Frontier AI models
for everything you build.
Real-time content moderation, multilingual classification, and developer-first APIs. Trained on 24,500 samples across 7 categories — built to scale.
POST /v1/moderate → { "text": "hello world", "lang": "auto" } ← { "label": "normal", "score": 0.857 }
Capabilities
Everything you need
From prototype to production, Pocai provides the tools and models you need.
Content Intelligence
Advanced ML models trained on real-world data to understand and classify text with high precision.
Moderation AI
State-of-the-art content moderation powered by Pocai Moderate. Detect harmful content with unmatched accuracy.
Low Latency
Sub-100ms response times with globally distributed infrastructure. Built for scale from day one.
Vision & Image (Soon)
Coming soon: image generation and analysis capabilities powered by multimodal AI models.
Usage Analytics (Soon)
Track your API usage, monitor costs, and view detailed metrics from your console dashboard.
Developer API
Simple REST API with detailed documentation. Integrate content moderation into your app in minutes.
How It Works
Three steps to moderate
No setup, no training. Just type and see results instantly.
Type any message
Write a sentence, question, or phrase in any language. Messages are processed in real time with sub-second latency.
AI analyzes across 7 categories
Our XLM-RoBERTa model scores your text for profanity, threats, harassment, hate speech, sexual content, jokes, and normal content — simultaneously.
Get instant confidence scores
Results show exact percentages for each category. If something looks wrong, submit a report to help improve the model.
Pocai Moderate
In-house content moderation model trained on 24,500 samples across 7 categories. Redesigned as a signal map so every category feels visible, measurable, and alive.
Signal score
Category confidence map
Dataset
24.5K
Accuracy
83%
Latency
<100ms
Category layer
Seven detection channels
Profanity
Very high94.7%
30 samples
Normal
Stable85.7%
125 samples
Sexual
Stable85.7%
50 samples
Hate Speech
Stable84%
50 samples
Joke
Stable83.8%
50 samples
Threat
Good80.9%
25 samples
Harassment
Improving62.7%
17 samples
Macro F1
Balanced performance across all categories
Weighted F1
Weighted by real evaluation support
Accuracy
Measured on the held-out test split
Our model
Purpose-built AI for content moderation.
Pocai Moderate
BetaContent Moderation Engine
Specialized model for detecting and filtering harmful content in real time. Built on XLM-RoBERTa-large with 560M parameters and 83% accuracy.
Get Started
Ready to build?
Start building with Pocai today. Free to try, pay only for what you use.
Start building
Free tier with generous limits. No credit card required.
- Free to start
- Pay as you go
- No subscriptions
- Sub-100ms latency