Installation and quick start

WriteLikeMe is published on npm as @ai-clyde/writelikeme. It has no runtime dependencies, runs fully offline, and needs Node.js 14 or newer (18+ recommended).

Install

npm

npm install @ai-clyde/writelikeme

pnpm

pnpm add @ai-clyde/writelikeme

yarn

yarn add @ai-clyde/writelikeme

bun

bun add @ai-clyde/writelikeme

Other ways to get it:

npx -y @ai-clyde/writelikeme --help           # run once, no install
npm install -g @ai-clyde/writelikeme          # global `writelike-me` command
npm install --save-dev @ai-clyde/writelikeme  # pin it for a project's agents/scripts

Check that it works:

npx writelike-me --version

Update with npm update @ai-clyde/writelikeme (or npm i -g @ai-clyde/writelikeme@latest). Remove with npm uninstall @ai-clyde/writelikeme.

Using an AI coding agent?

See AI harnesses for one-line MCP setup in Claude Code, Cursor, Codex, Qwen Code, Gemini CLI, VS Code, Windsurf and OpenCode, plus prompts to paste in.

Create your first SVG style code

From the command line:

npx writelike-me generate -f my-writing.md -o style/me.svg
npx writelike-me brief --style-code style/me.svg

generate -o writes the SVG (creating folders as needed). brief prints a short, plain-language description of the style that you can paste into any AI chat.

From Node.js:

const fs = require('fs');
const WriteLikeMe = require('@ai-clyde/writelikeme');

const analysis = WriteLikeMe.analyzeText(fs.readFileSync('my-writing.md', 'utf8'));
fs.writeFileSync('writelikeme-code.svg', WriteLikeMe.generateStyleSVG(analysis));
console.log(WriteLikeMe.describeStyle(analysis));

Or in the browser: use the demo on the home page, which can also load an existing SVG and copy its style brief.

Command line

writelike-me analyze  [text] [-f file]... [--brief]
writelike-me generate [text] [-f file]... [-o out.svg] [--width N] [--height N] [--no-words] [--context email]
writelike-me brief    --style-code me.svg
writelike-me inspect  me.svg [--brief]
writelike-me write    --style-code me.svg "Write a short paragraph about community gardens."
writelike-me check  --style-code me.svg [draft text | -f draft.md | -] [--min 0.8] [--json]
writelike-me compare  "reference text" "candidate text"
writelike-me mcp

Command	What it does
analyze	Prints style scores as JSON. --brief prints only categories, classification and the brief.
generate	Creates the SVG. Without -o it prints the markup; without text it uses the neutral default profile. --no-words leaves frequent words and signature phrases out of the SVG (privacy).
brief	Prints the style brief from an SVG (--style-code) or from text.
inspect	Prints the profile stored in an SVG's metadata.
write	Deterministic, template-based draft. Not a language model.
check	Scores a draft against an SVG style code and lists the largest style gaps with fixes. Exits non-zero when below --min.
compare	Heuristic style similarity (0–1) and per-signal differences.
mcp	Starts the MCP server for AI coding agents.

Text can come from positional words, -f/--file <path> (repeatable; a directory reads its .md, .txt and .markdown files non-recursively, sorted), or stdin with - (cat post.md | writelike-me analyze - --brief). generate --no-words leaves frequent words and signature phrases out of the SVG. In a checkout of this repository, use npm run analyze -- "text", npm run generate -- -o me.svg "text", and so on.

Imports

// CommonJS
const WriteLikeMe = require('@ai-clyde/writelikeme');

// ES modules (Node.js imports the CommonJS default export)
import WriteLikeMe from '@ai-clyde/writelikeme';

// Agent Handler (also available as WriteLikeMe.Handler / WriteLikeMe.createHandler)
const { createHandler } = require('@ai-clyde/writelikeme/handler.js');

index.js is also a UMD script. In a plain web page, load it with <script src="https://unpkg.com/@ai-clyde/writelikeme/index.js"></script> and use window.WriteLikeMe. This is how the demo on this site runs.

Schema v3: what's inside the SVG

The SVG <metadata> carries a complete, versioned JSON profile:

• Seven categories: formality, complexity, sentence_length, vocabulary, tone, creativity, punctuation
• Voice cues: pronouns, contractions, sentence variation, paragraph length, question/exclamation rates, transitions, hedging
• Habit signals: dialogue, all-caps, numbers, URLs, emoji, passive voice, punctuation rhythm, quote style
• Grammar patterns: comma splices, semicolon use, run-on sentences
• Sentence structure: simple / compound / complex rates
• Readability: Flesch-Kincaid grade, reading ease, grade label
• Style classification: formality level, dominant mood, voice, pacing
• Domain: business / technical / academic / creative / legal / medical / general
• Frequent words: up to five { word, count } entries
• Signature phrases: up to five { phrase, count } entries — repeated two-word sentence openers (optional, empty when none repeat)
• Context: optional context string (email / blog / docs ...) stored in the SVG metadata and used to name per-context style codes (style/<context>.svg)

v1 and v2 style codes are upgraded to v3 automatically when read.

The SVG contains the profile and frequent words, not the original sample. Still, frequent words can reveal personal details, so inspect the file (writelike-me inspect me.svg) before sharing it. To share style signals without the words, generate with generate --no-words (CLI), include_words: false (MCP generate_style_svg), or { includeWords: false } (generateStyleSVG / createStyleProfile in code); that empties frequentWords and signaturePhrases in the embedded profile.

Limitations

The analyzer uses dictionary and text-shape heuristics. It does not call an AI model or evaluate meaning, and results on short or non-English text are rough. Punctuation balance is not a grammar-correctness score. writeLikeMe() produces simple template drafts; for real writing, give the style brief to an AI model.