Bring the copy you already have. Entropy turns it into a human draft you can edit, refine, and rewrite in place.
Paste AI-generated text, upload a plain-text file, or begin with our sample.
Whole-text mode returns a complete rewritten document.
curl -X POST "https://www.getentropy.ai/v1/chat/completions" \
-H "Content-Type: application/json" \
-d '{
"model": "entropy-v2-main",
"messages": [
{ "role": "user", "content": "I always remembered the man who jumped into lava to save someone." }
],
"rewrite": {
"mode": "document",
"language": "en",
"domain": "social",
"structure": "preserve"
}
}'Entropy is a rewriting tool that transforms AI-generated text into prose that reads the way a person actually writes. Paste what you have, click Rewrite, and the output preserves your meaning while replacing the patterns that make AI text feel generic.
Completely. Every visitor gets roughly 2,500 words per day with no account at all. Sign in and that allowance increases to five times as much. There is no trial period, no credit card, and no paywall waiting at the end.
We log inputs and outputs so we can measure quality and improve the model over time. Your text is never sold, never shared with third parties, and never used to train anyone else’s product.
General-purpose chatbots rewrite by following instructions, and the result still carries the same predictable cadence. Entropy uses a purpose-built model trained exclusively on verified human writing, so the output picks up genuine stylistic variation rather than swapping one AI voice for another.
Rewrites may lower an AI-detection score, but that’s a side effect, not the goal. Entropy is designed to produce writing with real human variation — detection scores drop because the output genuinely reads differently, not because we’re gaming a classifier.
We’re expanding domain coverage, improving long-document handling, and preparing the multilingual Gemma 4 model for an open-source release. Selection rewriting and Mandarin support are already available in the Studio.
Yes. The V1 fine-tuned weights are published on HuggingFace (huggingface.co/ysong21/entropy-v1-lora), and the full codebase is public. The multilingual Gemma 4 V2 weights will follow the hosted release. Today you can integrate through the OpenAI-compatible API; the MCP server is coming soon.
It sets how far the rewrite can deviate from your original phrasing. All the way left keeps things close and predictable; all the way right gives the model more room for expressive, varied language. At higher settings you may occasionally see repetitive or unstable output — if that happens, just lower the slider a notch and rewrite again.