Summary

autopreso automates presentation slide creation by translating spoken words into real-time whiteboard visuals on an Excalidraw canvas. It targets presenters and educators seeking to focus on content delivery rather than manual slide design, offering a hands-free, dynamic visual generation experience.

How It Works

Audio input from the browser's microphone is processed by a Speech-to-Text (STT) engine (local Moonshine or cloud OpenAI). Transcribed text streams to a whiteboard agent (OpenAI, Codex, Ollama), which generates tool calls to manipulate a live Excalidraw scene. The system runs a local Express/WebSocket server bound to 127.0.0.1, ensuring data privacy. It supports a "staging" mode for initial content setup before transitioning to "live" speech-driven generation.

Quick Start & Requirements

• Installation: npm install -g autopreso or npx autopreso for immediate use. Source install: git clone, npm install, npm start.
• Prerequisites: Node.js/npm. OpenAI API key or Codex subscription for cloud models. Local operation requires Moonshine (STT, macOS-native) and Ollama (agent).
• Setup: npx autopreso offers the fastest startup.
• Links: Repository: https://github.com/kunchenguid/autopreso.git.

Highlighted Details

• Real-time, speech-driven Excalidraw whiteboard generation.
• Flexible model integration: OpenAI, Codex, Ollama supported for agents.
• Enables fully local operation with Moonshine (STT) and Ollama (agent).
• Local-only server binding (127.0.0.1) enhances privacy.
• Persistent configuration and agent instructions.

Maintenance & Community

The project is in alpha and under active development, welcoming bug reports. No specific community channels or notable contributors are detailed in the README.

Licensing & Compatibility

Described as "completely free and open source," but lacks a specific license declaration (e.g., MIT, Apache). This omission hinders assessment for commercial use or derivative works.

Limitations & Caveats

As an alpha project, expect instability, breaking changes, and occasional drawing inaccuracies. Local Moonshine transcription is macOS-only; other platforms require OpenAI Realtime STT. The unspecified license is a significant adoption blocker, especially for commercial applications.