PPTAgent (DeepPresenter)

PPTAgent is a system designed to automatically generate presentations from documents using a two-phase approach: an Analysis Phase that learns patterns from reference presentations, and a Generation Phase that creates structured outlines and visually cohesive slides.

PPTEval is the accompanying evaluation framework that assesses the quality of generated presentations across three dimensions:

• Content: Accuracy and relevance of the slides.
• Design: Visual appeal and consistency.
• Coherence: Logical flow of ideas.

The CLI is the fastest way for personal use or integrating with OpenClaw.

Note: Windows is not supported; use WSL. On macOS, the CLI may automatically install dependencies like Homebrew, Node.js, Docker, poppler, Playwright, and llama.cpp. On Linux, you must prepare the environment manually.

1. Install uv.
2. Run the interactive onboarding wizard to configure the environment.
3. Use pptagent generate to create presentations.

# Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh

# First-time interactive setup
uvx pptagent onboard

# Generate a presentation
uvx pptagent generate "Single Page with Title: Hello World" -o hello.pptx

# Generate with attachments
uvx pptagent generate "Q4 Report" \
  -f data.xlsx \
  -f charts.pdf \
  -p "10-12" \
  -o report.pptx

You can install PPTAgent via pip. For a full installation including all dependencies, use the [full] extra.

To install from PyPI:

pip install "pptagent[full]"

To install locally in editable mode:

pip install -e ".[full]"

For a stable server environment, use Docker Compose. This will pull the public images for the host service and the sandbox, then start the host service. The web UI will be available at http://localhost:7861.

# Pull the public images
docker pull forceless/deeppresenter-sandbox
docker pull forceless/deeppresenter-host
docker tag forceless/deeppresenter-sandbox deeppresenter-sandbox
docker tag forceless/deeppresenter-host deeppresenter-host

# Start the host service
docker compose up -d

You can run PPTAgent using a pre-built Docker image. Ensure you forward ports 8088 and 9297 if using a remote server.

docker pull forceless/pptagent:latest

# mapping home directory to /root to allow caching of models
docker run -dt --gpus all --ipc=host --name pptagent \
  -e OPENAI_API_KEY=$OPENAI_API_KEY \
  -e MINERU_API=$MINERU_API \
  -p 9297:9297 \
  -p 8088:8088 \
  -v $HOME:/root \
  forceless/pptagent

To view the running details, check the logs:

docker logs -f pptagent

Use this mode for full control over dependencies during development. This requires uv, playwright, npm, and docker.

1. Install the package in editable mode.
2. Install Playwright dependencies and browser.
3. Install HTML-to-PPTX components.
4. Download required models.
5. Pull and tag the necessary Docker images.
6. Start the application using python webui.py.

uv pip install -e .
playwright install-deps
playwright install chromium
npm install --prefix deeppresenter/html2pptx
modelscope download forceless/fasttext-language-id

docker pull forceless/deeppresenter-sandbox
docker pull forceless/deeppresenter-host
docker tag forceless/deeppresenter-sandbox deeppresenter-sandbox
docker tag forceless/deeppresenter-host deeppresenter-host

# Start the app
python webui.py

To run the full WebUI, follow these steps:

1. Initialize Models: Set the following environment variables:

   • OPENAI_API_KEY: Your API key
   • API_BASE: Your service provider endpoint
   • LANGUAGE_MODEL: The model name (e.g., openai/gpt-4.1)
   • VISION_MODEL: The vision model name
   • MINERU_API: The MinerU API endpoint (e.g., http://localhost:8000/file_parse)

2. Run Backend:

   python pptagent_ui/backend.py

3. Launch Frontend:

   cd pptagent_ui
   npm install
   npm run serve

Note: The backend API endpoint is configured in pptagent_ui/vue.config.js.

The desktop-commander is a Model Context Protocol (MCP) server that provides tools for filesystem management, terminal command execution, process control, and configuration. It is designed to be used by LLM clients to interact with a local or containerized environment.

Capabilities:

• Tools: Extensive set of filesystem and terminal tools.
• Resources: Currently returns an empty list.
• Prompts: Currently returns an empty list.
• Logging: Supports logging for console redirection.

Important Path Guidance: Unless explicitly requested, always use absolute paths. Relative paths may fail depending on the current working directory, and tilde paths (~/...) may not work in all contexts.

You can improve generation quality for research, PDF parsing, and visual assets by configuring optional services.

• Tavily: Improves web search. Set TAVILY_API_KEY in deeppresenter/mcp.json.
• MinerU: Improves PDF parsing. Set MINERU_API_KEY in deeppresenter/mcp.json or set MINERU_API_URL to a local deployment.
• Text-to-image model: Configure t2i_model in deeppresenter/config.yaml.

For a fully offline setup, deploy MinerU locally and set offline_mode: true in deeppresenter/config.yaml to disable network-dependent tools like web search.

Run the onboard command to perform the first-time interactive setup. This process:

• Checks for required dependencies (Docker, Playwright, NPM, Poppler).
• Creates configuration files (config.yaml and mcp.json).
• Allows you to configure LLM models (Research Agent, Design Agent, Long Context Model, Vision Model, and Text-to-Image Model).
• Supports configuring local model services (via llama-server) or external APIs.
• Configures MCP (Model Context Protocol) keys for web search (Tavily, SerpAPI) and PDF parsing (MinerU).

If a configuration already exists, you will be prompted to reconfigure (with a backup of existing files) or reuse the current setup.

deeppresenter onboard