Paper PPT Agent

Upload a paper, AI generates your presentation

中文 | English

A multi-agent pipeline for automatically generating editable PowerPoint presentations from academic papers. Upload a PDF or TeX source, and the AI handles content extraction, structural planning, layout design, and visual quality assurance.

Table of Contents

• ✨ Features
• 📸 Demo
• ⚙️ Requirements
• 🚀 Quick Start
• 📋 Changelog
• 🙏 Acknowledgements
• 📄 License

✨ Features

📸 Demo

⚙️ Requirements

An API key for at least one model provider: OpenAI / Anthropic / Gemini / DeepSeek or a custom BaseURL-compatible API.

Formula typesetting uses local MathJax to produce editable vector shapes. The backend also requires Node.js and the npm dependencies in frontend; the launcher checks and installs them.

Optional: workbench Agent generation requires Claude Code or Codex to be installed and configured locally. Template-import Agent mode currently uses Claude Code and requires it to be installed and configured locally.

🚀 Quick Start

# Clone the repository
git clone https://github.com/CRui5in/paper-ppt-agent.git
cd paper-ppt-agent

# One-click start (auto-installs deps + launches frontend & backend)
# Windows
.\start-dev.bat
# Linux
sh start-dev.sh

After starting: Frontend http://127.0.0.1:5173 · Backend http://127.0.0.1:8000

# Install dependencies
uv sync --upgrade-package openai-codex --upgrade-package openai-codex-cli-bin --upgrade-package claude-agent-sdk
cd frontend && npm install && cd ..

# Backend
uv run python -m uvicorn backend.app:app --host 127.0.0.1 --port 8000 --reload --reload-dir backend

# Frontend
cd frontend && npm run dev -- --host 127.0.0.1 --port 5173 --strictPort

📋 Changelog

June 2026

• 📚 Multi-source Data Support — Use papers, links, and text materials together as generation sources
• 🧠 Paper Parsing Enhancements — Enhanced paper parsing, section planning, and content filtering
• 🖼️ Figure Matching Improvements — Improved accuracy of body figures and page image matching
• 🧭 Report Structure Optimization — Refined table of contents and section division for better report structure
• 🛠️ Generation Stability Improvements — Improved Provider and Agent generation stability
• 🌐 Status and Preview Refinements — Refined SVG repair, preview parsing, and generation status feedback
• 🧩 Template Import Codex Support
• 🚀 Interactive Launcher

May 2026

• 👁️ Visual QA (Experimental) — Multimodal LLM renders slides as images for layout and contrast review
• 🖥️ Real-time SVG Preview + Log Panel + Critic Detail View — Live slide preview, Agent logs, and review details during generation
• 🎨 Template System & Custom Fonts — Pre-built industry-style templates with custom heading/body font configuration
• 🧩 Template Import — PPTX direct import, five-page template mapping, and Claude Code Agent mode for automated template analysis and templateization
• 🤖 Agent Generation Mode — Integrated Claude Code / Codex presentation generation in the workbench
• 📝 PPT Editor — Visual PPT editor integrated into generated results and template-import workflows, with slide editing, notes, saving, and re-export
• 🔬 Deep Research Workflow — External research enrichment (arXiv / Semantic Scholar / Web) with relevance filtering
• 🎨 Frontend UI Refactor — Refactored the workbench, result page, and template-import frontend experience, and upgraded the SVG-to-PPTX converter

April 2026

• 🔒 Static Critic Enhancements — Decorative-line occlusion detection, low-contrast text detection, multi-line text width estimation fix
• 📁 Version History Management — Automatic snapshot archival per feedback iteration with comparison and rollback
• 🔎 Token Log Filtering — Filter LLM calls by model, stage, page, and job with click-to-expand detail view
• ⏹️ Generation Cancellation — Cancel a running pipeline mid-execution
• 🤖 Multi-Agent Pipeline — Strategist → Executor → Critic three-stage collaboration with automatic SVG repair and feedback iteration

🙏 Acknowledgements

• PPTAgent — Pipeline design and Agent architecture reference
• ppt-master — Parts of the engineering approach reference
• PPTist — PPT editor reference and integration foundation. Thanks to the pipipi-pikachu/PPTist project.

⭐ Star History

📄 License

Released under the GNU Affero General Public License v3.0 (AGPL-3.0).

📬 Contact

• 💬 GitHub Issues: CRui5in/paper-ppt-agent/issues
• 📧 Email: [email protected]