Find the prose problems free tools ignore — echoes, crutch & filler words, reader-distancing filter words, adverb and weak-verb overuse, clichés, said-bookisms, repetitive sentence openers, and sentence-rhythm flatlines — all computed 100% in your browser. Your unpublished draft never touches a server.
▶ Try the live demo · no install, nothing uploaded
Good self-editing means catching what you can't see on a re-read: the word you used three times in two sentences, the just / really / suddenly crutches, the she saw / he felt / I noticed filter verbs that hold readers at arm's length, the wall of same-length sentences that flattens your pacing.
These are mechanical, detectable problems — but the tools that detect them well (ProWritingAid's "Echoes", Grammarly Premium) are paid, and the free options only do a fraction: word-frequency counters just rank words; Hemingway is general-purpose and knows nothing about fiction craft. Writers are also (rightly) wary of pasting an unpublished manuscript into a cloud service.
Inkspect fills that gap: the fiction-specific reports, free, and fully local.
| Report | What it finds |
|---|---|
| Echoes | The same word root repeated within a short window (e.g. walked…walking) — close repetition readers notice. Stemmed, stopword-aware, proper-noun-aware (won't flag character names), adjustable window. |
| Repeated phrases | Exact 3–5 word phrases reused across the manuscript. |
| Crutch & filler words | just, really, very, suddenly, kind of… with counts and density per 1,000 words. |
| Filter words | saw, heard, felt, knew, realized, noticed… that distance the reader. |
| Adverbs (-ly) | All -ly adverbs (minus a non-adverb allowlist), with extra emphasis on adverbs glued to dialogue tags (said quietly). |
| Weak verbs & passive | "To be" density plus a passive-voice heuristic. |
| Dialogue tags | Plain said/asked vs. ornate "said-bookisms" (exclaimed, hissed, retorted). |
| Clichés | A curated dictionary of worn-out stock phrases. |
| Sentence rhythm | Per-sentence length chart + flags for monotonous runs that flatten pacing. |
| Sentence openers | Flags runs of sentences that start with the same word ("She… She… She…") and shows opener frequencies. |
| Overused words | Document-wide frequency, grouped by root (name-aware). |
| Readability | Flesch Reading Ease, Flesch–Kincaid grade, word/sentence/paragraph/syllable stats, reading time, and an at-a-glance polish score. |
Plus: a readability dashboard with a dialogue-vs-narrative ratio and a
polish score; click any word in a report's breakdown to highlight just its
occurrences and step through them; click any finding to scroll to and
highlight it; the editor is live (edit and re-analyze); drag-and-drop
and .txt / .md / Word .docx import; light/dark themes; and a
one-click Markdown report export.
Inkspect makes zero network requests at runtime. Your manuscript is read,
analyzed, and highlighted entirely in your browser. Nothing is uploaded, logged,
or stored anywhere but your own machine (text is cached in your browser's
localStorage only, so a refresh doesn't lose your work). You can verify this in
your browser's Network tab — it stays empty.
git clone https://github.com/Skytuhua/Inkspect.git
cd Inkspect
npm install
npm run dev # http://localhost:5173- Download
inkspect-<version>-web.zipfrom the Releases page. - Unzip it and serve the folder over HTTP (a static site — opening
index.htmldirectly viafile://won't work because browsers block module workers there):
npx serve inkspect-1.1.0-web # or: python3 -m http.server- Open the printed URL. Click Load sample manuscript to see it in action, or open your own file.
- Bring text in: paste into the editor, drag-and-drop a file onto the
window, click Open file (
.txt,.md,.docx), or Sample to load the demo manuscript. - Read the dashboard: word/sentence stats, readability, the sentence-rhythm chart, and the polish score update automatically.
- Investigate a report: click it to expand. Its hits are highlighted in the text; use the ▲/▼ buttons to jump between them.
- Tune sensitivity: adjust the Echo window (how close counts as an echo) and Strictness (crutch/filter density threshold) in the toolbar.
- Export: Export report downloads a Markdown summary; Copy text copies the current manuscript.
| Command | Does |
|---|---|
npm run dev |
Start the Vite dev server. |
npm run build |
Type-check and build the static site into dist/. |
npm test |
Run the Vitest engine + importer suite. |
npm run lint |
ESLint (zero warnings allowed). |
npm run package |
Build the release zip + checksums into artifacts/. |
The analysis lives in src/engine/ as a pure, dependency-free TypeScript
library — analyze(text, options) → AnalysisResult. It tokenizes once
(carrying exact character offsets so every finding can be highlighted), segments
sentences/paragraphs, then runs each analyzer. The React UI runs the engine in a
Web Worker so even a full novel never freezes the page. See
ARCHITECTURE.md.
- English only in v1 (the craft word-lists and heuristics are English).
- It is not a grammar or spell checker, and does no AI rewriting — by design.
- Heuristics are approximate: the stemmer is light, the passive-voice and syllable detectors are rule-based. Treat reports as a guide, not a grader.
.doc(legacy) and.pdfimport are out of scope; export to.docx/.txt.
TypeScript · React 18 · Vite 5 · Vitest · a Web Worker · mammoth (for .docx) ·
hand-written CSS. All dependencies are MIT/Apache-2.0.
MIT © Inkspect contributors.