PPTAgent

PPTAgent generates presentations from documents or prompts. The accompanying paper, accepted to EMNLP 2025, describes a two-stage, edit-based approach: it analyzes reference slides, drafts an outline, then iteratively produces editing actions on selected reference slides, and it introduces PPTEval to score content, design and coherence. The README's news section lists V2 (deep research integration, free-form visual design, autonomous asset creation, text-to-image generation and an agent environment with a sandbox and 20+ tools), MCP server support, and in January 2026 freeform and template generation with PPTX export and an offline mode.

Setup is the heavy part. The README says all listed API keys, configuration and services are required: a model endpoint in config.yaml, a MinerU API key for document parsing (each key is valid for 14 days), and a Tavily key for search. It recommends Claude for the research agent and Gemini for the design agent, and offline mode works with limited capabilities if you deploy MinerU yourself. Start it with docker compose, or install locally with pip, Playwright and npm and run the web UI.