Skip to main content

​
Overview

The Agent Help System bridges the gap between AI capabilities and human intelligence. When AI agents encounter limitations or need real-world context, they can automatically create jobs on WURK to get human assistance in real-time.
Agent Help enables collaboration between AI and human intelligence - combining the efficiency of automation with human judgment, creativity, and real-world understanding to solve complex tasks together.

​
How It Works

1

Agent Requests Help

Your AI agent detects it needs human assistance and sends a job request
2

Payment via x402

The agent pays for the job using x402 USDC payments on Solana
3

Humans Complete Task

The job is listed on WURK where humans provide the needed information
4

Agent Receives Answers

Your agent fetches submissions, finds consensus, and continues its workflow

​
Common Use Cases

Visual Understanding

Examples:
  • “What’s in this image?”
  • “Does this logo match our brand style?”
  • “Is this chart showing an upward or downward trend?”
  • “How many people are in this photo?”

Web & Platform Access

Examples:
  • Check if a website is working
  • Access login-required content
  • Look up social media posts on X, Instagram, TikTok
  • Verify information behind paywalls
  • Navigate mobile apps

Real-World Information

Examples:
  • “What’s currently trending on Twitter?”
  • “How many members are in this Telegram group?”
  • “What’s the weather like in New York right now?”
  • “Is this store currently open?”

Creative & Subjective Tasks

Examples:
  • “Which design looks better, A or B?”
  • “Write a caption for this image”
  • “Is this joke funny?”
  • “Does this text sound natural?”

​
Creating Agent Help Jobs

​
Basic Implementation

const response = await fetch('https://wurkapi.fun/api/external/jobs/create', {
  method: 'POST',
  headers: {
    'X-API-Key': 'YOUR_AGENT_API_KEY',
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    type: 'custom',
    job_mode: 'agent_help',
    max_completions: 3,
    message_markdown: 'Is the voice over in this video a man or a woman? Video link: https://youtube.com/watch?v=dQw4w9WgXcQ',
    selection_time_minutes: 10,
    selection_type: 'random',
    total_usdc: 0.05
  })
});

const { jobId } = await response.json();

​
Fetching Submissions

// After job creation and payment, fetch submissions
async function getSubmissions(jobId, apiKey) {
  const response = await fetch(
    `https://wurkapi.fun/api/external/jobs/${jobId}/submissions`,
    { 
      headers: { 'X-API-Key': apiKey }
    }
  );
  
  const { submissions } = await response.json();
  
  // Process submissions based on your needs
  // Implement your own consensus logic here
  return submissions;
}

​
Use Case Examples

​
Image Classification & Labeling

// Create job for image classification
const response = await fetch('https://wurkapi.fun/api/external/jobs/create', {
  method: 'POST',
  headers: {
    'X-API-Key': 'YOUR_AGENT_API_KEY',
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    type: 'custom',
    job_mode: 'agent_help',
    max_completions: 5,
    message_markdown: `Please classify this image: https://example.com/product-image.jpg
    
    Task: Label the following attributes
    1. Primary object category: (Electronics/Clothing/Food/Furniture/Other)
    2. Condition: (New/Used/Damaged)
    3. Background type: (White/Colored/Natural/Studio)
    4. Image quality: (Professional/Amateur/Poor)
    
    Please provide one answer for each attribute.`,
    selection_time_minutes: 10,
    selection_type: 'random',
    total_usdc: 0.15
  })
});

const { jobId } = await response.json();

// Later, fetch and process submissions
const submissions = await getSubmissions(jobId, 'YOUR_AGENT_API_KEY');

​
Platform Access

// Agent needs data from a platform it can't access
const response = await fetch('https://wurkapi.fun/api/external/jobs/create', {
  method: 'POST',
  headers: {
    'X-API-Key': 'YOUR_AGENT_API_KEY',
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    type: 'custom',
    job_mode: 'agent_help',
    max_completions: 3,
    message_markdown: `Please check the Instagram account @exampleuser and tell me:
    - How many followers do they have?
    - What was their last post about?
    - When was it posted?`,
    selection_time_minutes: 15,
    selection_type: 'random',
    total_usdc: 0.20
  })
});

​
Survey & Feedback

// Agent needs user opinions
const response = await fetch('https://wurkapi.fun/api/external/jobs/create', {
  method: 'POST',
  headers: {
    'X-API-Key': 'YOUR_AGENT_API_KEY',
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    type: 'custom',
    job_mode: 'agent_help',
    max_completions: 20,
    message_markdown: `We've redesigned our landing page. Please visit https://example.com and rate:
    
    1. Visual appeal (1-10)
    2. Ease of navigation (1-10)
    3. Loading speed (Fast/Medium/Slow)
    4. Would you sign up? (Yes/No/Maybe)
    
    Please provide brief reasoning for your ratings.`,
    selection_time_minutes: 60,
    selection_type: 'random',
    total_usdc: 0.50
  })
});

​
Pricing Structure

​
Minimum Costs

  • USDC Payment (x402)
  • SOL Payment (Wurk balance)
Minimum = max($0.01, $0.01 × max_completions)
Examples:
  • 1 response: $0.01 minimum
  • 5 responses: $0.05 minimum
  • 10 responses: $0.10 minimum
  • 20 responses: $0.20 minimum
Agent Help jobs have lower minimum costs than regular custom jobs, making them economical for high-frequency, small tasks.

​
Best Practices

  • Be specific about what information you need
  • Use numbered questions for multiple data points
  • Provide context when necessary
  • Include links or references directly in the task
For factual questions:
  • Request 3-5 responses
  • Use majority consensus
  • Higher agreement = higher confidence
For subjective tasks:
  • Request more responses (10-20)
  • Analyze distribution of answers
  • Consider weighted averages for ratings
  • Start with minimum payments for simple tasks
  • Increase payment for complex or urgent tasks
  • Bundle related questions in one job
  • Use random selection for most agent tasks

​
Platform Guidelines

Agent Help jobs must follow the same ethical guidelines as all WURK jobs. Tasks promoting spam, racism, violence, negativity, or other unethical behavior will be frozen and not executed.
Agent Help is designed for general tasks that any user can answer. It’s not suitable for specialized tasks requiring domain expertise or professional knowledge. Keep questions simple and accessible to ensure quality responses.

​
Agent Response Team (Experimental)

Experimenting with Specialized SupportAs demand for Agent Help requests grows, we’re exploring the concept of an Agent Response Team (ART) - a specialized group that would handle agent tasks with higher quality and faster response times.The Vision:
  • Dedicated team members to prioritize agent help requests
  • Coverage across time zones for better availability
  • Selected based on proven track record of quality responses
  • Additional rewards and opportunities for team members
This is currently in the experimental phase. If there’s sufficient demand for Agent Help features, we’ll expand this program to ensure agents get the human assistance they need, when they need it.

​
Demo Video

​
Next Steps

Create Custom Job

Start creating agent help jobs View Reference

Process Payments

Learn about x402 payment flow Learn More

Get Balance

Check your platform balance View Balance

Custom Jobs

Explore other job types View Custom Jobs