POST /api/v1/name

Check a name

Usernames and display names, including the ones spelled to slip past filters.

Usernames and display names, including the ones spelled to slip past a word list. Check it at registration, and again when somebody changes it.

Parameters

Field	Type	Required
`name`	string	yes	Up to 200 characters.
`locales`	string[]	no	Your site's languages, up to ten.
`ai`	boolean	no	No model reads a name, so this is `false` and cannot be turned on. Sending `"ai": true` is refused with `422 ai_unavailable` rather than accepted and ignored.
`reference`	string	no	Your own id for the thing being judged, up to 120 characters. Never parsed, stored and handed back, so a verdict is findable later by the name you already use for it.
`policy`	string	no	Which of your policies to judge under. Absent means the project's policy, then your default; an account with none gets the shipped thresholds.
`project`	string	no	Which of your projects the verdict is filed under. Absent means your default project.
`actor`	string	no	Your own id for whoever wrote it. Filed with the verdict, and used to move the line when a policy asks for reputation.

Request

curl https://toxicfilter.com/api/v1/name \
  -H "Authorization: Bearer tf_live_xxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{ "name": "Ｈ4ｔ3Ｒ_88", "locales": ["en"] }'

Response

{
  "decision": "review",
  "flagged": ["evasion"],
  "scores": { "evasion": 0.74 },
  "signals": [
    {
      "category": "evasion",
      "score": 0.74,
      "detector": "evasion",
      "reason": "Full-width characters and digit substitution used to obscure a word.",
      "evidence": ["Ｈ4ｔ3Ｒ"]
    }
  ],
  "used_ai": false,
  "took_ms": 3,
  "credits": { "remaining": 4890, "renews_at": "2026-09-30T00:00:00+00:00" }
}

Evasion is its own answer

evasion is scored separately from whatever was being hidden, and that is useful on its own: leetspeak, homoglyphs, full-width characters and invisible codepoints in a username are a signal about intent even when the underlying word is ambiguous. A name written to get past a filter tells you something a clean name does not.

Why no model reads it

A name is a shape, not a piece of writing: there is no argument to follow, no sarcasm and no target, and the question is whether the characters were chosen to get past a list. The normaliser and the word lists answer that, and they answer it in under a millisecond for one credit. A model would add a sentence of prose and several credits to the same answer.

So ai here is not a default you can override. It is refused, because the one thing it must not be is accepted and ignored: used_ai: false with nothing anywhere saying why is how a caller pays for a habit and believes they bought something. The same is true of an address and a link. A bio is writing, so a signup can ask for the model.

Names are short, so context is thin and false positives cost more here than anywhere else. Treat review on a name as "look at it", never as "reject it": plenty of real people have names a filter finds interesting.