AI-powered document editing that learns your writing style
Why Styler? • Quick Start • How It Works • Features • Whitepaper • Technical Blog
Most AI writing tools have a problem: they make everyone sound the same.
When you ask ChatGPT to "improve" your writing, it applies its default preferences—certain hedging patterns, word choices, and formality levels. The result is polished but generic. It doesn't sound like you.
Styler is different. It learns your personal writing style and preserves it while improving clarity and flow.
| Generic AI Tools | Styler |
|---|---|
| One-size-fits-all edits | Learns your preferences |
| Strips your voice | Preserves your voice |
| No memory between sessions | Continuous learning |
| Same output for everyone | Adapts to you |
Download the latest release for your platform:
| Platform | Download |
|---|---|
| macOS | Styler-1.2.12.dmg |
| Windows | Coming soon |
| Linux | Coming soon |
- Download and open the DMG
- Drag Styler to Applications
- Remove quarantine (required for unsigned apps):
sudo xattr -rd com.apple.quarantine /Applications/Styler.app
- Launch Styler
- Configure your API key in Settings
Note: You'll need an API key from Anthropic or OpenAI to use Styler.
For developers who want to run from source or contribute:
# Clone and install
git clone https://github.com/p-koo/styler.git
cd styler && npm install
# Configure API key (choose one method)
# Option A: Via Settings UI (recommended)
npm run dev
# Open http://localhost:3000/settings and add your API key
# Option B: Via environment file
cp .env.example .env.local
# Edit .env.local with your key:
# ANTHROPIC_API_KEY=sk-ant-... (recommended)
# or OPENAI_API_KEY=sk-...
# Run development server
npm run dev
# Open http://localhost:3000# Build for macOS
npm run electron:build:mac
# Build for Windows
npm run electron:build:win
# Build for Linux
npm run electron:build:linux
# Output will be in dist/- macOS 10.15+ (Catalina or later)
- API Key from Anthropic or OpenAI (see below)
- Node.js 18+ — Download
- npm or yarn — Comes with Node.js
| Provider | Get API Key | Notes |
|---|---|---|
| Anthropic (recommended) | console.anthropic.com | Claude models, best quality |
| OpenAI | platform.openai.com | GPT-4 models |
| Ollama | ollama.ai | Free, runs locally |
| Package | Version | Purpose |
|---|---|---|
| Next.js | 16.x | React framework |
| React | 19.x | UI library |
| TypeScript | 5.x | Type safety |
| Tailwind CSS | 4.x | Styling |
| CodeMirror | 6.x | Text editor |
| Anthropic SDK | 0.71.x | Claude API |
| OpenAI SDK | 6.x | GPT API |
Full dependency list
{
"@anthropic-ai/sdk": "^0.71.2",
"@codemirror/commands": "^6.10.1",
"@codemirror/lang-markdown": "^6.5.0",
"@codemirror/language": "^6.12.1",
"@codemirror/language-data": "^6.5.2",
"@codemirror/legacy-modes": "^6.5.2",
"@codemirror/state": "^6.5.4",
"@codemirror/theme-one-dark": "^6.1.3",
"@codemirror/view": "^6.39.11",
"next": "^16.1.3",
"openai": "^6.16.0",
"react": "^19.2.3",
"react-dom": "^19.2.3",
"tailwindcss": "^4.1.18",
"typescript": "^5.9.3"
}Styler is powered by ADAPT (Adaptive Document Alignment via Prompt Transformations)—a multi-agent system that coordinates specialized AI agents.
You select text → Intent analysis → Generate edit → Critique → Refine → Present
↑ │
└──────────┘
(if score < 0.8)
- Intent Agent analyzes what the paragraph is trying to accomplish
- Prompt Agent builds a context-aware prompt with your style preferences
- LLM generates an edit aligned with your voice
- Critique Agent scores alignment (0-1) and identifies issues
- If score < 0.8, the system refines automatically (up to 3 times)
- You review the diff, toggle individual changes, and accept or reject
- Learning Agent updates your preferences based on your decision
| Layer | Scope | What It Controls |
|---|---|---|
| Base Style | Global | Verbosity, formality, hedging, format rules |
| Audience Profiles | Switchable | Context-specific overlays (academic, blog, business) |
| Document Adjustments | Per-document | Fine-tuned sliders, learned rules, document goals |
| Agent | Purpose |
|---|---|
| Orchestrator | Coordinates the edit-critique-refine loop |
| Intent | Analyzes document goals and paragraph purpose |
| Prompt | Builds style-aware, context-rich prompts |
| Critique | Fast edit evaluation during the edit loop (user waiting) |
| Learning | Thorough preference learning after decisions (user not waiting) |
Deep Dive: See the Whitepaper for full architecture details or the Technical Blog for a developer-focused walkthrough.
Styler Edit — Select a paragraph, add an optional instruction ("make concise"), get a style-matched suggestion.
Vibe Edit — Apply vibes (Polish, Concise, Formal, Engaging, Academic) to multiple paragraphs or the whole document.
- Word-level diffs — Toggle individual changes before accepting
- Iterative refinement — Not satisfied? Add feedback, click Refine, get a better edit
- Quick feedback chips — One-click feedback: "Too clunky", "Lost intent", "Too many edits"
- Cell controls — Move cells up/down, delete with toolbar buttons (Colab-style)
- Prettify — AI-powered cleanup for PDF imports: merge fragments, remove artifacts, fix formatting
- General Chat — Ask questions about writing, get advice based on your document profile
- Document Chat — Get feedback on selected cells, analyze alignment with your style
- Alignment Score — See how well your content matches your configured preferences
| Control | Range | Effect |
|---|---|---|
| Verbosity | Terse ↔ Detailed | Compression vs. expansion |
| Formality | Casual ↔ Formal | Contractions, register, tone |
| Hedging | Confident ↔ Cautious | Qualifiers and uncertainty |
- Auto-synthesized goals — System infers document objectives, guides all edits
- Paragraph intent — Understands each paragraph's role in the document
- Section awareness — Edits respect document structure
- LaTeX — Preserves commands, environments, math mode
- Markdown — Maintains headers, lists, code blocks
- Smart splitting — Syntax-aware document segmentation
| Shortcut | Action |
|---|---|
Delete / Backspace |
Delete selected cells |
Enter |
Edit selected cell |
Escape |
Exit edit mode / clear selection |
↑ / ↓ |
Navigate between cells |
Shift + ↑/↓ |
Extend selection |
Cmd/Ctrl + A |
Select all cells |
Cmd/Ctrl + C |
Copy selected cells |
Cmd/Ctrl + X |
Cut selected cells |
Cmd/Ctrl + V |
Paste cells |
Cmd/Ctrl + Z |
Undo |
Cmd/Ctrl + Shift + Z |
Redo |
- Audience profiles — Switch contexts instantly (journal → blog → grant)
- Version history — Full undo/redo, compare any versions
- Auto-save — Never lose work
- Dark mode — System, light, or dark theme
- Local storage — All data stays on your machine
Styler learns from your feedback without memorizing specific word choices:
| Signal | How It's Used |
|---|---|
| Rejection feedback | "Too formal" → decrease formality preference |
| Partial accepts | Toggled-off changes become avoid patterns |
| Decision history | Patterns extracted across multiple edits |
Conservative by design: Word avoidance rules require 5+ consistent rejections before activating. This prevents over-fitting to single decisions.
Academic Writing
- Research papers with consistent voice across sections
- Grant proposals with appropriate hedging for reviewers
- Thesis chapters that maintain your style
Professional Documents
- Technical docs with standardized terminology
- Reports matching organizational style guides
- Proposals calibrated for different audiences
Content Creation
- Blog posts that sound like you
- Documentation with consistent technical writing
- Batch editing across multiple documents
src/
├── agents/ # ADAPT multi-agent system
│ ├── orchestrator-agent.ts # Main coordination loop
│ ├── intent-agent.ts # Document goals & paragraph intent
│ ├── critique-agent.ts # Fast edit evaluation (latency-critical)
│ ├── learning-agent.ts # Preference learning from feedback
│ ├── prompt-agent.ts # Style-aware prompt building
│ └── constraint-extraction-agent.ts
├── app/
│ ├── editor/ # Main editor page
│ ├── settings/ # Global configuration
│ └── api/ # API routes
├── components/
│ ├── DiffView.tsx # Interactive diff display
│ ├── DocumentProfilePanel.tsx # Per-document preferences
│ ├── ChatPanel.tsx # Chat assistant sidebar
│ └── ...
├── memory/ # Preference storage
│ ├── preference-store.ts # Global preferences
│ └── document-preferences.ts # Document-specific learning
└── providers/ # LLM integrations
├── anthropic.ts
├── openai.ts
└── ollama.ts
| Category | Technologies |
|---|---|
| Desktop | Electron 40 |
| Frontend | Next.js 16, React 19, TypeScript, Tailwind CSS |
| Editor | CodeMirror 6 with syntax highlighting |
| AI | Anthropic Claude, OpenAI GPT-4, Ollama |
| Storage | Local JSON files (no cloud dependency) |
- Whitepaper — Comprehensive system documentation
- Technical Blog — Developer-focused architecture walkthrough
Contributions welcome! Please open an issue to discuss changes before submitting a PR.
MIT — see LICENSE
Peter Koo — koo@cshl.edu
Built with Claude Code by Anthropic