icip-cas/PPTAgent

Leverages a multi-agent architecture with deep research integration, autonomous asset creation, and sandbox-based tool execution (20+ tools including web search, PDF parsing, and text-to-image generation) to iteratively refine presentations. Supports both template-based and freeform visual design with PPTX export, offline mode with local model deployment, and integrates via MCP server protocol for broader agentic ecosystems. CLI-driven for personal use or embedding into systems like OpenClaw, with fine-tuned models available on Hugging Face.

3,491 stars. Actively maintained with 38 commits in the last 30 days. Available on PyPI.