OpenPrism

OpenPrism — Vibe Writing for Academia

中文 | English

---

✨ Highlights

🤖 AI Assistant	✍️ Compile & Preview	📚 Templates
Chat / Agent history Tools multi-step edits	TexLive / Tectonic / Auto PDF preview & download	ACL / CVPR / NeurIPS / ICML One-click conversion

🔄 Template Transfer
Legacy (LaTeX→LaTeX) / MinerU (PDF→MD→LaTeX) LLM-powered migration + auto compile fix + VLM layout check

🔧 Advanced Editing	🗂️ Project Management	⚙️ Configuration
AI autocomplete / Diff / diagnose	Multi-project + file tree + upload	OpenAI-compatible endpoint Local-first privacy

🔍 Search	📊 Charting	🧠 Recognition
WebSearch / PaperSearch	Chart from tables	Formula/Chart recognition

👥 Collaboration	📝 Peer Review
Multi-user real-time editing Cursor sync & online management	AI Review Report / Consistency Missing Citations / Compile Summary

---

📢 News

Warning

🚧 Template Transfer is under testing
The Template Transfer feature is currently in beta and may contain known or unknown bugs. If you encounter any issues, please report them via Issues.

Tip

🆕 2025-02 · Template Transfer (Dual Mode)
Two transfer modes are now available: Legacy mode (LaTeX→LaTeX direct migration) and MinerU mode (PDF→Markdown→LaTeX via MinerU API). Both modes feature LLM-powered content migration, automatic compile error fixing, and optional VLM-based layout checking.

Tip

🆕 2025-02 · Real-time Collaboration
Multi-user simultaneous editing is now available, powered by CRDT with automatic conflict resolution and cursor sync. Current version requires a server with a public IP; invite remote collaborators via token-based links.

---

_{✨ Home preview: three-panel workspace + editor + preview}

---

✨ Core Features

OpenPrism is a local-first LaTeX + AI workspace for academic writing, optimized for fast editing, controlled changes, and privacy.

🤖 AI Assistant

• Chat mode: read-only Q&A
• Agent mode: generate diffs for confirmation
• Tools mode: multi-step tools + multi-file edits
• Tasks: polish, rewrite, restructure, translate, custom
• Autocomplete: Option/Alt + / or Cmd/Ctrl + Space, Tab to accept

✍️ Compile & Preview

• Engines: TexLive / Tectonic / Auto fallback
• Preview toolbar: zoom, fit width, 100%, download PDF
• Compile log: error parsing + one-click diagnose + jump to error
• Views: PDF / Figures / Diff

📚 Template System

• Built-ins: ACL / CVPR / NeurIPS / ICML
• Conversion: one-click template switch with content preserved

🔄 Template Transfer

• Dual mode: Legacy (LaTeX→LaTeX) and MinerU (PDF→Markdown→LaTeX)
• MinerU integration: parse PDF via MinerU API, extract Markdown + images, then fill into target template
• LLM-powered migration: AI analyzes source/target structure, drafts transfer plan, and applies content mapping
• Auto compile fix: automatically detect and fix LaTeX compilation errors with retry loop
• VLM layout check: optional visual layout validation using VLM to detect overflow, overlap, and spacing issues
• Asset handling: automatic copy of images, bib files, and style files from source to target

🗂️ Project Management

• Projects panel: manage multiple projects
• File tree: create/rename/delete/upload/drag
• BibTeX: quick create references.bib

⚙️ Configuration

• LLM Endpoint: OpenAI-compatible, supports custom base_url
• Local storage: settings saved to browser localStorage
• TexLive config: customizable TexLive resources
• Language switch: toggle 中文/English in the top bar

🔍 Search & Reading

• WebSearch: online search with summaries
• PaperSearch: academic paper search with citation info

📊 Charts & Recognition

• Table-to-chart: generate charts directly from tables
• Smart recognition: formulas and charts auto-detected

📝 Peer Review

• AI Quality Check: automated paper quality assessment
• Full Review Report: generate detailed reviewer-style review comments
• Consistency Check: terminology and symbol consistency detection
• Missing Citations: find statements that need citations
• Compile Log Summary: summarize compile errors and fix suggestions

👥 Real-time Collaboration

• Multi-user editing: multiple users edit the same document simultaneously with real-time sync
• Cursor & selection sync: each user's cursor displayed in a distinct color, visible in real time
• Online user list: collaboration panel shows currently connected users and their status
• Invite to collaborate: invite others via link or token to join the editing session

---

🎨 Showcase

🖥️ Three-Panel Workspace

_{✨ AI Assistant | LaTeX Editor | PDF Preview}

✍️ Editor View

_{✨ Synchronized editing and preview}

🤖 Agent Mode

_{✨ Generate diff suggestions for review}

🧪 One-Click Diagnose

_{✨ Parse compile errors and jump to locations}

🌐 WebSearch

_{✨ Online search with concise summaries}

📄 PaperSearch

_{✨ Academic search and citation info}

📊 Table-to-Chart

_{✨ Turn tables into charts in one step}

🧠 Formula/Chart Recognition

_{✨ Recognize structures for editable outputs}

🔧 AI Autocomplete

_{✨ Option/Alt + / to trigger, Tab to accept}

📝 Peer Review

_{✨ AI Quality Check: Review Report / Consistency Check / Missing Citations / Compile Summary}

👥 Real-time Collaboration

_{✨ Multi-user real-time collaborative editing with cursor sync and online user management}

---

🚀 Quick Start

📋 Requirements

Basic Environment

• Node.js >= 18.0.0
• npm >= 9.0.0
• OS: Windows / macOS / Linux

LaTeX Compilation Environment (Required)

OpenPrism requires a LaTeX engine to generate PDFs. Choose one of the following options based on your OS:

Option 1: TexLive (Recommended)

• Linux (Ubuntu/Debian):

  sudo apt-get update
  sudo apt-get install texlive-full

• Linux (CentOS/RHEL):

  sudo yum install texlive texlive-*

• macOS:

  brew install --cask mactex

• Windows: Download TexLive installer

Option 2: Tectonic (Lightweight)

• Linux/macOS:

  curl --proto '=https' --tlsv1.2 -fsSL https://drop-sh.fullyjustified.net | sh

• Windows: Download Tectonic installer

Note: TexLive full installation is ~5-7GB, Tectonic is lighter but with fewer features. TexLive is recommended for Linux servers.

📦 Install & Run

Development Deployment

# 1. Clone repository
git clone https://github.com/OpenDCAI/OpenPrism.git
cd OpenPrism

# 2. Install dependencies
npm install

# 3. Start dev server (frontend + backend)
npm run dev

Access:

• Frontend: http://localhost:5173
• Backend: http://localhost:8787

Production Deployment

# 1. Build frontend and backend
npm run build

# 2. Start production server
npm start

Complete Linux Server Deployment Example

# 1. Install Node.js (Ubuntu example)
curl -fsSL https://deb.nodesource.com/setup_18.x | sudo -E bash -
sudo apt-get install -y nodejs

# 2. Install TexLive
sudo apt-get update
sudo apt-get install -y texlive-full

# 3. Verify installation
node --version  # Should show >= 18.0.0
pdflatex --version  # Should show TexLive version

# 4. Clone and deploy project
git clone https://github.com/OpenDCAI/OpenPrism.git
cd OpenPrism
npm install
npm run build

# 5. Configure environment variables (optional)
cat > .env << EOF
OPENPRISM_LLM_ENDPOINT=https://api.openai.com/v1/chat/completions
OPENPRISM_LLM_API_KEY=your-api-key
OPENPRISM_LLM_MODEL=gpt-4o-mini
OPENPRISM_DATA_DIR=/var/openprism/data
PORT=8787
EOF

# 6. Start service
npm start

# 7. Use PM2 for process management (recommended)
sudo npm install -g pm2
pm2 start npm --name "openprism" -- start
pm2 save
pm2 startup

---

⚙️ Configuration

Environment Variables

Create a .env file in the project root (optional):

# LLM Configuration
OPENPRISM_LLM_ENDPOINT=https://api.openai.com/v1/chat/completions
OPENPRISM_LLM_API_KEY=your-api-key
OPENPRISM_LLM_MODEL=gpt-4o-mini

# Data storage path
OPENPRISM_DATA_DIR=./data

# Backend service port
PORT=8787

# MinerU API Configuration (for PDF→MD→LaTeX transfer)
OPENPRISM_MINERU_API_BASE=https://mineru.net/api/v4
OPENPRISM_MINERU_TOKEN=your-mineru-token

LLM Configuration

OpenPrism supports any OpenAI-compatible endpoint, including custom base_url:

Method 1: Environment Variables

# .env file
OPENPRISM_LLM_ENDPOINT=https://api.openai.com/v1/chat/completions
OPENPRISM_LLM_API_KEY=sk-your-api-key
OPENPRISM_LLM_MODEL=gpt-4o-mini

Method 2: Frontend Settings Panel

• Click the "Settings" button in the frontend interface
• Fill in API Endpoint, API Key, and Model
• Configuration is automatically saved to browser localStorage

_{✨ LLM Configuration Settings Panel}

Supported Third-party Services:

• OpenAI: https://api.openai.com/v1
• Azure OpenAI: https://your-resource.openai.azure.com/openai/deployments/your-deployment
• Other compatible services: https://api.apiyi.com/v1

LaTeX Compilation Configuration

Supported Compilation Engines:

• pdflatex - Standard LaTeX engine
• xelatex - Supports Unicode and Chinese
• lualatex - Supports Lua scripting
• latexmk - Automated build tool
• tectonic - Modern lightweight engine

Configuration Method:

1. Select compilation engine in frontend "Settings" panel
2. Set to "Auto" for automatic fallback to available engines
3. Customize TexLive resource path

Data Storage Configuration

Default data storage is in ./data directory, can be modified via environment variable:

# Custom data directory
OPENPRISM_DATA_DIR=/var/openprism/data

Directory Structure:

data/
├── projects/           # User projects
│   ├── project-1/
│   │   ├── main.tex
│   │   └── references.bib
│   └── project-2/
└── templates/          # Template cache

---

👥 Collaboration Guide

OpenPrism includes a built-in real-time collaboration system based on CRDT (Yjs) + WebSocket, allowing multiple users to edit the same document simultaneously without any third-party service.

Collaboration Environment Variables

Add the following to your .env file:

# Token signing secret (must change for production)
OPENPRISM_COLLAB_TOKEN_SECRET=your-secure-random-string

# Require token for collaboration (default: true, set false for local dev)
OPENPRISM_COLLAB_REQUIRE_TOKEN=true

# Token TTL in seconds (default: 86400 = 24 hours)
OPENPRISM_COLLAB_TOKEN_TTL=86400

How to Use

1. Deploy: Deploy OpenPrism to a server with a public IP, configure a domain and HTTPS
2. Generate invite: Click "Generate Invite Link" in the collaboration panel on the editor page
3. Share link: Send the generated link to your collaborator
4. Join: Collaborator opens the link, token is verified automatically, and they enter the editor
5. Edit together: Multiple cursors visible in real time, edits sync automatically, conflicts resolved by CRDT

Nginx Reverse Proxy (Recommended, For Public Servers)

Collaboration requires WebSocket. Nginx must be configured with upgrade headers:

server {
    listen 443 ssl;
    server_name your-domain.com;

    location / {
        proxy_pass http://127.0.0.1:8787;
        proxy_http_version 1.1;
        proxy_set_header Upgrade $http_upgrade;
        proxy_set_header Connection "upgrade";
        proxy_set_header Host $host;
    }
}

Tip: Local access (127.0.0.1) bypasses token verification by default, suitable for local development.

No Public Server? Use Tunnel (ngrok)

You can collaborate remotely without a public server. OpenPrism has built-in tunnel support — one command exposes your local service to the internet.

Quick Start (ngrok, Recommended)

1. Sign up for a free ngrok account and get your authtoken
2. Run the following commands:

export NGROK_AUTHTOKEN=your_token_here

npm run tunnel:ngrok

3. On startup, the terminal prints a public URL. Share it with your collaborator:

  OpenPrism started at http://localhost:8787

  Tunnel active (ngrok):
  Public URL: https://xxxx.ngrok-free.app
  Share this URL to collaborate remotely!

4. Your collaborator opens the URL in their browser and starts editing in real-time

Other Tunnel Options

Option	Command	Notes
localtunnel	npm run tunnel	Zero-config, but may be unstable
Cloudflare Tunnel	npm run tunnel:cf	Requires cloudflared installed

Note: Tunnel is off by default. Regular npm start does not create a tunnel. You can also set it via env var: OPENPRISM_TUNNEL=ngrok npm start

---

🎯 Usage Guide (Quick)

1. Create Project: Create new project in Projects panel and select template
2. Write Paper: Edit LaTeX in Files tree
3. AI Edits: Switch to Agent / Tools, generate diff and confirm
4. Compile & Preview: Click "Compile PDF", preview on right side
5. Export PDF: Click "Download PDF" in preview toolbar

---

📁 Project Structure

OpenPrism/
├── apps/
│   ├── frontend/           # React + Vite frontend
│   │   ├── src/
│   │   │   ├── app/App.tsx    # Main application logic
│   │   │   ├── app/TransferPanel.tsx  # Template transfer UI
│   │   │   ├── api/client.ts  # API calls
│   │   │   └── latex/         # TexLive integration
│   └── backend/            # Fastify backend
│       └── src/
│           ├── index.js       # API / compile / LLM proxy
│           ├── routes/transfer.js  # Transfer API endpoints
│           └── services/
│               ├── mineruService.js        # MinerU API integration
│               └── transferAgent/          # LangGraph transfer workflows
│                   ├── graph.js            # Legacy transfer graph
│                   ├── graphMineru.js       # MinerU transfer graph
│                   ├── state.js            # Transfer state schema
│                   └── nodes/              # Workflow nodes
├── templates/              # LaTeX templates (ACL/CVPR/NeurIPS/ICML)
├── data/                   # Project storage directory (default)
└── README.md

---

🗺️ Roadmap

Feature	Status	Description
👥 Real-time Collaboration		Multi-user real-time editing with cursor sync and online user management (currently requires a server with public IP)
🌐 Serverless Collaboration		Local collaboration without a public server: ① built-in tunnel integration (ngrok / Cloudflare Tunnel) to expose local services in one click; ② WebRTC-based P2P direct connection without third-party relay
🔍 Enhanced WebSearch		Integrate third-party Search APIs (e.g. Google / Baidu / SerpAPI) for improved search quality and coverage
📚 Template Transfer (Dual Mode)		Legacy (LaTeX→LaTeX) and MinerU (PDF→MD→LaTeX) dual-mode template transfer with LLM-powered migration, auto compile fix, and VLM layout check
📸 Version Snapshots & Rollback		Project version management with snapshot saving and one-click rollback
📖 Citation Search Assistant		Auto-search related papers and generate BibTeX citations

---

🤝 Contributing

Welcome to submit Issues or PRs:

1. Fork the repository
2. Create a new branch
3. Commit your changes
4. Submit a PR

Development commands:

npm run dev
npm run dev:frontend
npm run dev:backend
npm run build

---

📄 License

MIT License. See LICENSE.

---

🙏 Acknowledgments

• Tectonic
• CodeMirror
• PDF.js
• LangChain
• React / Fastify

---

If this project helps you, please give us a ⭐️ Star!

_{Scan to join the community WeChat group}

Made with ❤️ by OpenPrism Team