Self-Hosted AI Moderation

Content Moderation.
Intelligent & Fast.

Multi-modal AI engine with native support for Uzbek, Kazakh, Russian, and English. Text, images, audio, video — zero external API calls.

Documentation
Text
Images
Speech
Music
Video
Text
Images
Speech
Music
Video

Multi-Modal Analysis

Analyze text, images, speech, music, and video through a single unified API. Powered by ONNX Runtime with CPU, CoreML, and CUDA support.

Detect profanity, toxicity, hate speech, NSFW content, violence, copyright violations, and dangerous audio events — all on your own infrastructure.

POST /v1/moderate/text

{

"text": "Check this content...",

"lang_hint": "uz",

"sync": true

}

// Response

{

"status": "completed",

"final_action": "allow",

"categories": [

{ "category": "toxicity", "confidence": 0.03 },

{ "category": "hate_speech", "confidence": 0.01 }

],

"processing_time_ms": 42

}

UZ

O'zbekcha

KK

Қазақша

RU

Русский

EN

English

Native Language Support

First-class support for Uzbek and Kazakh including dual-script normalization (Cyrillic/Latin), leetspeak detection, and language-specific profanity dictionaries.

Zero External APIs

Fully Self-Hosted

Deploy on your own infrastructure. All ML inference runs locally via ONNX Runtime. Your data never leaves your servers. Supports CPU, Apple Silicon, and NVIDIA GPUs.

PERFORMANCE

Benchmarks

Built for Speed

<500ms
Text
<2s
Image
<10s
Speech (1 min)
<30s
Music (3 min)
<45s
Video (1 min)
10K+
Jobs per day / GPU
0
External API calls
176
Languages detected
Detection

What We Detect

Toxicity & Hate Speech

Insults, discrimination, threats

NSFW & Nudity

Pornography, suggestive content

Violence & Gore

Weapons, drugs, dangerous content

Copyright

Audio fingerprint matching

Profanity

Multi-language dictionaries

Spam & Fraud

Scam detection, spam filtering

Dangerous Sounds

Gunshots, screams, explosions

Explicit Lyrics

Vocal separation + ASR analysis

Built With

Production-Ready Stack

Rust Backend

Axum + Tokio async runtime. Memory-safe, blazing fast, zero-cost abstractions. Handles thousands of concurrent requests.

ML Models

XLM-RoBERTa, CLIP ViT-L/14, Whisper, YAMNet, Demucs v4. All running via ONNX Runtime with hardware acceleration.

Docker Deploy

One-command deployment with Docker Compose. PostgreSQL + pgvector, RabbitMQ, MinIO, dedicated workers per content type.

Ready to Moderate?

Deploy SAPI AI on your infrastructure and start moderating content in minutes. Full API documentation and Docker images included.

SAPI Platform