zentropi

TL;DR: Copy/paste this into your AI agent...

Install the zentropi skills at https://github.com/zentropi-ai/skills

Zentropi Skills

Agent skills for building systems powered by the Zentropi classification engine. They give AI agents the superpower of fast, accurate, flexible, content labeling.

These skills follow the Agent Skills open standard. They teach AI coding agents how to integrate Zentropi into applications and workflows— from writing policies to optimizing performance to deploying live.

Skills

Skill	Description
zentropi-labeler	Create custom classifiers and label content against them using the Zentropi API.

Quick Start

0. Installation

Automatic installation:

Just ask your agent: "Install the zentropi skills at https://github.com/zentropi-ai/skills"

Skills.sh package manager:

Another fast way to install is with skills.sh:

# Install every skill in this repo
npx skills add https://github.com/zentropi-ai/skills

# …or install just the zentropi-labeler skill
npx skills add https://github.com/zentropi-ai/skills --skill zentropi-labeler

Manual installation:

Copy the skill directory into your agent's skills folder. The agent reads SKILL.md on demand and follows the instructions to build labelers, write policies, classify content, and handle errors.

1. Get an API key

Sign up at zentropi.ai and create an API key.

export ZENTROPI_API_KEY="your-key-here"

2. Classify content

Ask your agent to label content against your own criteria:

Label w/ Zentropi if this is funny: "I tried to catch fog. I mist."

The agent will write a basic policy, call the Zentropi API, and return results:

Label: 1 (criteria matched), Confidence: 0.60

3. Write better policies

Simple one-line criteria work for prototyping, but production policies should follow the CoPE policy format — a structured template with defined terms, inclusion criteria, exclusion criteria, and examples.

You may also use policies that our community has publicly shared on zentropi.ai. For example, try out this sexual content labeler with labeler_version_id = 'b5c41878-e659-4b3d-be70-fd85830af4d5'

What is CoPE?

CoPE (Content Policy Evaluator) is Zentropi's latest classification model. You write policies in plain English describing what to detect, and CoPE returns a binary label (match / no match) with a confidence score. No training data, no fine-tuning — just describe what you're looking for.

• Text and visual content — use cope-b-a4b for text; cope-b-a4b-mm for multimodal (subscriber only)
• Token context — 256k tokens (combined policy + content)
• Sub-second inference — typical response times under 1s; higher rate limits for subscribers

Learn more: CoPE model card

Resources

• Zentropi Platform — create labelers, manage API keys
• API Documentation
• Policy Writing Guide
• CoPE-B Model Card

License

See Zentropi Terms of Service.