Skip to content
BranchesTags

Repository files navigation

Deckium

AI-powered presentation software with human control.

Deckium is a desktop application that helps you create professional presentations using AI assistance. Built with Electron and React, it runs without a Deckium-hosted backend or account: you bring your own LLM API key and choose which provider receives your AI requests.

✨ Features

  • AI-Powered Slide Creation: Describe what you want in natural language; the agent builds and refines slides through a tool-based editing loop
  • Provider-Agnostic AI: Built on the Vercel AI SDK — works with OpenAI, Anthropic, Google AI, Azure OpenAI, and Ollama
  • Design System Built In: Compound components (cards, stat cards, header bars, takeaway bars, tables), automatic layout linting, text-density checks, and visual review keep AI output looking professional
  • Full Editor: Rich text (Quill), shapes, lines, images, SVG, bar charts and plots — with multi-select, group dragging, alignment guides, and smart snapping
  • AI Extras: Optional provider-native web search, image generation, data fetching from URLs, and slide screenshots for the agent's own visual review
  • Import & Export: PowerPoint import/export (full round-trip) and high-quality PDF export with real, selectable text
  • Privacy-Focused: No Deckium backend, account, or telemetry; files stay local and AI requests go directly to the provider you configure

🚀 Quick Start

Prerequisites

  • Node.js 20+ and npm
  • An API key from one of the supported LLM providers (or use Ollama for free local AI)

Installation

  1. Clone the repository

    git clone https://github.com/sleipner42/Deckium.git
cd Deckium

  2. Install dependencies

    cd app
npm install

  3. Start the application

    npm start

  4. Configure your LLM provider

    • Go to Edit > LLM Settings... (or press Cmd+, / Ctrl+,)
    • Select your provider, model, and enter your API key
    • Click Save

That's it! You're ready to create presentations.

🔧 Configuration

Supported LLM Providers

OpenAI

Anthropic (Claude)

Google AI (Gemini)

Azure OpenAI

Ollama (Free, Local)

  • Models: Llama 3.2, Llama 3.1, Mistral, Phi3, CodeLlama
  • Install: https://ollama.com/
  • No API key needed!

Web search: OpenAI, Anthropic, and Google support the provider's native web search tool. It's off by default (each search is billed by the provider) and can be enabled per provider in LLM Settings.

Settings File Location

Settings are stored locally at:

  • Linux: ~/.config/Deckium/llm-settings.json
  • macOS: ~/Library/Application Support/Deckium/llm-settings.json
  • Windows: %APPDATA%\Deckium\llm-settings.json

📖 Usage

Creating Slides

  1. Start a conversation with the AI assistant
  2. Describe what you want: "Create a title slide for my company presentation"
  3. The AI creates and designs slides, checking its own work against layout linting and visual review
  4. Continue refining with natural language — or edit anything directly by hand; the agent sees your manual changes on its next turn

Keyboard Shortcuts

Action macOS Windows/Linux
Open presentation Cmd+O Ctrl+O
Save Cmd+S Ctrl+S
Save As Cmd+Shift+S Ctrl+Shift+S
Export to PDF Cmd+E Ctrl+E
Export to PowerPoint Cmd+Shift+E Ctrl+Shift+E
LLM Settings Cmd+, Ctrl+,
Undo Cmd+Z Ctrl+Z
Redo Cmd+Shift+Z Ctrl+Y

Import & Export

  • PDF Export: File > Export to PDF... — real text (not images) with accurate positioning
  • PowerPoint Export: File > Export to PowerPoint...
  • PowerPoint Import: File > Import PowerPoint...

🏗️ Architecture

Project Structure

Deckium/
├── app/                        # Electron React application (the product)
│   ├── src/
│   │   ├── main/               # Electron main process
│   │   │   ├── ai/             # AI service, provider adapters, tool system
│   │   │   ├── presentation/   # Presentation state and operations
│   │   │   ├── powerpoint/     # PPTX import/export
│   │   │   ├── pdf-export/     # PDF export service
│   │   │   ├── settings/       # LLM settings service
│   │   │   └── main.ts         # Application entry point
│   │   ├── renderer/           # React frontend (editor UI)
│   │   └── common/             # Types and config shared by both processes
│   ├── e2e/                    # Playwright end-to-end tests
│   └── package.json
└── LICENSE

AI Integration

The AI system uses a tool-based agent loop (Vercel AI SDK streamText):

  1. The assistant receives the presentation context, the current slide, and a diff of any manual edits you made since its last turn
  2. It calls tools to manipulate the presentation — creating slides, adding elements, building compound components like cards and stat rows, aligning and distributing, generating images, fetching data
  3. After each edit, the tool result includes a slide grid and linting feedback so the model can catch and fix layout problems itself
  4. Conversation history (including tool calls and reasoning) is persisted per thread, so multi-turn sessions stay coherent and provider prompt caching stays warm

Agent-produced HTML is sanitized against a single shared formatting config, which also drives the editor's Quill setup and the PPTX converter — so what the AI writes, what you edit, and what exports are always in sync.

🛠️ Development

All commands run from the app/ directory.

npm start             # Development mode with hot reload
npm run typecheck     # Type-check (tsc --noEmit)
npm run check         # Lint + format check (Biome)
npm run check:fix     # Auto-fix lint and formatting issues
npm run build         # Production build (main + renderer)
npm run package       # Package for the local platform

Testing

npm test              # Unit/integration tests (Jest, jsdom)
npm run test:e2e      # End-to-end tests (Playwright drives the built Electron app)
npm run test:e2e:only # E2E without rebuilding (when the build is current)
npm run test:pptx-e2e # PowerPoint export → import round-trip

The E2E suite launches the packaged app with an isolated user-data directory and drives the real renderer, covering editor mounting, drag/resize behavior at measured zoom, and the editing lock. See CLAUDE.md for more detail on the test layers.

Code Quality

The project uses Biome for linting and formatting (configured in app/biome.json). Recommended before committing:

npm run check:fix && npm run typecheck

🔒 Privacy & Security

  • Local-First Storage: Presentation files and application state are stored on your machine
  • Direct API Calls: Prompts and relevant presentation context, including slide data or screenshots when requested by the agent, are sent directly to your configured LLM provider — there is no Deckium intermediary
  • API Keys: Keys are encrypted at rest with Electron safeStorage when the operating system provides a secure backend; systems without one fall back to local plaintext storage with a warning. Keys are sent only to the provider they configure
  • Optional Network Tools: Web search, image generation, URL data fetching, and logo lookup contact their respective external services only when used
  • No Tracking: No analytics or telemetry
  • Hardened Tooling: Agent HTML is sanitized before storage, and URL-fetching tools are SSRF-guarded

See SECURITY.md for private vulnerability reporting.

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

  1. Fork the repository
  2. Create your feature branch (git checkout -b feature/amazing-feature)
  3. Commit your changes using conventional commits (git commit -m 'feat: add amazing feature')
  4. Push to the branch (git push origin feature/amazing-feature)
  5. Open a Pull Request

📝 License

This project is licensed under the MIT License.

🙏 Acknowledgments

About

Open-source, local-first AI presentation software with human control.

Resources

Readme
License

Code of conduct

Code of conduct

Contributing

Contributing

Security policy

Security policy
Activity

Stars

0 stars

Watchers

1 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages