A quieter way to write
Same meaning.
Different words.
Rewrite text while preserving what it actually says.
Sentence-level restructuring.
Text is processed to generate and verify the rewrite.
The architecture
How it works
Rewriting is easy. Preserving meaning is the difficult part.
INPUT
Your original text
REWRITE
A language model reconstructs the wording and sentence structure.
VERIFY
A separate verification pass compares the rewrite against the original.
RETRY
If important meaning changed, the system automatically tries again.
OUTPUT
You receive the strongest rewrite that passed semantic verification.
Why not just ask an AI to paraphrase it?
A normal paraphraser has one objective: produce different wording.
That creates a problem. The model can accidentally remove qualifications, alter numbers, introduce claims, change certainty, or subtly change the author's argument.
NORMAL PARAPHRASER
OUR PIPELINE
Generation and verification are separate jobs.
Semantic verification
Different words. Same information.
ORIGINAL
The company reported revenue of $14.2 million in Q2, representing an 18% increase year over year.
REWRITE
Second-quarter revenue reached $14.2 million, up 18% compared with the same period last year.
Meaning
98%
Numbers
✓ $14.2M · 18%
Time period
✓ Q2
Relationship
✓ Year-over-year
Claims lost
0
Claims added
0
Similarity isn't measured by whether the sentences look alike. The verifier checks whether they communicate the same claims.
You control how different it becomes.
Light
Gentle wording changes.
The launch is planned for October.
Moderate
Sentence-level restructuring.
The team has planned an October launch.
Strong
Substantial structural rewriting.
October is when the team intends to launch.
Maximum
Maximum variation with meaning preserved.
The team’s launch target falls in October.
What we actually verify.
FACTS
Are factual claims preserved?
NUMBERS
Did any quantities change?
ENTITIES
Are names, companies and locations preserved?
CERTAINTY
Did “might” accidentally become “will”?
CAUSALITY
Are cause-and-effect relationships unchanged?
OMISSIONS
Was important information removed?
ADDITIONS
Did the model invent anything?
CONTRADICTIONS
Does the rewrite conflict with the original?
What rewriting can — and cannot — prove.
Rephrase verifies semantic preservation: whether the rewrite keeps the same factual claims, numbers, entities, certainty, and relationships.
That is different from proving anything about the origin of the text or removing a provider-specific watermark.
Rephrase makes no provenance claim beyond the checks it can actually perform.
WHAT WE VERIFY
- Semantic preservation
- Factual preservation
- Numbers and entities
- Missing information
- Added information
- Contradictions
WHAT REQUIRES AN EXTERNAL DETECTOR
Provider-specific watermark or AI-origin detection. Metadata removal also cannot prove that a textual signal is gone.
Your writing stays yours.
NO ACCOUNT REQUIRED
Start without creating a profile.
NO PERMANENT SERVER HISTORY
The server does not maintain submitted text history.
HISTORY STAYS IN YOUR BROWSER
Your saved rewrites stay in this browser.
Under the hood.
FAIL → RETRY
Up to three attempts with verifier feedback.
PASS → RETURN RESULT
Only a verified rewrite reaches the editor.
The quality gate prevents a rewrite from being accepted simply because it sounds good.
MODEL
Not configured
VERIFIER
Not configured
API
Operational