Tip
PPTAgent Skill for Claude Code, Codex & OpenCode Now Available!
Create, revise, and visually review editable PowerPoint decks with your coding agent.
Install the Skill → · Use Atria Dawn Preview →
Free Token Plan: Discovery · Atria API
You can use a multimodal vision model as a multimodal reviewer and apply for an API token through the Duanyan (端砚) Token Plan. We recommend deepseek-v4-flash-vision as the multimodal model. Setup guide →
Important
Looking for the previous runtime or reproducing the research papers? Use these pinned versions instead of the current development branch:
- PPTAgent (EMNLP 2025): paper · code at v0.2.0
- DeepPresenter (ACL 2026): paper · code at v1.1.38
- [2026/09] 🚀 Introducing Atria Dawn Preview, a new agentic model jointly released by Shanghai AI Laboratory, Fudan University, the Institute of Software (Chinese Academy of Sciences), Renmin University of China, the Institute of Automation (Chinese Academy of Sciences), and East China Normal University. Claim a generous free Token Plan: Discovery · Atria. Use Atria with PPTAgent Skill →
- [2026/09] 🧩 Released PPTAgent Skill for Claude Code, Codex & OpenCode — create, visually review, and export editable PowerPoint decks with your coding agent. Get started →
- [2026/04] 🎉 DeepPresenter accepted to ACL 2026!
- [2026/03] 🤗 We released fine-tuned models and taskset on Hugging Face.
- [2026/01] 🆕 Freeform & template generation now support PPTX export and offline mode. Context management added to prevent context overflow.
- [2025/12] 🔥 Released DeepPresenter codebase with major upgrades — Deep Research Integration, Free-Form Visual Design, Autonomous Asset Creation, Text-to-Image Generation, and an Agent Environment with sandbox & 20+ tools.
- [2025/09] 🛠️ MCP server support added — see the v1.1.37 documentation for details.
- [2025/08] 🎉 PPTAgent accepted to EMNLP 2025!
- [2025/05] ⭐ Reached 1,000 stars on GitHub!
- [2025/01] 🔓 Open-sourced the PPTAgent codebase.
Requirements: Claude Code, Codex CLI, or OpenCode; Linux (including WSL) or macOS; uv; npm; and LibreOffice available as libreoffice on PATH. On macOS, also install Google Chrome for the converter.
On Debian/Ubuntu, install the system dependencies first:
sudo apt-get install npm libreoffice-impressOn macOS, install the system dependencies with Homebrew:
brew install node
brew install --cask libreoffice google-chrome
ln -sf "$(command -v soffice)" "$(brew --prefix)/bin/libreoffice"Clone the repository and install the skill's dependencies. If you already have a checkout, start from its skills/pptagent/ directory:
git clone https://github.com/icip-cas/PPTAgent.git
cd PPTAgent/skills/pptagent
uv venv --python 3.12 .venv
uv pip install --python .venv/bin/python -r requirements.txt
.venv/bin/python -m playwright install --with-deps chromiumRun the command for your client from skills/pptagent/.
Claude Code
.venv/bin/python scripts/install.py --client claudeCodex CLI
.venv/bin/python scripts/install.py --client codexOpenCode
.venv/bin/python scripts/install.py --client opencodeThe installer prepares Node dependencies and registers the skill for the selected client. Keep the repository in place, then check the installation:
.venv/bin/python scripts/pptagent.py doctorFor OpenCode visual MCP configuration and the complete workflow, continue with the OpenCode setup guide.
This example uses Atria Dawn Preview to write and revise the slides, with an external visual model to review them. Complete the installation above, then create an API key in the Atria console.
You can use a multimodal vision model as a multimodal reviewer and apply for an API token through the Duanyan (端砚) Token Plan. We recommend deepseek-v4-flash-vision as the multimodal model.
In skills/pptagent/config.yaml, use the example below and replace base_url with the OpenAI-compatible API base URL shown in your Duanyan console:
mode: text
visual:
base_url: "<OpenAI-compatible API base URL from the Duanyan console>"
model: "deepseek-v4-flash-vision"
api_key_env: VISUAL_API_KEY
timeout_seconds: 300
delivery:
mode: strictSave your Duanyan API key in skills/pptagent/.env:
VISUAL_API_KEY=<your-duanyan-api-key>Atria writes the slides through the text workflow; deepseek-v4-flash-vision reviews the rendered images. The skill appends /chat/completions to base_url; use the API prefix from the console, not the Token Plan webpage URL. With an image-capable host model, you can use mode: multimodal instead. See visual review configuration for details.
Set your Atria key and open a separate task folder:
export ATRIA_API_KEY="<your-atria-api-key>"
mkdir -p ~/pptagent-demo
cd ~/pptagent-demoChoose the client you registered during installation.
Claude Code — launch through Atria's Messages API:
ANTHROPIC_BASE_URL=https://api.atria-asi.ai \
ANTHROPIC_AUTH_TOKEN="$ATRIA_API_KEY" \
claude --model Atria-Dawn-PreviewCodex CLI — merge the following into ~/.codex/config.toml. Keep model and model_provider at the top level, before any section headers, and update existing entries rather than duplicating them:
model = "Atria-Dawn-Preview"
model_provider = "atria"
[model_providers.atria]
name = "Atria"
base_url = "https://api.atria-asi.ai/v1"
env_key = "ATRIA_API_KEY"
wire_api = "responses"Then run codex from the same terminal. It uses Atria's Responses API and reads ATRIA_API_KEY from your environment.
OpenCode — configure Atria through your normal OpenCode provider settings, then follow the OpenCode setup guide to connect the visual review tools.
Send this request in your Atria-powered session:
Use the pptagent skill to create a 6-slide presentation about how our
engineering team can adopt AI coding assistants. Use a clean 16:9 layout.
Render and visually review the slides and the exported deck, then deliver
an editable answer.pptx.
You can also invoke the skill explicitly with /pptagent in Claude Code or $pptagent in Codex. In OpenCode, ask it to use the pptagent skill. Continue in the same session to revise the deck:
Turn slide 3 into a workflow diagram and shorten the recommendations.
Keep the deck at 6 slides, rebuild it, and review the updated PPTX.
Your task folder keeps answer.pptx, editable HTML sources, previews, and the review report. See the Skill guide for details and optional MinerU and search tools for richer source material.
If you find this project helpful, please use the following to cite it:
@inproceedings{zheng-etal-2025-pptagent,
title = "{PPTA}gent: Generating and Evaluating Presentations Beyond Text-to-Slides",
author = "Zheng, Hao and
Guan, Xinyan and
Kong, Hao and
Zhang, Wenkai and
Zheng, Jia and
Zhou, Weixiang and
Lin, Hongyu and
Lu, Yaojie and
Han, Xianpei and
Sun, Le",
editor = "Christodoulopoulos, Christos and
Chakraborty, Tanmoy and
Rose, Carolyn and
Peng, Violet",
booktitle = "Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing",
month = nov,
year = "2025",
address = "Suzhou, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.emnlp-main.728/",
doi = "10.18653/v1/2025.emnlp-main.728",
pages = "14413--14429",
ISBN = "979-8-89176-332-6",
abstract = "Automatically generating presentations from documents is a challenging task that requires accommodating content quality, visual appeal, and structural coherence. Existing methods primarily focus on improving and evaluating the content quality in isolation, overlooking visual appeal and structural coherence, which limits their practical applicability. To address these limitations, we propose PPTAgent, which comprehensively improves presentation generation through a two-stage, edit-based approach inspired by human workflows. PPTAgent first analyzes reference presentations to extract slide-level functional types and content schemas, then drafts an outline and iteratively generates editing actions based on selected reference slides to create new slides. To comprehensively evaluate the quality of generated presentations, we further introduce PPTEval, an evaluation framework that assesses presentations across three dimensions: Content, Design, and Coherence. Results demonstrate that PPTAgent significantly outperforms existing automatic presentation generation methods across all three dimensions."
}
@misc{zheng2026deeppresenterenvironmentgroundedreflectionagentic,
title={DeepPresenter: Environment-Grounded Reflection for Agentic Presentation Generation},
author={Hao Zheng and Guozhao Mo and Xinru Yan and Qianhao Yuan and Wenkai Zhang and Xuanang Chen and Yaojie Lu and Hongyu Lin and Xianpei Han and Le Sun},
year={2026},
eprint={2602.22839},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2602.22839},
}
