Sexual Content
Detects explicit sexual content, suggestive flirting, solicitation, minor-related sexual risk, and adult content distribution.
Suitable for communities, chat, comments, fiction, UGC, and LLM-generated text moderation.
Moderation Coverage
Six major directions covering mainstream text risks
The platform provides standard capabilities, policy templates, and manual policy controls so teams can integrate quickly and then refine rules for their own business logic. It also supports moderation for both Chinese and English text.
Illegal Content
Covers gambling, drugs, fraud, underground trade, privacy abuse, and circumvention-related violations.
Suitable for content platforms, ecommerce, tools, AI apps, and enterprise compliance workflows.
Political Sensitivity
Detects sensitive expressions related to public institutions, political figures, regional conflict, historical controversy, and ideology.
Suitable for LLM output review, pre-publication moderation, and manual review routing.
Abuse & Hate
Detects personal attacks, discriminatory insults, hate speech, and hostile expressions.
Suitable for comment moderation, customer support chats, communities, and live interactions.
Spam & Promotion
Detects off-platform diversion, contact info drops, ecommerce promotion, financial marketing, and recruitment spam.
Suitable for anti-spam and anti-diversion controls across communities, social platforms, and brand sites.
Violence & Extremism
Detects violent harm, terrorism, self-harm risk, and weapons or dangerous goods related content.
Suitable for high-risk moderation, AI safety, and strict risk-control scenarios.
Bilingual Moderation
The model supports moderation for both Chinese and English text
The model supports Chinese social content, community comments, customer-service conversations, danmu, news comments, and other text scenarios. It also supports English moderation with a unified result schema, making it easier for cross-region teams to share one policy set and one API.
Chinese moderation
Covers common Chinese internet expressions, abbreviations, obfuscated variants, and business content scenarios for communities, content platforms, and LLM applications.
English moderation
Supports English risk-text detection for overseas products, global communities, and English generated-content moderation, reducing multilingual integration cost.
Unified output schema
Whether the input is Chinese or English, the API returns one consistent schema of public labels, categories, and scores for client apps, policy engines, and operations systems.
Policy Controls
Multiple policy modes with manual configuration support
Multiple policy templates
Supports allow, block, downgrade, review, and alert templates for different moderation goals and business scenarios.
Manual policy configuration
Policies can also be configured manually by top-level category, public label code, threshold, and traffic source.
Bonus quota after company verification
New users receive 1,000 free credits, and approved company verification unlocks another 50,000 credits for evaluation and onboarding.
Delivery Modes
Supports both public cloud and private deployment
Public cloud service
A cloud API for text moderation, directly callable by API or HTTP SDK, with high concurrency capacity and a 99.9999% service availability target.
Private deployment
Deploy the moderation package on the customer's own servers and run text moderation inside a LAN or private network for stronger data privacy.
Result Schema
Public label preview, definitions, and business guidance
Sexual Content
Business-facing definitions, handling advice, and representative detection scope. Three labels are shown here; total labels: 7 Contact support for the full catalog.
Sexual Content
sexual_solicitation
Sexual Solicitation
The text contains prostitution, sexual service solicitation, or adult traffic diversion.
Recommended to block directly and record account risk.
Sexual Content
sexual_teasing
Sexual Suggestiveness
The text contains teasing, suggestive, or borderline sexual expressions.
Block, fold, or send to review depending on the business scenario.
Sexual Content
explicit_sexual_content
Explicit Sexual Content
The text directly describes sexual acts, sexual crimes, or explicit adult scenes.
Recommended for high-priority blocking.
Total 7 labels. Contact support for the full catalog.
Illegal Content
Business-facing definitions, handling advice, and representative detection scope. Three labels are shown here; total labels: 8 Contact support for the full catalog.
Illegal Content
gambling_and_betting
Gambling & Betting
The text involves gambling platforms, betting activities, or abnormal wagering guidance.
Recommended to block directly and raise account risk level.
Illegal Content
drugs_and_contraband
Drugs & Contraband
The text involves drugs, prohibited medicine, or other restricted goods.
Recommended to block directly.
Illegal Content
fraud_and_blackmarket
Fraud & Underground Trade
The text involves fraud, forgery, pyramid schemes, organized crime, or other black/grey market activities.
Recommended for high-priority blocking and linkage with fraud control.
Total 8 labels. Contact support for the full catalog.
Political Sensitivity
Business-facing definitions, handling advice, and representative detection scope. Three labels are shown here; total labels: 5 Contact support for the full catalog.
Political Sensitivity
public_affairs_sensitive
Public Affairs Sensitivity
The text involves governments, institutions, regulations, or sensitive public governance discussion.
Block, downgrade, or review depending on the business scenario.
Political Sensitivity
political_figure_sensitive
Political Figure Sensitivity
The text involves political figures, major public figures, or related controversial expressions.
Handle with context-aware and region-specific policies.
Political Sensitivity
separatism_and_regional_conflict
Separatism & Regional Conflict
The text involves sovereignty disputes, separatist rhetoric, or escalated regional conflict expressions.
Recommended for high-priority review or blocking.
Total 5 labels. Contact support for the full catalog.
Abuse & Hate
Business-facing definitions, handling advice, and representative detection scope. Three labels are shown here; total labels: 2 Contact support for the full catalog.
Abuse & Hate
abusive_attack
Abusive Attack
The text contains obvious personal attacks, verbal abuse, or malicious provocation.
Block, downgrade, or route to manual review.
Abuse & Hate
discriminatory_hate
Discriminatory Hate
The text contains discriminatory insults toward groups based on region, gender, profession, appearance, or ethnicity.
Recommended for high-priority blocking.
Total 2 labels. Contact support for the full catalog.
Spam & Promotion
Business-facing definitions, handling advice, and representative detection scope. Three labels are shown here; total labels: 5 Contact support for the full catalog.
Spam & Promotion
offsite_lead_gen
Off-platform Diversion
The text tries to move users off the current platform through add-friend, DM, or diversion wording.
Recommended to block or rate-limit.
Spam & Promotion
contact_exchange
Contact Info Promotion
The text contains WeChat, QQ, phone numbers, URLs, public accounts, or other contact entry points.
Recommended to block or mask based on account profile.
Spam & Promotion
commerce_marketing
Commerce & Course Promotion
The text includes product promotion, ecommerce sales, or training-course marketing.
Throttle, fold, or block according to anti-ad rules.
Total 5 labels. Contact support for the full catalog.
Violence & Extremism
Business-facing definitions, handling advice, and representative detection scope. Three labels are shown here; total labels: 4 Contact support for the full catalog.
Violence & Extremism
violent_harm
Violent Harm
The text involves gore, physical harm, murder, or hired violence.
Recommended for high-priority blocking.
Violence & Extremism
terror_extremism
Terror & Extremism
The text involves terrorist organizations, terrorist incidents, or extremist content.
Recommended to block directly with strict retention.
Violence & Extremism
self_harm
Self-harm Risk
The text involves suicide, self-harm, or expressions of harming oneself.
Recommended for priority blocking and, where applicable, care escalation or human intervention.
Total 4 labels. Contact support for the full catalog.
Other Risks
Business-facing definitions, handling advice, and representative detection scope. Three labels are shown here; total labels: 2 Contact support for the full catalog.
Other Risks
noisy_text
Low-quality Noise
The text is dominated by gibberish, repetition, spammy filler, or low-information content.
Recommended to discard, fold, or route to a low-priority queue.
Other Risks
other_risk_signal
General Risk Signal
Non-standard but noteworthy risk signals are detected and should be judged with additional business rules.
Recommended for manual review or a secondary rule engine.
Total 2 labels. Contact support for the full catalog.
Safe
Business-facing definitions, handling advice, and representative detection scope. Three labels are shown here; total labels: 3 Contact support for the full catalog.
Safe
safe_pass
Safe Pass
No clear policy violation is detected in the current text.
Allow by default, or combine with business dictionaries and random review if needed.
Safe
safe_context
Neutral Public Discussion
The text contains neutral discussion, objective statements, or ordinary mentions that do not directly form a risk.
Usually safe to allow, with optional low-priority review depending on the business context.
Safe
lifestyle_normal
Normal Lifestyle Content
The text is mainly about daily life, health, parenting, or ordinary social topics.
Usually safe to allow directly.
Total 3 labels. Contact support for the full catalog.
Delivery Value
A clear path from trial to production
Policy-ready moderation results
Results include top-level risk type, public label code, score, and summary buckets so they can be applied directly to policy actions.
Consistent web and API outputs
The website, user center, and API share the same result schema, making it easy to move from trial to production integration.
Operations dashboard included
The admin console shows registrations, visits, moderation trends, and remaining user quota for ongoing operations.
Chinese and English support
The main website, product pages, API docs, and user-facing pages support both Chinese and English for international delivery.
Typical Scenarios
Covers mainstream moderation and LLM risk-control scenarios
LLM-generated text moderation
Review LLM-generated text before release and identify toxic, illegal, abusive, or politically sensitive content.
Community text moderation
Suitable for posts, comments, DMs, and group chats to quickly identify sexual, political, spam, and abusive content.
Live comment moderation
Detect high-risk text in live comments and video chat streams to support real-time blocking, warnings, or human review.