Moderate Chinese text without a Chinese entity.

zhsafe is one API over several major Chinese moderation engines. Send chat lines, nicknames, posts or model output and get back pass, review or block, with one set of categories and an English explanation of every verdict. Billed in USD.

Example: an in-game chat message, checked with scene chat
blockads 0.92, agreement 1.0

Asks players to add an outside WeChat account to claim a discount, which moves them off your platform.

Try it on your own text

The playground runs the same pipeline as the API, up to 500 characters per check. Results come back in the exact shape of POST /v1/text/check.

0 / 500
Examples

Pick an example or write your own text, then select Check text. The verdict, the flagged categories and the matched characters appear here.

The same check with the API (needs an API key)

For products that serve Chinese-speaking users

Global moderation tools are tuned for English and for Western policy. zhsafe checks Chinese text against the rules that apply to it, and tells you what it found in English.

  • Games with China servers

    Chat, player nicknames and guild names, checked before other players see them.

    chatnickname
  • Cross-border communities

    Posts, comments and replies from Chinese-speaking members, in the same queue as the rest of your moderation.

    postcomment
  • AI products

    Chinese prompts before they reach your model, and model responses before they reach your users.

    llm_inputllm_output
  • Cross-border e-commerce

    Product titles, listings and ad copy, including off-platform contact details and absolute claims like 最 and 第一 that China's Advertising Law restricts.

    ad

What you get instead of a mainland cloud account

Mainland moderation services ask for a local business license and real-name verification, then bill in RMB. zhsafe puts them behind one account you can open from anywhere.

No Chinese entity needed

Sign up, get an API key and pay in USD. Docs and support are in English.

One API, one category set

Several engines behind one endpoint, their labels mapped onto 13 shared categories. Write your policy once, per category, and it holds whichever engine answered.

Consensus instead of one opinion

With the consensus strategy several engines check each text. Agreement means fewer false positives and fewer misses. When they disagree you get review, not a forced answer.

Every verdict explained in English

Each flagged category comes with the matched characters and a one-line explanation, so your trust and safety team can review decisions without a translator.

How a verdict is made

Every request goes through the same four steps, whether it takes one engine or several.

  1. Normalize

    Full-width characters, spacing tricks and variant forms are folded together, so look-alike text gets the same result.

  2. Route

    The scene decides which engines see the text and which thresholds apply. A nickname is judged differently from a forum post.

  3. Merge

    Engine labels are mapped to one category set. Each category records its score and the share of engines that flagged it.

  4. Explain

    You get pass, review or block, the categories with matched spans, and a short English explanation for each.

The categories

  • politics
  • terrorism
  • violence
  • porn
  • gambling
  • drugs
  • illegal
  • abuse
  • minors
  • ads
  • spam
  • religion
  • other

What each category covers

Request early access

zhsafe is in early access. We're onboarding a small number of teams and issuing API keys by email.

  • Tell us what you moderate and roughly how much of it.
  • We'll reply to talk through your scenes and issue a key.
  • Billing is in USD, per check.