Word AI Assistant

Task Pane Add-in for Microsoft Word with a ChatGPT interface for document work.

Features

• Document Chat — ask questions about document content
• Rephrase — rewrite selected text while preserving meaning
• Compress — reduce text to 30-50% of original
• Expand — add clarifications and details to text
• Summarize — create structured document summary
• Generate Sections — create new sections from description

Requirements

• Node.js 18+
• Microsoft Word for Mac (Microsoft 365) or Word Online
• OpenAI API key

Installation

# Clone the repository
cd WordAI

# Install dependencies
npm install

# Create .env file in server folder
cp server/env.example server/.env

# Edit .env and add your OpenAI API key

Configuration (server/.env)

OPENAI_API_KEY=sk-your-api-key-here
OPENAI_API_BASE=https://api.openai.com/v1
LLM_MODEL=gpt-4o-mini
PORT=3001
MAX_CONTEXT_CHARS=8000

Running

# Start client and server simultaneously
npm run dev

# Or separately:
npm run dev:client  # Frontend at https://localhost:3000
npm run dev:server  # Backend at http://localhost:3001

On first client launch, vite-plugin-mkcert will automatically create and install an SSL certificate for localhost.

Sideload in Word for Mac

Method 1: Via ~/Library/Containers folder (recommended)

1. Close Word
2. Copy the manifest:

   mkdir -p ~/Library/Containers/com.microsoft.Word/Data/Documents/wef
   cp manifest/word-addin-manifest.xml ~/Library/Containers/com.microsoft.Word/Data/Documents/wef/

3. Open Word
4. Go to: Insert → Add-ins → My Add-ins
5. Find "AI Assistant" and activate it

Method 2: Via developer menu

1. Open Word
2. Go to: Insert → Add-ins → My Add-ins
3. Select the Shared Folder tab
4. Click Browse and select manifest/word-addin-manifest.xml

Method 3: Via Office Online

1. Open Word Online (office.com)
2. Create or open a document
3. Insert → Add-ins → Manage My Add-ins
4. Upload My Add-in → select word-addin-manifest.xml

Usage

1. Open a document in Word
2. Click the AI Assistant button on the Insert ribbon
3. Select the mode from the dropdown
4. Enter your request and click Send

Modes

Mode	Requires Selection	Description
Document Chat	No	Ask questions about document content
Rephrase	Yes	Rewrite selected text
Compress	Yes	Reduce text by 50-70%
Expand	Yes	Add details and clarifications
Summarize	No	Create structured summary
Generate Section	No	Create new section from description

Options

• Use selected text — take context from selection (otherwise from entire document)
• Replace selected text — automatically replace selection with result

Settings

Click ⚙️ in the panel to access settings:

• LLM Model — OpenAI model (gpt-4o-mini, gpt-4o, etc.)
• Max context characters — text limit for sending to LLM
• Legal style — use legal terminology in prompts

Project Structure

WordAI/
├── manifest/
│   └── word-addin-manifest.xml   # Office Add-in manifest
├── client/                        # React + Vite + TypeScript
│   ├── src/
│   │   ├── components/           # UI components
│   │   ├── hooks/                # React hooks
│   │   ├── office/               # Office.js wrappers
│   │   ├── api/                  # HTTP client
│   │   └── utils/                # Prompts and utilities
│   └── vite.config.ts
├── server/                        # Express + TypeScript
│   ├── src/
│   │   ├── index.ts              # HTTP server
│   │   └── llm.ts                # OpenAI integration
│   └── .env                      # Configuration (not in git)
├── config/
│   └── config.example.json       # Example configuration
└── package.json

Troubleshooting

Certificate not trusted

On first launch, the browser may show a certificate warning. Add an exception or:

# macOS - trust certificate
sudo security add-trusted-cert -d -r trustRoot -k /Library/Keychains/System.keychain ~/.vite-plugin-mkcert/cert.pem

Add-in not loading

1. Make sure dev server is running (npm run dev)
2. Open https://localhost:3000 in browser and accept the certificate
3. Restart Word
4. Check Word console for errors (Cmd+Option+I in Word Online)

API Error

1. Check OPENAI_API_KEY in server/.env
2. Check server logs in terminal
3. Make sure the model is available for your API key