Presenton is an open-source AI presentation generator and API, written in TypeScript and licensed under Apache-2.0, that creates, edits, and exports fully editable PowerPoint and PDF decks from a prompt or an uploaded document while running on your own infrastructure or your own model provider keys.

What it is

Presenton is a self-hostable presentation generation platform in the AI development tooling ecosystem. It ships as a web application deployed through a Docker package, as downloadable desktop applications for macOS, Windows, and Linux, and as an AI Presentation Generation API. Generation is driven by a model provider of your choice, and output is a fully editable .pptx or PDF deck that can be refined in a drag-edit interface.

The concrete problem it solves is lock-in. Gamma, Canva, Beautiful AI, Decktopus, and Presentations AI deliver presentation generation as hosted SaaS with subscription pricing and vendor-held data. Presenton replaces that arrangement with software you can run yourself, templates you own, and API keys you control, so generation costs flow directly to whichever provider you select rather than through a fixed platform fee.

Key capabilities

• Generates presentations from a prompt, an uploaded document, or an existing PowerPoint design.
• Exports fully editable PowerPoint (.pptx) files and PDFs with professional formatting.
• Supports custom templates and themes authored in HTML and Tailwind CSS, plus AI template generation from an existing PPTX.
• Works with Ollama, LM Studio, OpenAI, Gemini, Vertex AI, Azure OpenAI, Amazon Bedrock, Fireworks, Together AI, Anthropic, or any other OpenAI-compatible provider.
• Brings your own key (BYOK) and allows mixing text and image generation providers in one workflow.
• Generates images through DALL-E 3, Gemini Flash, Pexels, or Pixabay.
• Exposes a built-in MCP server for generating presentations over Model Context Protocol, alongside the AI Presentation Generation API.

Who uses it and how

• Teams that need deck generation to stay inside their own infrastructure deploy the Docker package and point it at an internal or approved model endpoint.
• Individual users who want local execution or BYOK install the desktop app: .dmg for macOS Apple Silicon and Intel, .exe for Windows x64, .deb for Linux x64.
• Organisations with established brand decks convert existing PowerPoint files into AI-ready templates instead of rebuilding layouts from scratch.
• Product and platform builders embed presentation generation into their own tools by calling the AI Presentation Generation API or the MCP server.
• Privacy-sensitive users run open-source models through Ollama or LM Studio so prompts, documents, and generated decks never leave their machines.

Getting started

Self-hosting runs through the Docker package documented in the quickstart at https://docs.presenton.ai/v3/get-started/quickstart. Desktop builds are downloadable from https://presenton.ai/download for macOS, Windows, and Linux.

How it compares

Against Gamma, Canva, Beautiful AI, Decktopus, and Presentations AI, Presenton is Apache-2.0 licensed open source rather than a proprietary hosted service, so the code can be inspected, modified, and redeployed. Those products are hosted-only, whereas Presenton runs as a self-hosted Docker deployment or a local desktop app, which keeps generated content and prompts within the user's own environment. The cost model differs accordingly: instead of subscription tiers, users pay their chosen model provider directly, or nothing beyond their own hardware when running local models.

When to use it — and when not to

Choose Presenton when data ownership, template control, or avoidance of subscription pricing matters more than zero-operations convenience. A self-hoster must run the Docker deployment and supply model provider credentials unless running a local runtime, and custom designs require familiarity with HTML and Tailwind CSS. Teams that want a fully managed product with no infrastructure or API key handling, or that need a hosted service with an SLA, are not the intended audience.