Run an open-source AI presentation generator on desktop or your own infrastructure

A Presenton self-hosted AI presentation setup can run through its Windows, macOS, or Linux desktop app, or as a Docker-deployable web service. The Apache-2.0 project can use your own text and image providers, offers API and MCP access, custom HTML and Tailwind CSS templates, multi-user workspaces, and editable PPTX or PDF export.

Choose the right entry point

For occasional presentation generation, the official Windows x64 .exe, macOS .dmg, or Linux .deb desktop app is the direct path. For internal workflows, batch generation, or stronger control over data location, use Docker and the API. The official PowerShell Docker example maps a local service to port 5001 and persists app_data; a real team deployment still needs authentication, TLS, backup, API-key handling, and network controls.

You can generate from a prompt or uploaded document and use Ollama, LM Studio, OpenAI, Gemini, Anthropic, or another compatible provider. Source-level Electron development also requires Node.js LTS, npm, Python 3.11, and uv, so it is not the lowest-friction path for ordinary users.

Privacy and output checks

Self-hosting does not automatically make every input local. Using Ollama and local assets can keep data on your device or server, but cloud text or image providers can receive prompts, documents, and image requests. Check each provider and sign-in path separately.

The upstream describes PPTX export as fully editable. That is useful compared with flat page images, but still requires page-by-page checks in PowerPoint for font replacement, text overflow, chart data, image resolution, and cross-version compatibility. The current v0.9.3-beta release makes preserving original inputs and a rebuild path especially important.

Scope boundary

Presenton is a full application with UI, API, MCP, templates, and self-hosting. Gwarket's separate PPT Master candidate focuses on Agent-produced native PowerPoint decks; the two should not be presented as the same product.

Source and license