sarihصريح
Scan Arabic text for toxicity, hate speech, and spam. Dialect-aware. Fully offline.
13 filters · 3 severity levels · fully offline — Arabic-first content moderation for social apps, chatbots, and UGC platforms.
$
pip install sarih
Chapter I
The Problem
Arabic platforms need Arabic moderation.
- English-First Tools
Existing moderation APIs are built for English. They miss Arabic slang, dialect-specific insults, and culturally specific toxicity. - Dialect Blindness
An insult in Egyptian Arabic looks nothing like the same insult in Gulf Arabic. MSA-only tools miss dialect-specific content entirely. - Cloud Dependency
Sending user content to third-party APIs. Privacy concerns, latency, cost, and single point of failure. - Data Privacy
Sensitive user content leaving your infrastructure. Compliance nightmares. sarih runs entirely on your machine.
Chapter II
13 Filters
Every type of toxic content. Caught.
- Profanity
Swear words and vulgar language across all Arabic dialects. Not just MSA dictionaries. - Hate Speech
Slurs and dehumanizing language targeting ethnic, religious, or social groups. - Spam
Repetitive text, promotional patterns, keyword stuffing, and bot-like content. - Adult Content
Sexually explicit language, innuendo, and inappropriate material. - Violence
Threats, incitement to harm, and graphic descriptions of violence. - PII
Phone numbers, emails, national IDs, and other personally identifiable information in Arabic text. - Misinformation
Common health and political misinformation patterns circulating in Arabic.
Chapter III
Severity Levels
Three levels. Clear action for each.
| Level | Meaning |
|---|---|
| BLOCK | Must remove. Clearly toxic, dangerous, or illegal content. |
| FLAG | Needs human review. Likely problematic but context-dependent. |
| REVIEW | Soft signal. May be fine, but worth a second look. |
Chapter IV
Dialect Aware
Five dialects. One tool.
| Dialect | Arabic |
|---|---|
| MSA | فصحى |
| Egyptian | مصري |
| Gulf | خليجي |
| Levantine | شامي |
| Moroccan | مغربي |
Demo
See It
Chapter V
Commands
6 commands. Zero config.
- scan
Scan a JSONL file for toxic content - check
Check a single text string - pipe
Read from stdin for pipelines - stats
Moderation statistics by filter and severity - clean
Remove or redact flagged content - explain
Describe all filters and severity levels
Chapter VI
Get Started
# Install
$ pip install sarih
# Check a single text
$ sarih check "text to moderate"
# Scan a dataset
$ sarih scan data.jsonl
# Clean a dataset
$ sarih clean data.jsonl --output clean.jsonl
# View statistics
$ sarih stats data.jsonl
# Pipe from stdin
$ cat texts.jsonl | sarih pipe
# Learn about filters
$ sarih explain
Chapter VII
As a Library
Import and moderate in two lines.
from sarih import moderate
result = moderate("text to check")
print(result.severity) # BLOCK, FLAG, or REVIEW
print(result.filters) # list of triggered filters
print(result.dialect) # detected dialect