Skip to content
DocInfra

AI humanizer APIPublic Beta

An AI humanizer API that keeps the facts.

POST a text, get it back sounding like a person wrote it: same meaning, same language, and every number, date, link, email, citation and protected term exactly where it was. One key, one JSON call.

Available now in Public Beta. Open to any workspace.

# Rewrite a text so it reads naturally

curl -X POST https://api.apidocinfra.com/v1/text/rewrite \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "In today’s fast-paced world, it is important to note that our platform delivers value to 1,200 customers.",
    "mode": "natural",
    "intensity": "medium",
    "protected_terms": ["DocInfra"]
  }'

Real endpoints and fields from the API contract. Replace the key and the file.

Why developers pick it

  • Preservation by construction

    What must not change is swapped for markers before the model sees the text and verified on the way back. If a marker is lost, you get an error, not a guess.

  • Meaning first

    The prompt is versioned in code and ranks meaning, facts, intent and language above naturalness. Nothing is inflated to look human.

  • Output checks

    Length ratio, language and stray instructions are checked on every chunk, with one strict retry before the call fails.

  • Your text is not stored

    Only counts, timings and cost are recorded. The text never goes to the database or the logs.

  • Stable API

    Modes, intensity, protected terms and idempotency are part of the contract; the model behind it can change without breaking you.

  • Documents too

    The same engine rewrites DOCX files with the layout kept, as an async job.

One call

  1. 1

    POST /v1/text/rewrite with text, an optional mode (natural, professional, conversational, concise), intensity and protected_terms.

  2. 2

    The API protects numbers, dates, URLs, emails, citations, code and your terms, rewrites the rest in chunks, and checks that everything protected came back once and in order.

  3. 3

    You get the rewritten text with the detected language and metrics: words in, words out, percentage changed.

We do not promise that the output evades AI detectors, and we do not measure against them. The product is judged on naturalness, preserved meaning and exact facts.

Questions

Which languages?
English and Spanish are benchmarked; other languages are accepted but not measured.
How long can a text be?
Up to 10,000 words per call. Longer inputs are split on paragraphs and rejoined in order.
Is it available now?
The API is implemented and in beta. Create a workspace to get a key with it enabled.