Risk management: Critical decisions require human approval
  • Compliance: Meet regulatory requirements for human oversight
  • Learning: Collect feedback to improve AI performance

​
Review Workflow

A review workflow in HITL.sh follows this pattern:
  1. AI generates content - Your AI system creates output (text, images, decisions)
  2. Review request - Send the output to HITL.sh for human review
  3. Human review - Reviewers examine and approve/reject/modify content
  4. Result processing - Your workflow receives the review result and continues

Workflow Examples

See real-world examples of review workflows in action.

​
Architecture Components

​
Review Templates

Templates define the structure of what gets reviewed and what input you collect from humans. Template elements:
  • Fields: Input areas for reviewers (text, checkboxes, file uploads)
  • Layout: How the review interface is organized
  • Validation: Rules for required fields and data formats
  • Styling: Custom appearance and branding

Workflow Designer

Learn how to create and customize review templates.

​
Agent Inbox

The Agent Inbox is the central hub where reviewers access pending reviews. Features:
  • Unified view: All reviews in one place
  • Filtering: Sort by status, assignee, priority
  • Bulk actions: Process multiple reviews efficiently
  • Mobile optimized: Review on any device

​
Integration Methods

HITL.sh integrates with your existing systems through multiple channels:

SDKs

Python and TypeScript SDKs for direct integration

Workflow Tools

Native nodes for n8n, Make.com, and other platforms

HTTP API

RESTful API for custom integrations

Webhooks

Receive review results in real-time

​
Review Types

​
Content Review

Review AI-generated content before publication or distribution. Use cases:
  • Blog posts and articles
  • Marketing copy
  • Social media content
  • Product descriptions

​
Decision Approval

Approve AI-generated decisions or recommendations. Use cases:
  • Financial transactions
  • Risk assessments
  • Compliance decisions
  • Strategic recommendations

​
Quality Assurance

Ensure AI outputs meet quality standards. Use cases:
  • Code generation
  • Data analysis
  • Image generation
  • Translation services

​
Training Data

Collect human feedback to improve AI models. Use cases:
  • Model fine-tuning
  • Dataset creation
  • Performance evaluation
  • Bias detection

​
Workflow Patterns

​
Simple Approval

The most basic pattern: AI generates content, human approves/rejects.

​
Iterative Improvement

AI generates content, human provides feedback, AI regenerates.

​
Multi-Step Review

Complex workflows with multiple review checkpoints.

​
Conditional Review

Only review content that meets certain criteria.

​
Security & Privacy

​
Data Protection

  • Encryption: All data encrypted in transit and at rest
  • Access control: Role-based permissions for team members
  • Audit logs: Complete history of all review activities
  • Compliance: GDPR and CCPA compliant

​
Integration Security

  • API keys: Secure authentication for integrations
  • Webhook verification: Signed payloads for security
  • Rate limiting: Protection against abuse
  • IP restrictions: Optional IP whitelisting

​
Performance & Scalability

​
Review Processing

  • Real-time updates: Instant notifications and status changes
  • Batch processing: Handle multiple reviews efficiently
  • Priority queuing: Important reviews get processed first
  • Load balancing: Automatic scaling based on demand

​
Integration Performance

  • Low latency: API responses in milliseconds
  • High throughput: Handle thousands of reviews per minute
  • Reliability: 99.9% uptime guarantee
  • Global distribution: Low-latency access worldwide

​
Best Practices

​
Template Design

  • Keep it simple: Focus on essential review criteria
  • Clear instructions: Help reviewers understand what to look for
  • Efficient layout: Minimize clicks and scrolling
  • Mobile first: Ensure good experience on mobile devices

​
Workflow Integration

  • Async processing: Don’t block your main workflow
  • Error handling: Plan for review failures or timeouts
  • Fallback options: Provide alternatives when reviews aren’t available
  • Monitoring: Track review completion times and success rates

​
Team Management

  • Clear roles: Define who reviews what
  • Training: Help reviewers understand the process
  • Feedback loops: Continuously improve the review process
  • Automation: Automate routine review assignments

​
Next Steps

Quickstart Guide

Get hands-on experience with your first review workflow.

Integration Guide

Learn how to integrate HITL.sh with your existing systems.

Examples

See real-world implementations and use cases.