✍️ Vibe Writing is Coming — Write Faster and Better!

Litewrite: AI-Powered Collaborative LaTeX Writing Platform

News • Features • Screenshots • Architecture • AI Features • nanobot • Quick Start • Contributing

---

Try Litewrite: (https://litewrite.ai/)

📰 News

• 2026-02-10 🤖 **nanobot × Litewrite — nanobot empowers your vibe writing (setup guide)
• 2026-02-03 🎉 Litewrite v1.0.7 — Real-time collaboration, LaTeX compilation & AI chat.

---

✨ Features

🚀 TAP Smart Completion — Type a few words, AI continues writing for you. Just hit Tab and you're all set!

💬 ASK Mode — Summon your AI assistant anytime. Grammar issues? Format problems? Just ask!

🤖 Agent Mode — The ultimate hands-off writing experience! AI automatically edits, polishes, formats!

🔍 Deep Research — AI conducts in-depth research and auto-generates reports. Research while you write!

📱 nanobot — Chat with your projects from IM apps. Compile, edit, and manage — all from your phone!

---

💬 Chat with your projects from IM apps

---

📸 Screenshots

Click to expand screenshots

🏠 Landing Page

AI-integrated vibe writing — lite, fast, effortless

📚 Template Gallery

10+ pre-built templates: NeurIPS, ACM, Beamer, CV, Thesis, and more

🔣 Symbol Panel

Quick insert Greek letters, math operators, and LaTeX constructs

📖 Reference Search

Search and insert BibTeX citations with fuzzy matching

📜 Version History

Create snapshots, compare diffs, and restore any previous version

🔗 Project Sharing

Share via link or invite collaborators by email

💡 Try it yourself: Visit litewrite.ai for the live demo!

---

🏗️ Architecture

Litewrite is a microservices architecture with 6 core services communicating over a Docker network.

Service	Port	Technology	Role
🌐 Next.js	3000	TypeScript, React	Web app, API routes, auth, project management
🔄 WebSocket	1234	Node.js, Yjs	Real-time collaboration, document sync, chat rooms
🧠 AI Server	6612	Python, FastAPI	TAP completion, Chat Agent, Deep Research
📄 Compile	3002	Node.js, TeXLive	LaTeX compilation (pdfLaTeX/XeLaTeX/LuaLaTeX)
🤖 nanobot	—	Python, LiteLLM	Telegram/Feishu bot, Litewrite project operations
⚡ Redis	6379	Redis	Yjs persistence, rate limiting, caching

🔒 All internal service-to-service communication is authenticated via X-Internal-Secret header.

---

🧠 AI Features

Litewrite's AI capabilities are powered by a modular Agent system built on Python FastAPI. The architecture follows a MainAgent + SubAgent orchestration pattern with a unified Tool layer.

⚡ TAP Completion (Type-Ahead Prediction)

Ghost-text AI completion as you type, similar to GitHub Copilot.

Stage	Description
Input	prefix (text before cursor) + suffix (text after cursor)
Processing	Boundary extraction → Scenario detection → LLM inference
Output	Action type (insert / complete_word / fix / skip) + suggested text
Post-processing	Boundary fixups → Diff computation → Frontend rendering

🤖 Agent System (Ask & Agent Mode)

The Agent system uses a ReAct-style loop with tool calling for complex writing tasks.

Mode-based Tool Availability:

Tool	Ask Mode	Agent Mode	Description
read_file	✅	✅	Read file content with line ranges
list_files	✅	✅	List project files and structure
web_search	✅	✅	Search the web for information
arxiv_search	✅	✅	Search arXiv papers with RAG
plan	✅	✅	Generate step-by-step plan
done	✅	✅	Signal task completion
edit_file	—	✅	Edit file with diff preview
task	—	✅	Delegate to SubAgent

Context Management:

• 📦 Session history limit: 64K tokens (configurable)
• 📦 Execution context limit: 64K tokens (configurable)
• 🔄 Automatic LLM-based compression when threshold (95%) is exceeded
• 🛡️ Shadow documents for safe edit preview (or direct-apply for trusted callers)

🔍 Deep Research

Multi-source research with automatic report generation.

🔧 Unified Tool Layer

All tools inherit from a base Tool class with standardized interfaces:

class Tool(ABC):
    name: str              # Tool identifier
    description: str       # LLM-readable description
    parameters: dict       # JSON Schema for parameters
    mode: ToolMode         # ask / agent / all

    async def execute(params, context) -> ToolResult

The ToolContext provides:

• 📋 Project and user information
• 📡 SSE event emitter for real-time frontend updates
• 🔓 direct_apply flag for trusted callers (bypasses shadow documents)
• 📊 Shared usage accumulator for token tracking

---

📱 nanobot — Chat Platform Integration

nanobot is an AI assistant that connects messaging platforms (Telegram, Feishu/Lark) to Litewrite, enabling users to manage LaTeX projects through natural language conversations.

📖 For detailed setup instructions, see the nanobot Deployment Guide.

🛠️ Key Capabilities

nanobot provides 20 Litewrite tools for comprehensive project management:

Category	Tools
🎯 Core	List projects, invoke Litewrite AI agent, compile to PDF
📁 File Operations	List, read, edit, create, rename, delete, upload files
📦 Project Management	Create, delete, rename projects
📜 Version Control	List, save, restore versions
📥 Import	Import from arXiv, GitHub/GitLab, or file upload

⚙️ How It Works

1. 💬 User sends a message in Telegram or Feishu
2. 🧠 nanobot's agent processes the message with LLM reasoning
3. 🔧 Agent decides which Litewrite tools to call
4. 🔐 Tools communicate with Litewrite via Internal API (X-Internal-Secret)
5. 📤 Results (text, PDF, etc.) are sent back to the chat platform

💾 Session Management

• 🔗 Per-project session cache: consecutive calls reuse the same conversation
• 👁️ Sessions appear in the Litewrite web UI as "nanobot" conversations
• 🤝 The litewrite_agent tool delegates complex tasks to Litewrite's built-in AI agent, combining nanobot's chat interface with Litewrite's full editing capabilities

---

🚀 Quick Start

Prerequisites

• Docker Desktop (includes Docker Compose)
• 8GB+ RAM recommended (TeXLive image is ~5GB)

Development Setup

# Clone the repository
git clone https://github.com/hkuds/litewrite.git
cd litewrite

# Start with one command
./scripts/up-dev.sh

This will:

1. Create .env from env.example.oss if missing
2. Pull/build all Docker images
3. Start the full stack (web, ws, ai-server, compile, minio, redis)

🔗 Access Points

Service	URL	Description
🌐 App	http://localhost:3000	Main application
📦 MinIO	http://localhost:9001	Object storage console (minioadmin/minioadmin)
🧠 AI Server	http://localhost:6612/health	AI services health check
📄 Compile	http://localhost:3002/health	LaTeX compile service

🏭 Production Deployment

# With custom env files
./scripts/up-prod.sh --env-file .env.production --env-file .env.secrets

📋 Required Environment Variables

Variable	Description
NEXTAUTH_SECRET	Random string for session encryption
INTERNAL_API_SECRET	Internal service authentication
DATABASE_URL	PostgreSQL connection string
S3_BUCKET	S3 bucket name
S3_REGION	S3 region
S3_ACCESS_KEY_ID	S3 access key
S3_SECRET_ACCESS_KEY	S3 secret key
OPENROUTER_API_KEY	AI service API key

See env.example.oss for the complete configuration reference.

---

🛠 Tech Stack

🎨 Frontend Framework: Next.js 14 (App Router) Language: TypeScript 5 Editor: CodeMirror 6 + y-codemirror.next UI: Tailwind CSS + Shadcn UI	⚙️ Backend API: Next.js API Routes ORM: Prisma Auth: NextAuth.js 5 Collaboration: Yjs + y-websocket
🧠 AI Services AI Server: Python FastAPI LLM: OpenRouter (multi-model) nanobot: Python + LiteLLM Channels: Telegram, Feishu/Lark	🏗️ Infrastructure Database: SQLite (dev) / PostgreSQL (prod) Cache: Redis Storage: S3-compatible (MinIO/AWS) LaTeX: TeXLive (pdfLaTeX, XeLaTeX, LuaLaTeX) Container: Docker + Docker Compose

---

✏️ Editor Experience

Feature	Description
📝 LaTeX Editor	Built on CodeMirror 6 with syntax highlighting, smart completion, bracket matching, and folding
📄 Markdown Editor	Full GFM support with code highlighting, math rendering (KaTeX), and live preview
👁️ Visual Editing	Typora-like WYSIWYG: renders content while editing, shows source at cursor position
📂 Multi-file Projects	Organize large documents with \input{}, \include{}, and proper file tree management

🤖 AI-Powered Writing

Feature	Description
⚡ TAP Completion	Ghost-text AI completion as you type (Cursor/Copilot-like experience)
💬 Ask AI	Select text and ask AI to explain, improve, translate, or rewrite
🤖 Agent Mode	Full autonomous editing — AI reads, plans, and applies multi-file edits
🔍 Deep Research	Multi-iteration research: web + arXiv search, structured report generation with BibTeX
📊 AI Table	Generate LaTeX tables from natural language descriptions
📱 nanobot	Manage projects from Telegram/Feishu: edit files, compile PDFs, import papers

👥 Collaboration & Sharing

Feature	Description
🔄 Real-time Collaboration	Multiple users editing simultaneously with cursor presence indicators
📜 Version History	Create snapshots, compare diffs, and restore any previous version
🔗 Sharing	Share projects via link with granular view/edit permissions

📄 Compilation & Preview

Feature	Description
🔧 Multiple Engines	pdfLaTeX, XeLaTeX, LuaLaTeX with full TeXLive distribution
👁️ Live Preview	Auto-compile on save with instant PDF preview
🔀 SyncTeX	Bi-directional navigation: click PDF to jump to source, click source to highlight PDF
📦 Export Options	Download PDF, source ZIP, or individual files

---

📁 Project Structure

litewrite/
├── app/                      # Next.js App Router pages & API routes
│   ├── api/                  # REST API + Internal API endpoints
│   ├── editor/               # Editor page
│   └── (home)/               # Dashboard pages
├── components/               # React components
│   ├── editor/               # Editor-specific components
│   ├── pdf-viewer/           # PDF preview components
│   └── ui/                   # Shadcn UI components
├── lib/                      # Shared utilities
├── prisma/                   # Database schema & migrations
│
├── server/                   # WebSocket collaboration server
│   ├── ws-server.ts          # Yjs WebSocket + HTTP endpoints
│   └── yjs-persistence.ts    # Redis-based Yjs persistence
│
├── ai-server/                # AI service (Python FastAPI)
│   ├── api/                  # Chat, TAP, Deep Research endpoints
│   ├── services/             # Service implementations
│   │   ├── chat_1_5/         # Agent system (MainAgent + SubAgents)
│   │   ├── tap/              # TAP completion service
│   │   └── deep_research/    # Multi-iteration research
│   ├── tools/                # Agent tools (read, edit, search, etc.)
│   └── core/                 # Config, LLM client, caching
│
├── compile-server/           # LaTeX compilation service
│
├── nanobot/                  # Chat platform AI assistant
│   └── nanobot/
│       ├── agent/            # LLM agent loop + 20 Litewrite tools
│       ├── channels/         # Telegram, Feishu, WhatsApp adapters
│       ├── bus/              # Async message bus
│       ├── config/           # Environment-based configuration
│       └── skills/           # Extensible skill system
│
├── docker-compose.yml        # Development configuration
├── docker-compose.prod.yml   # Production configuration
└── scripts/                  # Setup & deployment helpers

---

🔧 Development

Common Commands

# Start development environment
docker compose up

# Start in background
docker compose up -d

# View logs
docker compose logs -f web
docker compose logs -f ai-server
docker compose logs -f nanobot

# Rebuild specific service
docker compose up --build web

# Reset database
docker compose exec web npx prisma migrate reset

# Run TypeScript check
npm run typecheck

🔥 Hot Reload

Directory	Service	Auto-reload
app/, components/, lib/	Next.js	✅
server/	WebSocket	✅
ai-server/	AI Server	✅
nanobot/	nanobot	✅
compile-server/	Compile	❌ (rebuild needed)

---

⚙️ Environment Variables

📋 Click to expand full list

Core

Variable	Description	Default
NEXTAUTH_SECRET	NextAuth encryption secret	(required)
INTERNAL_API_SECRET	Internal service auth	(required)
DATABASE_URL	Database connection	file:./dev.db

AI Providers

Variable	Description	Default
OPENROUTER_API_KEY	LLM API key (required for AI features)	—
OPENROUTER_API_KEYS	Multiple keys for rotation	—
OPENROUTER_KEY_STRATEGY	Key rotation: round_robin, random, least_used	round_robin
LLM_MODEL	Default LLM model	google/gemini-3-flash-preview
EMBEDDING_API_BASE	Embedding API URL	https://openrouter.ai/api/v1
EMBEDDING_API_KEY	Embedding API key	—
EMBEDDING_MODEL	Embedding model	text-embedding-3-small
SERPER_API_KEY	Web search (for Deep Research)	—

nanobot

Variable	Description	Default
TELEGRAM_ENABLED	Enable Telegram bot	false
TELEGRAM_BOT_TOKEN	Telegram bot token from @BotFather	—
FEISHU_ENABLED	Enable Feishu bot	false
FEISHU_APP_ID	Feishu app ID	—
FEISHU_APP_SECRET	Feishu app secret	—
NANOBOT_DEFAULT_LITEWRITE_USER_ID	Litewrite user UUID for project ownership	—

Storage

Variable	Description	Default
STORAGE_PROVIDER	s3 or local	s3
S3_ENDPOINT	S3 endpoint URL	http://localhost:9000
S3_BUCKET	S3 bucket name	litewrite
S3_REGION	S3 region	us-east-1
S3_ACCESS_KEY_ID	S3 access key	—
S3_SECRET_ACCESS_KEY	S3 secret key	—

Optional

Variable	Description	Default
REDIS_URL	Redis URL (enables Yjs persistence + rate limiting)	—
NEXT_PUBLIC_WS_URL	Public WebSocket URL for clients	ws://localhost:1234
TAP_MODEL	Override model for TAP completion	(uses LLM_MODEL)

📄 See env.example.oss for the complete template.

---

🤝 Contributing

We welcome contributions! Please see our Contributing Guide for details.

# Quick contribution workflow
git clone https://github.com/hkuds/litewrite.git
cd litewrite
git checkout -b feature/your-feature
# Make your changes
git commit -m "feat: add amazing feature"
git push origin feature/your-feature
# Open a Pull Request

Contributors

---

🙏 Acknowledgments

Litewrite is built on the shoulders of giants:

• Next.js — React framework
• CodeMirror — Code editor
• Yjs — CRDT for collaboration
• Prisma — Database ORM
• TeXLive — TeX distribution
• Shadcn UI — UI components
• OpenRouter — LLM routing
• LiteLLM — LLM provider abstraction
• nanobot — Ultra-lightweight AI assistant

---

⭐ Star History

Thanks for visiting ✨ litewrite!

⭐ Star us on GitHub!