Jevfanity

A tiny profanity + moderation API powered by Jev and deployed on Cloudflare Workers.

One request to Jev evaluates multiple moderation categories in parallel and returns a probability for each one.

Hosted version

A public instance is available at api.jevfanity.app.

The hosted API is free for the time being. Please use it responsibly and do not abuse the service, evade rate limits, or send unnecessary automated traffic. Availability and free access are not guaranteed and may change as usage and operating costs evolve.

curl -X POST https://apijevfanity.app/v1/moderate \
  -H "Content-Type: application/json" \
  -d '{"text":"your message here","level":"medium"}'

API

POST /v1/moderate

{
  "text": "your message here"
}

Optional:

{
  "text": "your message here",
  "level": "medium",
  "threshold": 0.75,
  "categories": ["profanity", "slur", "harassment"]
}

Moderation levels:

• low: Allows isolated f-bombs, casual swearing, and mild insults; flags extreme or dangerous content.
• medium: PG-13-style moderation and the default level.
• strict: Child-oriented moderation that flags any profanity, slur, targeted abuse, threat, or age-inappropriate sexual language.

Supported categories:

• profanity
• slur
• harassment
• threat
• sexual

Example response:

{
  "flagged": true,
  "score": 0.97,
  "level": "medium",
  "threshold": 0.75,
  "categories": {
    "profanity": 0.97,
    "slur": 0.03,
    "harassment": 0.12,
    "threat": 0.01,
    "sexual": 0.02
  },
  "flagged_categories": ["profanity"],
  "model": "jev-1.13.0",
  "usage": {
    "input_tokens": 320,
    "output_tokens": 40
  },
  "processing": {
    "estimated_tokens": 14,
    "chunks": 1,
    "max_tokens_per_chunk": 100,
    "unicode_confusable_tokens": 0
  }
}

flagged is true when at least one selected category violates the chosen level's policy with a score that meets or exceeds the confidence threshold. needs_review is true for a non-flagged result with a category score of at least 0.5, providing a review state for moderate signals without treating them as definitive violations.

Setup

npm install
cp .dev.vars.example .dev.vars

This repository is an npm workspace monorepo. The Cloudflare Worker lives in packages/api, and the publishable client lives in packages/sdk. Run npm run dev or npm run deploy from the repository root for the API, and npm run build to build the SDK.

Only the SDK is published to npm. After logging in with npm login, preview it with npm pack --dry-run --workspace jevfanity, then publish it with npm run publish:sdk.

Put your TypeSafe key in .dev.vars:

TYPESAFE_API_KEY=...

Then:

npm run dev

Deploy

Store the TypeSafe key as a Cloudflare Worker secret:

npx wrangler secret put TYPESAFE_API_KEY

Optional: protect the public Jevfanity endpoint with your own bearer token:

npx wrangler secret put JEVFANITY_API_KEY

Then deploy:

npm run deploy

If JEVFANITY_API_KEY is configured, clients must send:

Authorization: Bearer YOUR_JEVFANITY_API_KEY

Example

curl -X POST https://jevfanity.app/v1/moderate \
  -H "Content-Type: application/json" \
  -d '{"text":"this is fucking awful","level":"medium"}'

Browser usage

const result = await fetch("https://jevfanity.app/v1/moderate", {
  method: "POST",
  headers: {
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    text: message,
    level: "medium",
  }),
}).then((response) => response.json());

if (result.flagged) {
  console.log(result.flagged_categories);
}

npm SDK

Install the typed client:

npm install jevfanity

import Jevfanity from "jevfanity";

const jevfanity = new Jevfanity({
  baseUrl: "https://api.jevfanity.app",
  apiKey: process.env.JEVFANITY_API_KEY,
});

const result = await jevfanity.moderate(message, {
  level: "medium",
  categories: ["profanity", "harassment"],
});

if (result.flagged) {
  console.log(result.flagged_categories);
}

The SDK uses the global fetch, works in Node.js 18+ and browsers, and accepts a custom fetch implementation through the constructor for testing or runtimes with their own HTTP client. Failed HTTP requests throw JevfanityError with status and parsed body properties.

Why multiple questions?

Jev's Noul primitive returns the probability that a yes/no proposition is true. Jevfanity asks one focused question per moderation category rather than asking one broad question like "is this bad?"

That gives applications both:

{
  "flagged": true
}

and the underlying category probabilities for custom policy decisions.

Notes

• The default threshold is 0.75.
• The default moderation level is medium.
• Override the deployment default with DEFAULT_LEVEL set to low, medium, or strict.
• Override the deployment default with DEFAULT_THRESHOLD in wrangler.jsonc.
• Clients can supply a request-specific threshold from 0 to 1.
• The Worker limits input to 10,000 characters per request.
• Text is split into chunks of at most 100 whitespace-delimited tokens with a 10-token overlap. Category scores are the maximum score across all chunks.
• Unicode text is allowed. Tokens made from lookalike Unicode characters are also compared in normalized form so disguised profanity or slurs can be evaluated by meaning and context. Unicode characters alone never cause a flag.
• The API allows 20 moderation requests per minute for each client IP in each Cloudflare location. A limited request returns 429 with Retry-After: 60.
• IP-based rate limiting means users on the same shared network may share a limit bucket. Use an authenticated user or API-key identifier instead when the API gains per-user authentication.
• CORS is open by default. Tighten Access-Control-Allow-Origin before using a browser-facing production deployment with sensitive data.

Contributing

Contributions are welcome. See CONTRIBUTING.md for setup, development, and pull request guidance.

License

Jevfanity is available under the MIT License.