DEV Community: Samrat Dutta

Rewrite: a free AI rewriter, paraphrasing tool and document editor for Windows, macOS and Linux

Samrat Dutta — Mon, 26 Oct 2026 10:11:04 +0000

I write research papers in LaTeX and reports in Word. For years my routine was the same: draft a paragraph, paste
it into a web paraphraser, hit the word limit, paste half of it again, copy the result back, then fix the LaTeX
commands the tool had mangled. I built Rewrite so that all of that happens inside one desktop app that also opens the
files I actually work in.

Rewrite is free. There is no account and no word limit. This post walks through everything it does.

An AI rewriter that makes drafts read as human

If you start from an AI draft, the text has a recognisable sound: the same openers, the same padded phrases, the
same sentence shapes. Rewrite's Auto-rewrite goes through a selection or the whole document and changes that.

You choose how far it goes:

Light rewords wordy phrases and swaps a few words, about 15% of them. The meaning stays where it was.
Medium also replaces about 30% of the eligible words with synonyms that fit the context.
Strong changes up to half of the eligible words and rephrases whole sentences.

You can also pick the kinds of change: rewording wordy phrases, swapping words, and rephrasing whole sentences. Then
Rewrite shows every change one at a time. Accept it, reject it, or pick a different alternative before you move on.
Nothing is applied behind your back.

In my own tests, text that went through Auto-rewrite passed AI detectors and plagiarism checkers. Test it on your
own writing; detectors change often.

One honest note, which is also in the app: rewording someone else's text does not make it your own work. Cite your
sources.

A paraphrasing tool you can steer word by word

Auto-rewrite is the fast path. Most of the time I work more slowly, one spot at a time.

Click a word and the side panel lists alternatives, ranked by how well each one fits that exact sentence. Each
has a fit bar and a tag that says whether it is a synonym, a related word or a word that fits the context. Many come
with a short definition, so you can tell "rapid" from "hurried" without leaving the page. Alt+1 to Alt+9 applies the
first nine without touching the mouse.

Select a phrase to reword it as a unit. Rewrite also highlights wordy phrases on its own. "Due to the fact that"
gets a peach highlight, and clicking it offers "because" (keeps 97% of the meaning) or "as".

Press Ctrl+E for new versions of the whole sentence under the cursor. Every version shows two numbers: how much
of the meaning it keeps and how much of the wording it changed. A rewrite that keeps 98% of the meaning while
changing 60% of the words is usually the one you want; one that keeps 85% deserves a second look.

Spelling and grammar

Spelling slips get a red underline and grammar slips a blue one. Click either for the fix. It works in every file
type Rewrite opens.

Word documents, without Word

Rewrite opens and saves .docx, .doc, .odt, .fodt and .rtf with their formatting intact: styles, lists, pictures,
tables, charts, formulas, footnotes and links. A formatting toolbar handles the everyday edits, with Ctrl+B, Ctrl+I,
Ctrl+U and Ctrl+K for a link. Ctrl+P shows the document as it will print.

So you can open a report someone emailed you, rewrite three paragraphs, fix the grammar and send it back as .docx,
and the other person sees their own styles and tables.

LaTeX, with the PDF beside it

Open a .tex file and Rewrite shows the typeset PDF next to the source, built with your own LaTeX installation
(MiKTeX, MacTeX or TeX Live). F5 rebuilds it. When you rewrite, commands, maths and \cite keys are never changed;
only the prose moves. That was the reason I started the project, and it is still the part I use most.

Markdown gets the same treatment, with a rendered preview as you type. HTML tags and source code are left alone too.

PDF editing

PDFs open in their own view. You can highlight, underline, strike out, add notes, draw, white out an area, add text
and pictures, and change a line of text in place. Pages can be turned, moved, deleted, inserted or merged from
another PDF.

Rewrite can also turn a PDF into an editable document with its headings, lists, tables and pictures, mending the
words that were broken across line ends. That is useful when the only copy of something is a PDF.

Export and convert

Anything Rewrite opens can be written out as PDF, Word, OpenDocument, RTF, HTML, Markdown, LaTeX or plain text
(Ctrl+Shift+E). A Markdown draft can become a Word file for a supervisor; a Word report can become LaTeX for a
journal template.

The small things

Find and replace with Ctrl+F and Ctrl+H, and F3 for the next match.
Open a file with Ctrl+O, or drop it on the window.
A dark theme for late nights.
F1 opens the full guide inside the app.

Price, account, privacy

Rewrite is free, with no account and no word limit. On first start it offers a one-time download of about 400 MB
for its best suggestions; after that it works offline. If you skip the download it runs in Lite mode with a
built-in thesaurus.

Download

System	File
Windows 10 and 11 (Intel, AMD, ARM)	Setup .exe, no admin rights needed
macOS 15 or later (Apple Silicon and Intel)	.dmg
Debian and Ubuntu (x64 and ARM)	.deb
Any Linux (x64 and ARM)	AppImage

Get it at https://samratduttaofficial.github.io/Rewrite/ . The source is on GitHub at
https://github.com/SamratDuttaOfficial/Rewrite . The Windows installer is not signed yet, so SmartScreen may ask you
to confirm the first time.

For the curious: Rewrite is written in Rust with GTK 4, and one GitHub Actions workflow builds all six installers.

If a suggestion is bad or a file opens wrong, open an issue on GitHub and tell me. That is the fastest way it gets
better.

WaterSheep: an open-source alternative to Jev

Samrat Dutta — Sat, 10 Oct 2026 10:02:26 +0000

TypeSafe's Jev made a simple idea popular: a model that doesn't write text, only makes decisions. You send some state and typed questions, and you get back choices, scores and probabilities your code can act on. Jev itself is closed, and you pay per token.

When it launched, I saw everyone hyping it up and thought: I can build an open one. So I did. WaterSheep is an open-source alternative to Jev. It answers the same kinds of questions about any text:

noul: yes/no, with the probability of yes
choice: one of your options
score: a rating scale, with the expected level
multi: every option that applies (Jev doesn't have this one)

Every answer includes a probability for every option.

Already using Jev? Keep your code

Run WaterSheep as a local server:

pip install git+https://github.com/SamratDuttaOfficial/WaterSheep
watersheep --model samratduttaofficial/WaterSheep --serve

It answers POST /v1/systemone with the same request and response shape as Jev, so TypeSafe's own Python SDK works against it. Point the SDK at it and run your code as usual:

export TYPESAFE_BASE_URL=http://127.0.0.1:8766
python triage.py

The script is plain Jev code:

from typesafe_sdk import Choice, Noul, Score, TypeSafeClient

client = TypeSafeClient()  # or TypeSafeClient(base_url="http://127.0.0.1:8766")

r = client.system_one(
    state={"ticket": "I was charged twice for order #4411. Refund the duplicate today."},
    questions={
        "refund": Noul(instructions="Does the customer ask for a refund?"),
        "team": Choice(
            instructions="Which team should handle this?",
            criteria={"billing": None, "shipping": None, "technical": None},
        ),
        "urgency": Score(
            instructions="How urgent is this?",
            criteria=["low", "medium", "high"],
        ),
    },
)
a = r.answers
print("refund:", a["refund"].noul)                               # 0.9451
print("team:", a["team"].choice, a["team"].probabilities)        # billing {'billing': 0.8817, ...}
print("urgency:", a["urgency"].score, a["urgency"].legend)       # 1.4805 {0: 'low', 1: 'medium', 2: 'high'}

Any API key value works locally. I tested this with typesafe-sdk 0.7.2; Jev's JavaScript SDK is untested so far.

No SDK? It's plain HTTP

curl http://127.0.0.1:8766/v1/systemone -d '{
  "state": "My card was charged twice and the box arrived crushed.",
  "questions": {
    "team": {"type": "choice", "instructions": "Which team should handle this?",
             "criteria": {"billing": "payments, refunds", "shipping": "delivery, damage"}},
    "issues": {"type": "multi", "instructions": "Which problems are reported?",
               "criteria": ["double charge", "damaged item", "late delivery"]}
  }
}'

The multi question returns both problems, each with its own probability. A single choice has to pick one.

Python without a server

from transformers import pipeline

ws = pipeline(model="samratduttaofficial/WaterSheep", trust_remote_code=True)
ws("I was charged twice.", question="Which team should handle this?", options=["billing", "shipping", "support"])

It also runs as ONNX, and the live demo lets you try it without installing anything.

How good is it?

On the in-distribution test split it gets 77.8% accuracy; on held-out datasets it never saw, 61.2%. The probabilities are calibrated (expected calibration error 0.026 and 0.043), so you can set thresholds on them: act automatically when the model is confident, and send the rest to a person. Per-benchmark results, including the weak ones, are in the README, and the details are in the paper.

Limits: English only, long inputs are truncated, rating questions are the weakest type, and it isn't meant for high-stakes decisions on its own.

Open all the way down

Code, weights and the training pipeline are Apache 2.0, and you can retrain it on your own data with one script. WaterSheep is independent and not affiliated with TypeSafe.

GitHub: https://github.com/SamratDuttaOfficial/WaterSheep
Hugging Face: https://huggingface.co/samratduttaofficial/WaterSheep

I'd love feedback, especially examples where it gets things wrong.