icip-cas/PPTAgent

What it solves

PPTAgent (and its upgraded version, DeepPresenter) automates the creation of professional presentations. It removes the manual effort of designing slides, researching content, and sourcing visual assets by transforming prompts or documents into fully formatted PPTX files.

How it works

The system operates as an AI agent with a sandbox environment and a suite of over 20 tools. It integrates deep research capabilities, autonomous asset creation, and text-to-image generation to build slides. It can process various inputs, including plain text prompts, Excel files, and PDFs, and supports both free-form visual design and template-based generation. For optimal performance, the project provides a fine-tuned model called DeepPresenter-9B.

Who it’s for

It is designed for users who need to generate presentations quickly from data or research, such as students creating classroom materials, professionals preparing reports, or researchers presenting documents.

Highlights

• End-to-End Automation: Generates complete PPTX exports from simple prompts or attached files.
• Rich Toolset: Includes an agent environment with a sandbox and 20+ tools for research and asset creation.
• Multimodal Integration: Supports text-to-image generation and high-quality PDF parsing via MinerU.
• Flexible Deployment: Available via CLI, Docker Compose, or as a source build, with support for offline mode.
• Research-Backed: Based on papers accepted to EMNLP 2025 and ACL 2026.