Support & Documentation

Getting Started

What is Zentropi?

Zentropi is a system that allows you to create our own AI-powered content labelers. Drawing upon our state of the art CoPE small language model, the labelers you build are flexible, accurate, and fast. No coding required -- build your labeler with a simple prompt.

Quick Start Guide
  1. Create a Labeler: Go to Create Labeler and describe what you want to classify
  2. Add Test Examples: Upload or paste text samples to test your labeler
  3. Refine & Optimize: Use our auto-optimization feature to improve accuracy
  4. Deploy: Save your labeler and use it via our API or web interface

Frequently Asked Questions

Our labelers leverage our state-of-the-art CoPE model to provide high accuracy. Most users experience 80-95% accuracy for their classification tasks. Use our benchmarking feature to test your labeler and our auto-optimization to improve results.

You can label any text content including:
  • Social media posts and comments
  • Customer reviews and feedback
  • Email content and support tickets
  • News articles and blog posts
  • User-generated content and forums
  • Product descriptions and listings
Common use cases include sentiment analysis, content moderation, topic classification, PII detection, feed ranking, and LLM guardrails.

We provide a comprehensive REST API that allows you to integrate our labelers with your existing systems.
  • API Keys: Create secure API keys in your API dashboard
  • Simple Integration: Send POST requests to /v1/label with your content
  • Real-time Results: Get instant classification results with confidence scores
Check out our API documentation for code examples and detailed integration guides.

  • Community Tier: Basic labeling and criteria optimization with access to our open models and community support
  • Enterprise Tier: Auto-optimization, higher rate limits, access to our latest models, and premium support
Visit our subscription page for detailed feature comparisons.

Our auto-optimization feature uses advanced AI to analyze your test examples and automatically improve your labeling criteria. It:
  • Reviews your current criteria against test examples
  • Identifies areas for improvement
  • Suggests refined criteria and label definitions
  • Flags potentially inaccurate labels
  • Provides performance metrics to track improvements
The process typically takes between 2 and 20 minutes and can significantly boost accuracy.

Zentropi currently has these limitations:
  • Binary labels only: All classifications are binary (0 or 1 -- i.e., yes/no)
  • Text content only: No support for images, audio, video, or other media types
  • US English only: Optimized for US English text; other languages have untested accuracy
For multi-class needs, you can create multiple binary labelers (e.g., separate labelers for "contains sports content" and "contains politics content").

Yes! Anyone can upload their own labeled datasets for bulk evaluation. This allows you to:
  • Test your labeler against known ground truth data
  • Calculate precise accuracy metrics
  • Identify specific areas where your labeler needs improvement
  • Compare performance across different model versions
Upload your datasets in the test examples section of your labeler. Remember that all labels must be binary (0 or 1).

Common Use Cases & Examples

Content Moderation
"Identify usernames that contain offensive language, sexual references, or discriminatory terms."
Returns: 1 = contains offensive content, 0 = clean content
Sentiment Analysis
"Determine if a product review is positive based on the customer's direct experience."
Returns: 1 = positive review, 0 = negative/neutral review
PII Detection
"Identify content that contains personally identifiable information (PII) such as emails, phone numbers, addresses, or social security numbers."
Returns: 1 = contains PII, 0 = no PII detected
Topic Classification
"Determine if a support ticket is about billing issues."
Returns: 1 = billing-related, 0 = not billing-related
Note: Create separate labelers for each topic.
Spam Detection
"Identify promotional emails that contain discount codes, sales language, or marketing offers."
Returns: 1 = spam/promotional, 0 = legitimate content
Compliance Checking
"Flag text that contains medical terms, diagnoses, treatments, or references to health conditions."
Returns: 1 = contains medical content, 0 = no medical content

Troubleshooting

My labeler isn't accurate enough
  • Add more diverse test examples to better represent your data
  • Make your criteria more specific and detailed
  • Try auto-optimization (upgrade to Enterprise for more power)
API integration issues
  • Ensure you're using the correct API endpoint: https://api.zentropi.ai/v1/label
  • Check that your API key is valid and not expired
  • Verify the request format matches our documentation
  • Review error messages for specific validation issues
Can't access certain features
  • Some features require subscriptions - check your subscription status
  • Contact support if you believe you should have access