Output
Free tool
TextClean
Paste any text and get it cleaned, inspected and scored — hidden characters, AI-style signals, personal data, readability, originality against your own sources, and deterministic rewrites. Everything runs in your browser; nothing is uploaded.
—
Text trust snapshot
Paste text to begin
Scores are local heuristics, not proof of authorship or factual truth.
Working document
Source text
Live diagnostics
What changed
Important: AI-authorship signals from text alone are probabilistic heuristics. They can be wrong, especially after editing, translation, short text, or model updates. Treat these as signals, never proof.
AI-style analysis
—/100
Not enough text
Authorship continuity
No analysis yet
Paragraph map
AI-style by segment
Artifact scanner
Prompt & AI residue
Neural model check
On-device AI detector
Optional second opinion from a neural detector (RoBERTa fine-tuned on the RAID benchmark, MIT-licensed) running entirely in your browser. The first run downloads ~130 MB of model weights from a CDN and caches them — your text still never leaves this page.
Stylometry
Compare with a known author sample
Originality
Local plagiarism comparison
Document comparison
Version A vs B
Citation audit
References & citation integrity
Claim inventory
Statements to fact-check
Offline mode can identify check-worthy claims, but cannot verify them against the web. Export the list for a fact-checking workflow.
Local provenance
File metadata
Writing quality
—/100
Paste text to analyze
Readability
—
Privacy risk
—
Sensitive content
PII & secret scanner
Style
Tone fingerprint
Quality signals
Clarity & repetition
Persuasion
Manipulation & loaded-language signals
Local transformations
Rewrite without an LLM
These modes use transparent deterministic edits. They improve hygiene and style, but do not pretend to be a full generative rewrite.
Preview
Natural cleanup
Transparent edits
What the rewrite changed
What it does
Five tools in one page
Clean
Strip zero-width and control characters, normalize smart quotes, dashes and Unicode look-alikes, remove markdown residue. See exactly what changed in a diff.
Detect
Heuristic AI-style scoring per paragraph, an optional on-device neural detector (RoBERTa, RAID-trained), prompt/AI residue scanner, and stylometric comparison with a known writing sample. Signals, not proof.
Verify
Local plagiarism check against reference texts you supply, version A/B comparison, citation audit and a list of claims worth fact-checking.
Analyze
Readability and writing-quality scores, tone fingerprint, repetition and filler detection, PII & secret scanner (emails, phones, IBANs, API keys…).
Rewrite
Transparent, rule-based rewrites — natural, concise, professional, simplified or anonymized — with every edit listed. No LLM involved.
Private by design
No server, no account, no tracking of your text. Close the tab and it is gone. Export a JSON report if you want to keep the analysis.
Questions
Is my text sent anywhere?
No. All processing happens in your browser; there is no API call with your text, no analytics on it, no storage other than your rule preferences in localStorage. The optional neural detector downloads model files (~130 MB, once) from a CDN — that download contains no information about your text, which still never leaves the page.
Can it prove a text was written by AI?
No tool can. The Detect tab combines stylistic heuristics with an optional neural detector trained on the RAID benchmark. Both report probabilities, and both are known to misfire — especially on edited, translated, short, or formal text. Treat scores as a hint to look closer, never as evidence.
What does the plagiarism check compare against?
Only the reference texts you paste or open on the Verify tab. It does not search the web. That keeps it private — and means it will not find sources you did not provide.
Why is the "rewrite" so conservative?
It is deliberately rule-based: filler removal, contraction handling, quote normalization, anonymization of names, emails and numbers. You can see every change, and it never invents content.
Can I use it on confidential documents?
Yes — that is the point. Nothing leaves your machine. For files, only text formats are read (txt, md, csv, json, html); PDF and DOCX are not parsed in this version.
Need this at scale, or with a real search index / LLM behind it?
That is what we build for clients: document pipelines, RAG assistants over your own knowledge base, PII redaction at volume. Talk to us.
Need the production version?
Web-scale originality search, real fact-checking, PDF/DOCX parsing and LLM-powered rewriting need servers and models. We build those — on your data, under your control.