autopreso
Converts spoken words into live presentations by using AI to automatically draw and organize elements on an Excalidraw canvas.
About autopreso
autopreso is an innovative open-source tool designed to bridge the gap between verbal communication and visual storytelling. By leveraging real-time speech-to-text and AI agents, it transforms a live Excalidraw canvas into a dynamic whiteboard that draws, labels, and organizes itself as you speak.
For the Non-Technical Reader
Imagine giving a presentation where you never have to click a 'Next' button or fumble with a laser pointer. autopreso acts like an invisible illustrator sitting next to you; as you explain a concept, the whiteboard populates with diagrams, notes, and connections automatically. It turns a static speech into a living visual map, allowing speakers to stay fully engaged with their audience while the 'deck' builds itself in real-time.
For the Technical Reader
The system architecture is built on a local Express + WebSocket server that binds to 127.0.0.1 for maximum privacy. Key technical highlights include:
- Multi-Model Support: Users can choose between OpenAI Realtime, Codex, or fully local instances via Ollama.
- Transcription Engines: Supports Moonshine for local STT (optimized for macOS) or OpenAI's cloud-based transcription.
- State Management: A 'Staging' mode allows for seeding content and instructions, while 'Live' mode hands canvas control to the agent.
- Latency Optimization: Implements a warmup loop to prime the agent against staging content, ensuring immediate responsiveness once the presentation begins.
- Privacy: Designed to run locally, ensuring that sensitive presentation data does not leave the host machine unless cloud providers are explicitly chosen.
Why It Matters
This project represents a significant shift toward Agentic UI, where the interface is no longer a passive tool but an active collaborator. By supporting local-first workflows (Ollama + Moonshine), it addresses the primary enterprise concerns of data privacy and API costs. It challenges the traditional slide-deck paradigm, moving toward a more fluid, generative form of visual communication.
The Voice AI Space Lab Idea
Why stop at presentations? You could use autopreso to build a 'Live Courtroom Sketch Artist' for digital meetings. As participants debate or brainstorm, the agent could automatically generate a visual 'argument map,' linking conflicting points and highlighting consensus areas in real-time, providing a visual receipt of the decision-making process without anyone taking a single note.
Explore the repository here: https://github.com/kunchenguid/autopreso